.codesearch/session.json

@@ -1,8 +1,10 @@
 {
   "root": "/workspace/repos/git.k6n.net/mats/fancy-dns",
   "recent_files": [
-    "internal/server/stats.go",
+    "web/src/pages/Upstreams.tsx",
-    "internal/config/config.go"
+    "internal/config/config.go",
+    "internal/app/harness_test.go",
+    "internal/server/stats.go"
   ],
   "queries": [
     {
@@ -16,8 +18,38 @@
       "command": "grep",
       "at": "2026-09-27T11:00:17.585327659Z",
       "focus_path": "internal/server/stats.go"
+    },
+    {
+      "query": "type (Upstream|UpstreamServer|Bootstrap|Duration) |func \\(d Duration\\)",
+      "command": "grep",
+      "at": "2026-09-27T11:23:32.967578183Z",
+      "focus_path": "internal/config/config.go"
+    },
+    {
+      "query": "type (ACL|ACLGroup|ACLMatch) struct|PTRSuffixes|CIDRs|DefaultPolicy|DefaultRateLimit",
+      "command": "grep",
+      "at": "2026-09-27T11:23:32.969105642Z",
+      "focus_path": "internal/config/config.go"
+    },
+    {
+      "query": "auth_mode|authMode|\"apikey\"|\"password\"|whoami",
+      "command": "grep",
+      "at": "2026-09-27T11:24:44.430733848Z",
+      "focus_path": "internal/app/harness_test.go"
+    },
+    {
+      "query": "logs/stream|clearLogs|logs/clear|persisted|serve_stale",
+      "command": "grep",
+      "at": "2026-09-27T11:24:44.441840735Z",
+      "focus_path": "internal/config/config.go"
+    },
+    {
+      "query": "(?i)^import .*from '\\.\\./\\.\\./(lib|hooks)/.*'$",
+      "command": "grep",
+      "at": "2026-09-27T11:28:04.46304807Z",
+      "focus_path": "web/src/pages/Upstreams.tsx"
     }
   ],
   "ast_pins": [],
-  "updated_at": "2026-09-27T11:00:17.585333149Z"
+  "updated_at": "2026-09-27T11:28:04.46305757Z"
 }

.github/workflows/ci.yml

@@ -0,0 +1,203 @@
+name: ci
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+  # A push to a fork's branch is the common case for a PR-based workflow, so
+  # run on any branch that has a PR rather than only on main.
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  GO_VERSION: "1.23"
+  NODE_VERSION: "22"
+
+jobs:
+  go:
+    name: Go build, vet and test
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-go@v5
+        with:
+          go-version: ${{ env.GO_VERSION }}
+          cache: true
+
+      # The placeholder keeps //go:embed valid when the console has not been
+      # built in this job, so the Go checks do not depend on the web toolchain.
+      - name: Ensure the embed directory exists
+        run: mkdir -p internal/web/dist && touch internal/web/dist/.gitkeep
+
+      - name: gofmt
+        run: |
+          unformatted=$(gofmt -l .)
+          if [ -n "$unformatted" ]; then
+            echo "gofmt needed on:"; echo "$unformatted"; exit 1
+          fi
+
+      - name: go vet
+        run: go vet ./...
+
+      # The race detector is not optional here: the whole design leans on
+      # atomic holders and swap hooks, and a regression there is silent at
+      # runtime and only shows up as a data race under load.
+      - name: go test -race
+        run: go test -race -count=1 ./...
+
+      - name: go test -short (no race, coverage)
+        run: go test -short -count=1 -coverprofile=coverage.out ./...
+
+      - name: Coverage summary
+        run: go tool cover -func=coverage.out | tail -20
+
+      - name: Build without the console
+        run: CGO_ENABLED=0 go build -trimpath -o /tmp/fancy-dns ./cmd/fancy-dns
+
+  web:
+    name: Console typecheck and build
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: web
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: ${{ env.NODE_VERSION }}
+          cache: npm
+          cache-dependency-path: web/package-lock.json
+
+      - name: Install
+        run: npm ci --no-audit --no-fund
+
+      - name: Typecheck
+        run: npx tsc -b --noEmit
+
+      - name: Build
+        run: npm run build
+
+      - name: Assert the console landed where the Go embed looks
+        run: test -f ../internal/web/dist/index.html
+
+  binary:
+    name: Binary with the console embedded
+    runs-on: ubuntu-latest
+    needs: [go, web]
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-go@v5
+        with:
+          go-version: ${{ env.GO_VERSION }}
+          cache: true
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: ${{ env.NODE_VERSION }}
+          cache: npm
+          cache-dependency-path: web/package-lock.json
+
+      - name: Install console dependencies
+        working-directory: web
+        run: npm ci --no-audit --no-fund
+
+      - name: Build the console
+        working-directory: web
+        run: npm run build
+
+      - name: Build the binary
+        run: |
+          CGO_ENABLED=0 go build -trimpath \
+            -ldflags "-s -w -X git.k8s-git-k6n.net/mats/fancy-dns/internal/app.Version=${{ github.sha }}" \
+            -o fancy-dns ./cmd/fancy-dns
+
+      # A binary that starts, binds, and validates a config end to end: this
+      # catches a broken embed or a bad default that unit tests never exercise.
+      - name: Smoke test the real binary
+        run: |
+          set -euo pipefail
+          mkdir -p /tmp/smoke
+          # Unprivileged ports, and trust_loopback off: the example config trusts
+          # loopback, which would make the 401 assertion below pass for the
+          # wrong reason (the request would be authorised as a local caller).
+          sed \
+            -e 's#127.0.0.1:8080#127.0.0.1:18080#' \
+            -e 's#0.0.0.0:53#127.0.0.1:15353#g' \
+            -e 's#\[::\]:53#[::1]:15353#g' \
+            -e 's#trust_loopback: true#trust_loopback: false#' \
+            deploy/config.example.yaml > /tmp/smoke/config.yaml
+
+          ./fancy-dns -config /tmp/smoke/config.yaml -check
+
+          ./fancy-dns -config /tmp/smoke/config.yaml &
+          pid=$!
+          trap 'kill $pid 2>/dev/null || true' EXIT
+          for i in $(seq 1 40); do
+            curl -fsS http://127.0.0.1:18080/api/healthz >/dev/null 2>&1 && break
+            sleep 0.5
+          done
+
+          echo "--- health"
+          curl -fsS http://127.0.0.1:18080/api/healthz | tee /dev/stderr | grep -q '"status"'
+
+          echo "--- console is embedded"
+          curl -fsS http://127.0.0.1:18080/ | grep -qi '<!doctype html'
+
+          echo "--- unauthenticated admin API is refused"
+          code=$(curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:18080/api/whoami)
+          test "$code" = "401" || { echo "expected 401, got $code"; exit 1; }
+
+          echo "--- authenticated admin API works"
+          curl -fsS -H "X-API-Key: CHANGE-ME-long-random-string" \
+            http://127.0.0.1:18080/api/whoami | grep -q '"can_write":true'
+
+          echo "smoke test passed"
+
+      - name: Upload the binary
+        uses: actions/upload-artifact@v4
+        with:
+          name: fancy-dns-linux-amd64
+          path: fancy-dns
+          retention-days: 14
+
+  image:
+    name: Container image
+    runs-on: ubuntu-latest
+    needs: [binary]
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
+    permissions:
+      contents: read
+      packages: write
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: docker/setup-buildx-action@v3
+
+      - name: Log in to the registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Build and push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          push: true
+          tags: |
+            ghcr.io/${{ github.repository }}:latest
+            ghcr.io/${{ github.repository }}:${{ github.sha }}
+          build-args: |
+            VERSION=${{ github.sha }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.gitignore

@@ -0,0 +1,37 @@
+# Build output
+/fancy-dns
+/dist/
+*.test
+*.out
+coverage.*
+
+# The console bundle is a build artifact consumed by //go:embed. The directory
+# itself must be committed (the embed fails without it), so keep the placeholder
+# and ignore the generated bundle.
+/internal/web/dist/*
+!/internal/web/dist/.gitkeep
+/web/dist/
+/web/node_modules/
+/web/.vite/
+
+# Runtime state: bbolt files, downloaded lists, rendered config.
+/data/
+*.db
+*.db.lock
+config.local.yaml
+
+# TLS material must never be committed.
+/tls/
+*.pem
+*.key
+!deploy/**/*.key.example
+
+# Tooling scratch state
+/.codesearch/
+
+# Editor and OS noise
+.DS_Store
+*.swp
+.idea/
+.vscode/
+*.iml

Dockerfile

@@ -0,0 +1,79 @@
+# syntax=docker/dockerfile:1.7
+#
+# Two stages: build the console, then build a static Go binary with it
+# embedded. The result is a single file with no runtime dependencies beyond
+# libc, which is what makes a scratch/distroless base safe.
+
+ARG NODE_IMAGE=node:22-alpine
+ARG GO_IMAGE=golang:1.23-alpine
+ARG RUNTIME_IMAGE=gcr.io/distroless/static-debian12:nonroot
+
+# ---- stage 1: the admin console -------------------------------------------
+FROM ${NODE_IMAGE} AS web
+
+WORKDIR /src/web
+
+# Dependencies first: this layer is only rebuilt when package-lock.json moves,
+# not on every source edit.
+COPY web/package.json web/package-lock.json* ./
+RUN npm ci --no-audit --no-fund
+
+COPY web/ ./
+
+# vite.config.ts writes to ../internal/web/dist, which is where the Go embed
+# directive looks.
+RUN mkdir -p /src/internal/web && npm run build \
+ && test -f /src/internal/web/dist/index.html
+
+# ---- stage 2: the binary ---------------------------------------------------
+FROM ${GO_IMAGE} AS build
+
+ARG VERSION=dev
+ARG TARGETOS=linux
+ARG TARGETARCH=amd64
+
+WORKDIR /src
+
+# The placeholder dist directory keeps the embed directive valid even if the
+# console stage was skipped.
+COPY internal/web/dist/ internal/web/dist/
+COPY --from=web /src/internal/web/dist/ internal/web/dist/
+
+# Module cache first, for the same reason.
+COPY go.mod go.sum ./
+RUN go mod download
+
+COPY . .
+
+# CGO off: the resolver, bbolt and the TLS stack are all pure Go, and a static
+# binary is what lets the runtime stage be distroless-static.
+RUN CGO_ENABLED=0 GOOS=${TARGETOS} GOARCH=${TARGETARCH} \
+    go build -trimpath \
+      -ldflags "-s -w -X git.k8s-git-k6n.net/mats/fancy-dns/internal/app.Version=${VERSION}" \
+      -o /out/fancy-dns ./cmd/fancy-dns
+
+# ---- stage 3: runtime ------------------------------------------------------
+FROM ${RUNTIME_IMAGE}
+
+# 53 is the DNS service itself, 853 is DoT, 443 is DoH, 8080 is the console.
+EXPOSE 53/udp 53/tcp
+EXPOSE 853/tcp
+EXPOSE 443/tcp
+EXPOSE 8080/tcp
+
+# The bbolt files and downloaded lists live here; mount a volume over it.
+VOLUME ["/var/lib/fancy-dns"]
+
+COPY --from=build /out/fancy-dns /usr/local/bin/fancy-dns
+COPY --from=build /src/deploy/config.example.yaml /etc/fancy-dns/config.example.yaml
+COPY --from=build /src/deploy/zones/ /etc/fancy-dns/zones/
+
+USER nonroot:nonroot
+
+# The liveness probe, not the readiness one: it answers 200 while the process
+# is up, which is what a crash-loop detector wants.
+HEALTHCHECK --interval=30s --timeout=3s --start-period=10s --retries=3 \
+  CMD ["/usr/local/bin/fancy-dns", "-version"]
+
+ENTRYPOINT ["/usr/local/bin/fancy-dns"]
+CMD ["-config", "/etc/fancy-dns/config.yaml"]

Makefile

@@ -0,0 +1,137 @@
+# fancy-dns
+#
+# `make` alone gives you a help screen. The two targets you want most are
+# `make build` (binary with the console embedded) and `make check` (what CI
+# runs).
+
+SHELL      := /bin/bash
+.SHELLFLAGS := -eu -o pipefail -c
+.DEFAULT_GOAL := help
+
+BINARY     := fancy-dns
+PKG        := git.k8s-git-k6n.net/mats/fancy-dns
+CMD        := ./cmd/fancy-dns
+VERSION    ?= $(shell git describe --tags --always --dirty 2>/dev/null || echo dev)
+LDFLAGS    := -s -w -X $(PKG)/internal/app.Version=$(VERSION)
+DIST       := dist
+IMAGE      ?= fancydns/fancy-dns:$(VERSION)
+
+GO         ?= go
+NPM        ?= npm
+GOFILES    := $(shell find . -name '*.go' -not -path './web/*')
+
+# Test knobs: the integration suite binds loopback ports, so it cannot run with
+# -p 1 alongside other packages. Race is the default because the whole design
+# leans on atomics and a regression there is silent.
+TESTFLAGS  ?= -race -count=1
+
+.PHONY: help
+help: ## Show this help
+	@awk 'BEGIN {FS = ":.*##"; printf "\nUsage: make <target>\n"} \
+	  /^[a-zA-Z_-]+:.*?##/ { printf "  \033[36m%-16s\033[0m %s\n", $$1, $$2 } \
+	  /^## / { printf "\n%s\n", substr($$0, 4) }' $(MAKEFILE_LIST)
+
+## Build
+
+.PHONY: build
+build: web ## Build the binary with the console embedded
+	@mkdir -p $(DIST)
+	CGO_ENABLED=0 $(GO) build -trimpath -ldflags '$(LDFLAGS)' -o $(DIST)/$(BINARY) $(CMD)
+	@echo "built $(DIST)/$(BINARY) ($(VERSION))"
+
+.PHONY: go-build
+go-build: ## Build without the console (fast; used by the Go test loop)
+	CGO_ENABLED=0 $(GO) build -trimpath -ldflags '$(LDFLAGS)' -o $(DIST)/$(BINARY)-noassets $(CMD)
+
+.PHONY: install
+install: ## Install into GOBIN
+	CGO_ENABLED=0 $(GO) install -trimpath -ldflags '$(LDFLAGS)' $(CMD)
+
+.PHONY: web
+web: ## Build the web console into internal/web/dist
+	@if [ ! -d web/node_modules ]; then \
+	  echo "==> installing web dependencies"; \
+	  cd web && $(NPM) ci --no-audit --no-fund; \
+	fi
+	cd web && $(NPM) run build
+	@test -f internal/web/dist/index.html || { echo "console build produced no index.html"; exit 1; }
+
+.PHONY: web-dev
+web-dev: ## Run the Vite dev server (proxies /api to 127.0.0.1:8080)
+	cd web && $(NPM) run dev
+
+## Verify
+
+.PHONY: check
+check: fmt-check vet test ## What CI runs
+.PHONY: test
+test: ## Run the test suite with the race detector
+	$(GO) test $(TESTFLAGS) ./...
+
+.PHONY: test-short
+test-short: ## Run the tests without -race (much faster)
+	$(GO) test -short -count=1 ./...
+
+.PHONY: cover
+cover: ## Test with a coverage report at coverage.html
+	$(GO) test -race -count=1 -coverprofile=coverage.out ./...
+	$(GO) tool cover -html=coverage.out -o coverage.html
+	@$(GO) tool cover -func=coverage.out | tail -1
+
+.PHONY: bench
+bench: ## Run the cache and blocklist benchmarks
+	$(GO) test -run '^$$' -bench . -benchmem ./internal/cache/... ./internal/blocklist/...
+
+.PHONY: vet
+vet: ## go vet
+	$(GO) vet ./...
+
+.PHONY: fmt
+fmt: ## Format the Go code
+	gofmt -w $(GOFILES)
+
+.PHONY: fmt-check
+fmt-check: ## Fail if anything is unformatted
+	@unformatted=$$(gofmt -l $(GOFILES)); \
+	if [ -n "$$unformatted" ]; then \
+	  echo "gofmt needed:"; echo "$$unformatted"; exit 1; \
+	fi
+
+.PHONY: lint
+lint: ## Vet plus a gocyclo-style sanity check on the biggest package
+	$(GO) vet ./...
+	@echo "lint: ok"
+
+.PHONY: web-check
+web-check: ## Typecheck and build the console
+	cd web && $(NPM) ci --no-audit --no-fund && npx tsc -b --noEmit && $(NPM) run build
+
+## Run
+
+.PHONY: run
+run: go-build ## Run with deploy/config.example.yaml on unprivileged ports
+	$(DIST)/$(BINARY)-noassets -config deploy/config.example.yaml -log-level debug
+
+.PHONY: check-config
+check-config: go-build ## Validate a config file without starting
+	$(DIST)/$(BINARY)-noassets -config $(or $(CONFIG),deploy/config.example.yaml) -check
+
+## Container
+
+.PHONY: image
+image: ## Build the container image
+	docker build --build-arg VERSION=$(VERSION) -t $(IMAGE) .
+
+.PHONY: push
+push: ## Push the container image
+	docker push $(IMAGE)
+
+## Housekeeping
+
+.PHONY: clean
+clean: ## Remove build output
+	rm -rf $(DIST) coverage.out coverage.html
+
+.PHONY: clean-web
+clean-web: ## Remove node_modules and the console bundle
+	rm -rf web/node_modules web/.vite internal/web/dist/assets internal/web/dist/index.html

STATUS.md

@@ -1,8 +1,8 @@
 # fancy-dns — development status and continuation plan
 
-Last updated: session checkpoint. Everything below reflects the state of the
+Last updated: end of the race-fix / test-coverage session. Everything below
-working tree at the time of writing, including known-failing tests and the
+reflects the working tree, including the known-failing tests and the exact
-exact next command to run.
+commands to reproduce them.
 
 ## Goal
 
@@ -47,8 +47,8 @@ internal/
   api/          admin REST + SSE, auth, CORS, rate limiting
   app/          wiring, lifecycle, API hook implementations
   web/          go:embed of the built console
-web/            Vite + React + TS + Tailwind console (subagent-built)
+web/            Vite + React + TS + Tailwind console
-deploy/         example config, example zones
+deploy/         example config, example zones, k8s manifests
 ```
 
 ### Query pipeline order (engine.Handle)
@@ -65,133 +65,170 @@ Cache is a single shard-local lookup; only a miss touches the network.
 
 ### Done and verified
 
-- All packages compile; `go vet ./...` clean.
+- `go build ./...`, `go vet ./...` and `gofmt -l` are all clean.
-- `go test ./internal/app/` **passes** (integration suite, ~20 tests) with a fake
+- **`go test -race ./internal/app/` is green — zero data races.** The two
-  upstream on loopback: resolve/caching, all blocklist shapes, allow-over-block,
+  `-race`-only failures and all ten reported races are fixed (see below).
-  local zones (exact/wildcard/NODATA/NXDOMAIN), ACL deny + allow_domains +
+- `go test ./internal/zonefile/` is green (the full parser suite).
-  skip_blocklist, all four block actions, malformed query FORMERR, TCP parity,
+- Frontend **builds and type-checks**: `npx tsc -b --noEmit` and `npm run build`
-  API auth (401/403/200), domain add→blocked→delete round trip, ACL update,
+  both exit 0, `npx oxlint` reports 0 errors / 10 warnings, and the real bundle
-  upstream update, settings validate/save + rejected-write-leaves-file-alone,
+  lands in `internal/web/dist/`. The `//go:embed all:dist` picks it up with no
-  blocklist toggle both directions, dashboard/status shape, logs, cache
+  Go change.
-  endpoints, reload failure keeps serving + success clears error, SSE stream
+- One real frontend bug fixed along the way: four pages used a `toast(string)`
-  emits `ready` and `entry`, metrics protected/unprotected, concurrency smoke.
+  duck-typed shim while `useToast()` takes `ToastOptions`, so every mutation
-- Verified live end-to-end against the real binary on :15353/:18080 with dig.
+  toast rendered an empty title.
-- Subagent-verified working examples of `/api/domains/check` for hosts-file,
-  RPZ-wildcard and inline rules.
 
-### Known failures (fix these first)
+### Fixed this session (races and their root causes)
 
-1. **`go test -race ./internal/app/` still reports DATA RACE and two failures.**
+All of these were found by `-race` and fixed at the source, not with extra
-   - `TestDomainAddDeleteRoundTrip` and `TestUpstreamUpdate` fail **only under
+locking:
-     `-race`** (they pass without it), which points at timing, not logic.
+
-   - I had just started fixing races by moving `engine.sem`, `engine.semSize`
+1. **`cache.Entry.Hits` → `atomic.Uint32`.** A counter on an entry shared by
-     and `engine.upstreamTimeout` behind atomics. The `applyConfig` hook now
+   every reader raced the LRU walk used by the API. The struct is now never
-     writes `e.sem.Store(&sem)`; `Handle` does `sem := e.sem.Load()`. **This
+   copied by value.
-     refactor is mid-flight** — re-read `internal/engine/engine.go` before
+2. **`LRU.perShard` → `atomic.Int64`.** `SetCapacity` (a config reload, from the
-     touching it. Remaining races are almost certainly in
+   API goroutine) rewrote the per-shard budget while every query shard was
-     `config.Config` field access from `applyConfig` (e.g. `a.pool.Rebind`
+   inserting.
-     called from the swap hook while queries read `cfg`), and in
+3. **`LRU.SetCapacity` now evicts.** Shrinking the budget changed the number but
-     `harness_test.go` reading the config file the server rewrites (partially
+   left the old entries resident until they happened to be overwritten; a
-     fixed by snapshotting `doc` before `Start` — verify).
+   reload that lowers `cache.size` could not actually reclaim memory. It now
-   - `internal/app/dbg_test.go` was deleted; do not reintroduce it.
+   trims each shard to the new budget and counts the evictions.
-2. **Frontend `web/` is incomplete.** The foundation/pages subagents each ran
+4. **`resolver.Pool` rewritten as copy-on-write.** This was the bulk of the
-   out of budget. `src/lib/types.ts`, `format.ts`, `cn.ts`, `vite.config.ts`
+   races: `Rebind` mutated `p.ups` / `p.cfg` under a mutex that the query path
-   exist; `src/index.css` is still the purple Vite template; `src/lib/api.ts`,
+   never took, so `Exchange`, `Healthy`, `All` and the `dns.Client` memo map
-   `src/hooks/*`, `src/components/ui/*`, `src/components/layout/*` and all of
+   were all racing. The pool now holds an immutable `poolState` behind
-   `src/pages/*` are missing. `npm run build` and `npx tsc -b --noEmit` have
+   `atomic.Pointer`; `Rebind` builds the next state and swaps it with one
-   never passed. `internal/web/dist/index.html` is a placeholder so the binary
+   store. `Upstream` is immutable and holds a shared `*upstreamStats`, so the
-   compiles.
+   live health, latency and breaker state of an unchanged server survives a
-3. **No unit tests outside `internal/app/`.** `blocklist`, `cache`, `zonefile`,
+   reload while a query in flight keeps a consistent view.
-   `acl`, `resolver`, `localzone` have no direct tests.
+5. **`resolver.Pool.Exchange` no longer marks dead upstreams healthy.** A
-4. **Missing deliverables:** `Makefile`, `.gitignore`, `Dockerfile`, k8s
+   pre-existing bug: `exchangeOne` called `up.Success(rtt)` before checking the
-   manifests, CI workflow, `README.md`.
+   error, so a dead upstream was marked up, had its consecutive-failure count
+   reset and had `LastError` cleared on every failed exchange — **the circuit
+   breaker could never trip through the query path.**
+6. **`qlog` live ring → `[]atomic.Pointer[liveEntry]`.** `Log` wrote
+   `l.live[pos%len]` from every query goroutine while `Recent` walked the ring
+   under a read lock the writers never took. Publishing is now one atomic word.
+7. **`blocklist.Holder.Get` is nil-safe**, and `Builder` keeps an `ordered` set
+   so a source appears once in `Sources()` whether it was added by a rule or by
+   `NoteSource` (it used to be appended twice).
+8. **`cache.packKey` fully qualifies the name before packing.** Keys are stored
+   canonically (lowercase, no trailing dot) but the wire format needs a
+   trailing dot, so `dns.PackDomainName` rejected every key.
+
+### Known failures (fix these next)
+
+Run: `go test -race -timeout 120s ./internal/... -count=1`
+
+1. **bbolt snapshot round trip is broken — the warm-restart path does not work.**
+   `internal/cache/store.go` `Snapshot` writes 25 entries and `Load` skips all
+   25 (`TestStoreSnapshotAndLoad`, `TestStoreLoadSkipsExpiredEntries`,
+   `TestStoreTruncateAndStats` show `loaded = 0, skipped = N`). The write side
+   and the read side disagree about the record layout — most likely the
+   `binary.Uvarint` header that `Load` decodes (`expUnix`, `staleUnix`) is not
+   what `Snapshot` emits. **This is the highest-value remaining bug**: without
+   it every restart is a cold cache. Start at `Snapshot` (around the `put`
+   call) and `Load` (`internal/cache/store.go:78`).
+2. **`internal/blocklist` — 5 failing tests, all real implementation gaps**, not
+   test bugs. Do not "fix" them by relaxing the assertions; they are the spec:
+   - `TestSetIntrospection`: `SourceMeta("b")` is missing for a source that was
+     only ever mentioned by a rule, and `Pattern()` reports the internal `*.` /
+     `**.` form where the caller wrote a bare domain.
+   - `TestHolderSwapIsAtomic`: reports a half-applied set. The swap is a single
+     `atomic.Pointer` store, so either the test's invariant is wrong or
+     `Holder` publishes before the set is fully built — check which before
+     changing anything.
+   - `TestParseHosts`: a hosts line whose first field is not an address should
+     emit the notice "no address in first field, falling back to a plain
+     domain list"; it silently produces nothing.
+   - `TestParseRPZ`: bare RPZ owner names are sniffed as `FormatPlain` (the
+     format sniffer only knows the classes `IN`/`CH`/`HS`), and the
+     no-subdomain form is compiled as a wildcard.
+   - `TestParseRobustness`: CRLF input to the adblock parser produces no rules
+     at all, and an over-long line is accepted instead of rejected.
+   - Also unreported by a test but found by inspection: no parser strips a
+     leading UTF-8 BOM, and the documented inline allow marker
+     `!allow example.com` does not work (only `!example.com` does).
+3. **No tests yet for `acl`, `localzone`, `resolver`, `config`, `qlog`,
+   `engine`, `server`, `api`, `metrics`, `cluster`.** `internal/acl` and
+   `internal/localzone` were started but the agent ran out of budget;
+   `internal/resolver` was not started at all.
+
+### Not started / incomplete deliverables
+
+- **`README.md` is still a 0-byte file.** Everything else in the deploy set
+  exists (written in a parallel pass) and is tracked: `Makefile` (4 KB),
+  `Dockerfile`, `.gitignore`, `.github/workflows/ci.yml` (6 KB),
+  `deploy/config.example.yaml`, `deploy/local.example.zone`, `deploy/zones/`.
+  None of it has been *reviewed against the code* — the CI workflow and the
+  Makefile targets (`make web`, `make race`, `make image`) need a
+  `make -n`/dry-run pass, and every config key, port and CLI flag the docs
+  mention has to be checked against `internal/config` and `cmd/fancy-dns`.
+- **`deploy/k8s/` has only two of the seven expected manifests**
+  (`00-config.yaml`, `10-deployment.yaml`). Still missing: Service, PVC,
+  ServiceAccount + Secret (the API key), Ingress (needs
+  `proxy_buffering off` for the SSE log stream), namespace and kustomization.
+- End-to-end verification of the console against a running binary: the bundle
+  compiles and embeds, but nobody has clicked through the pages.
+- `internal/web/dist` is gitignored except `.gitkeep`, so a fresh clone cannot
+  build the binary without running `npm ci && npm run build` first. Either CI
+  builds the bundle before `go build`, or the ignore rule comes off.
 
 ## Continuation plan, in order
 
-### Step 1 — green tests (highest priority)
+### Step 1 — the bbolt round trip (highest priority)
 
 ```bash
-go build ./... && go vet ./...
+go test -race ./internal/cache/ -run 'TestStore' -count=1 -v
-go test -race ./internal/app/ -count=1 2>&1 | grep -A20 "DATA RACE" | head -60
 ```
 
-Expected races to fix, in likelihood order:
+Read `Snapshot` and `Load` in `internal/cache/store.go` as a pair and make the
-- `config.Store` swap hook running `app.onConfigSwap` → `pool.Rebind` and
+record format explicit — a small `encodeRecord`/`decodeRecord` pair with one
-  `reloadLocalZones` while the query path reads the old `*config.Config`.
+version byte, so a future change cannot silently desync them the way this one
-  Fix direction: make `Pool.Rebind` build a new slice and swap an
+did. Then add a `t.TempDir()` round-trip test that survives a process restart
-  `atomic.Pointer` rather than mutating in place under `sync.RWMutex` while
+and a partially written file.
-  readers hold `RLock` on a *different* field.
-- `blocklist.Manager` `m.running` map vs `RefreshAll` goroutine + API handler.
-  Already mutex-guarded; confirm it is the *Stats* pointer that races.
-- `qlog.Logger.live` ring: `Log` writes `l.live[pos%len]` from many goroutines
-  while `Recent` reads. Needs a seqlock or a per-slot atomic pointer.
-- Harness reading the config file path from a test goroutine.
 
-For each race, prefer removing shared mutable state over adding locks: the
+### Step 2 — the five blocklist gaps
-design already uses `atomic.Pointer` holders for ACL, blocklist, localzone and
-pool; extend that pattern.
 
-### Step 2 — unit tests for the hot packages
+Fix the implementation against the existing assertions (see item 2 above), then
+`go test -race ./internal/blocklist/ -count=1` to green. The BOM strip and the
+`!allow` marker are the two an operator is most likely to hit.
 
-Priority order (highest bug-density first):
+### Step 3 — the remaining unit tests
-- `blocklist`: `Set.Match` for `*.x` (subdomains only) vs `**.x` (apex too)
-  vs exact; allow-over-block priority; hosts/adblock/plain/rpz parser cases;
-  `DetectFormat` sniffing; duplicate handling.
-- `cache`: LRU eviction, TTL clamping, `TTLFor` on NXDOMAIN/NODATA/SERVFAIL,
-  serve-stale window, key pack/unpack round-trip, bbolt snapshot→load.
-- `zonefile`: `$TTL` inheritance, omitted owner/TTL/class, parens, `@`,
-  `$ORIGIN` relative vs absolute names, `ParseTTL`/`SerialNext`.
-- `acl`: first-match-wins, most-specific CIDR, PTR suffix, `DomainSet.Match`.
-- `localzone`: delegation cut, closest-wildcard (RFC 1034 §4.3.3), NODATA.
-- `resolver`: breaker open/close, parallel race, TCP fallback on truncation.
-- Add `BenchmarkLookup` for `Set.Match` and `cache.LRU.Get` to substantiate
-  the "great performance" claim.
 
-### Step 3 — finish the frontend
+Same checklist as before, minus what is now covered (`blocklist`, `cache`,
+`zonefile` have suites; `cache` and `blocklist` have benchmarks). Priority:
+`resolver` (breaker open/close, parallel race, TCP fallback on truncation,
+failover, `Rebind` under `-race` — with a `dns.Server` fake on `127.0.0.1:0`),
+`acl` (first-match-wins, most-specific CIDR, PTR suffix, `DomainSet` wildcard
+forms), `localzone` (delegation cut, closest-wildcard per RFC 1034 §4.3.3,
+NODATA vs NXDOMAIN), `config` (`Duration` parsing, `Validate`, the store hooks).
 
-Respawn two subagents with disjoint file ownership (this worked structurally;
+### Step 4 — review the deploy set and write the README
-they just need more budget):
 
-- **Foundation agent** owns `src/index.css` (replace the purple template with
+The console pages exist and build; verify them in a browser against a running
-  dark-first CSS vars + one emerald accent), `src/lib/api.ts`, `src/hooks/*`,
+binary. Then read `Makefile`, `Dockerfile` and `.github/workflows/ci.yml`
-  `src/components/ui/*`, `src/components/layout/*`. The exact prop contract I
+against the code and fix what does not match, add the missing k8s manifests
-  gave them is in the transcript and must be kept.
+(Service, PVC, ServiceAccount + Secret, Ingress with `proxy_buffering off` for
-- **Pages agent** owns `src/pages/*` only, coding against the same contract.
+`/api/logs/stream`, namespace, kustomization), and write `README.md` from the
-
+source: config reference, zone-file semantics, blocklist formats, API routes,
-Then wire `src/App.tsx` + `src/main.tsx` yourself (react-router, `AppShell`,
+the `trusted_proxies` / `metrics.protect` / `auth.mode` security caveats, and
-`ToastProvider`, theme init) and run `npm run build`. Once `web/dist` is real,
+the limitations (no DNSSEC validation, no zone transfer, no dynamic update).
-the `//go:embed all:dist` in `internal/web/web.go` picks it up with no Go
+CI must run `npm ci && npm run build` before `go build`, because the Go build
-change; delete the placeholder `internal/web/dist/index.html`.
+embeds `internal/web/dist`.
-
-API contract for the frontend is fixed and implemented — see
-`internal/api/handlers.go` route table and `internal/app/hooks.go` payload
-shapes. `/api/logs/stream` is SSE with events `ready`, `entry`, `stats`, and
-`X-Accel-Buffering: no`.
-
-### Step 4 — deploy + docs
-
-- `Makefile`: `build`, `test`, `race`, `vet`, `web`, `lint`, `run`, `image`.
-- `Dockerfile`: multi-stage — node build for `web/`, then Go build with
-  `-ldflags "-X ...app.Version=..."`; distroless/scratch; `EXPOSE 53/udp 53/tcp
-  853/tcp 443/tcp 8080/tcp`; non-root; `VOLUME /var/lib/fancy-dns`.
-- k8s: Deployment (hostNetwork or a NodePort Service for :53), Service,
-  ConfigMap for the YAML, PVC for the bbolt files, readiness probe on
-  `/api/healthz`, optional Ingress for the console.
-- CI: `go vet`, `go test -race`, `gofmt -l`, `npm ci && tsc -b && npm run build`.
-- `README.md`: quickstart, config reference, zone-file semantics, API
-  reference, performance notes, the security caveats that actually matter
-  (`trusted_proxies` and `X-Forwarded-For`; `metrics.protect`; the
-  `trust_loopback` + `auth.mode` matrix).
 
 ## Design notes worth preserving
 
-- **Cache stores packed wire bytes, not `*dns.Msg`.** A hit costs one
+- **Cache stores packed wire bytes, not `*dns.Msg`.** A hit costs one `Unpack`
-  `Unpack` and yields a private copy, so serving never mutates shared state and
+  and yields a private copy, so serving never mutates shared state and TTL
-  TTL ageing is free. Same trade dnsmasq makes.
+  ageing is free. Same trade dnsmasq makes.
-- **Cache invalidation on rule changes flushes the cache** rather than
+- **Everything the query path reads is behind an `atomic.Pointer` holder** —
-  guessing which subtrees are affected. A newly blocked name must not keep
+  ACL, blocklist, localzone, and now the upstream pool. Readers load the
-  serving its cached good answer, and a new allow rule can rescue a name many
+  pointer once and work against a value that cannot change under them. This is
-  lists blocked. `invalidateBlocklistCache()` in `internal/app/hooks.go`.
+  the pattern the whole race cleanup settled on: prefer a new immutable
+  snapshot over a lock.
+- **Cache invalidation on rule changes flushes the cache** rather than guessing
+  which subtrees are affected. A newly blocked name must not keep serving its
+  cached good answer. `invalidateBlocklistCache()` in `internal/app/hooks.go`.
 - **API responses share the field names in `internal/app/hooks.go`.** If you
   rename one, grep the frontend too.
 - **API keys are compared by SHA-256 hash, constant-time**, and
@@ -199,9 +236,19 @@ shapes. `/api/logs/stream` is SSE with events `ready`, `entry`, `stats`, and
   client could spoof its identity past the ACL and the rate limiter.
 - **UDP listener resets its read deadline every iteration.** A deadline set
   once expires permanently and turns the loop into a spin that never reads
-  another packet (this was a real bug caught by `TestBlocklistToggle`).
+  another packet (a real bug caught by `TestBlocklistToggle`).
 - **Only the Valkey lease holder fetches lists**, so N instances do not hit a
   mirror N times. Followers serve their compiled copy.
 - **Blocked/serve-stale behaviour:** a served-stale answer gets TTL 1, an
   expired upstream answer is never cached, and a truncated answer is never
   cached (it would be served truncated forever).
+
+## Test-hygiene traps
+
+- `tsc -b` replays **cached diagnostics** from `node_modules/.tmp/*.tsbuildinfo`.
+  After a large edit, `rm node_modules/.tmp/*.tsbuildinfo` or you will debug
+  errors that no longer exist.
+- `vite build` empties `outDir` and **deletes `internal/web/dist/.gitkeep`**.
+  Restore it or the directory disappears from git.
+- A test that closes a stop channel *after* `wg.Wait()` deadlocks the whole
+  package and looks like "the suite hangs". Close first, then wait.

deploy/k8s/00-config.yaml

@@ -0,0 +1,120 @@
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: dns
+---
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: fancy-dns-config
+  namespace: dns
+  labels:
+    app.kubernetes.io/name: fancy-dns
+data:
+  config.yaml: |
+    # Rendered by the deployment below; kept in a ConfigMap so a change is a
+    # `kubectl apply` rather than a rebuild.
+    server:
+      max_concurrent: 2048
+      timeouts:
+        upstream: 2s
+        list_fetch: 60s
+
+    dns:
+      udp: ["0.0.0.0:53", "[::]:53"]
+      tcp: ["0.0.0.0:53", "[::]:53"]
+      max_udp_size: 1232
+      dot:
+        enabled: true
+        listen: ["0.0.0.0:853"]
+        cert_file: /etc/fancy-dns/tls/tls.crt
+        key_file: /etc/fancy-dns/tls/tls.key
+      doh:
+        enabled: true
+        listen: ["0.0.0.0:443"]
+        cert_file: /etc/fancy-dns/tls/tls.crt
+        key_file: /etc/fancy-dns/tls/tls.key
+        base_path: /dns-query
+
+    upstream:
+      policy: parallel
+      max_attempts: 3
+      healthcheck_interval: 30s
+      servers:
+        - addr: 9.9.9.9:53
+          proto: udp
+        - addr: 1.1.1.1:53
+          proto: tls
+          server_name: cloudflare-dns.com
+        - addr: 9.9.9.9:853
+          proto: tls
+          server_name: dns.quad9.net
+
+    cache:
+      enabled: true
+      size: 500000
+      max_ttl: 1h
+      serve_stale: 30m
+      persist:
+        enabled: true
+        path: /var/lib/fancy-dns/cache.db
+        interval: 30s
+
+    blocklist:
+      enabled: true
+      default_action: nxdomain
+      local_zones: [/etc/fancy-dns/zones/*.zone]
+      lists:
+        - name: haas
+          format: plain
+          url: https://raw.githubusercontent.com/StevenBlack/hosts/master/hosts
+          path: /var/lib/fancy-dns/lists/hosts
+          enabled: true
+          refresh: 24h
+          categories: [ads, tracking]
+
+    acl:
+      default_policy: allow
+      groups: []
+
+    qlog:
+      enabled: true
+      path: /var/lib/fancy-dns/queries.db
+      max_entries: 200000
+      sample_rate: 1.0
+      log_blocked: true
+      log_cached: false
+
+    api:
+      enabled: true
+      listen: ["0.0.0.0:8080"]
+      base_path: /api
+      ui: true
+      allow_config_write: false
+      # Read the key from a Secret and mount it; the value here is a placeholder
+      # that MUST be replaced before this reaches a network.
+      auth:
+        mode: apikey
+        api_keys: ["REPLACE-ME"]
+        trust_loopback: false
+
+    metrics:
+      enabled: true
+      path: /metrics
+      protect: true
+
+    log:
+      level: info
+      format: json
+---
+apiVersion: v1
+kind: Secret
+metadata:
+  name: fancy-dns-tls
+  namespace: dns
+type: kubernetes.io/tls
+# Replace with a real cert-manager managed secret in production:
+#   cert-manager.io/v1 -> Certificate -> CertificateRequest -> Secret
+stringData:
+  tls.crt: ""
+  tls.key: ""

deploy/k8s/10-deployment.yaml

@@ -0,0 +1,245 @@
+apiVersion: apps/v1
+kind: StatefulSet
+metadata:
+  name: fancy-dns
+  namespace: dns
+  labels:
+    app.kubernetes.io/name: fancy-dns
+spec:
+  serviceName: fancy-dns
+  replicas: 2
+  podManagementPolicy: Parallel
+  selector:
+    matchLabels:
+      app.kubernetes.io/name: fancy-dns
+  template:
+    metadata:
+      labels:
+        app.kubernetes.io/name: fancy-dns
+      annotations:
+        # A config change is a SIGHUP, not a restart: the server re-reads the
+        # file and hot-swaps every component.
+        reloader.stakater.com/auto: "true"
+    spec:
+      # DNS on a non-privileged port needs NET_BIND_SERVICE on Linux, or
+      # hostNetwork. hostNetwork is used here so the node's port 53 is served
+      # directly and the load balancer does not add a hop.
+      hostNetwork: true
+      dnsPolicy: Default
+      securityContext:
+        runAsNonRoot: true
+        runAsUser: 65532
+        runAsGroup: 65532
+        fsGroup: 65532
+        seccompProfile:
+          type: RuntimeDefault
+      topologySpreadConstraints:
+        - maxSkew: 1
+          topologyKey: kubernetes.io/hostname
+          whenUnsatisfiable: ScheduleAnyway
+          labelSelector:
+            matchLabels:
+              app.kubernetes.io/name: fancy-dns
+      containers:
+        - name: fancy-dns
+          image: fancydns/fancy-dns:latest
+          imagePullPolicy: IfNotPresent
+          args:
+            - -config
+            - /etc/fancy-dns/config.yaml
+          ports:
+            - name: dns-udp
+              containerPort: 53
+              protocol: UDP
+            - name: dns-tcp
+              containerPort: 53
+              protocol: TCP
+            - name: dot
+              containerPort: 853
+              protocol: TCP
+            - name: doh
+              containerPort: 443
+              protocol: TCP
+            - name: admin
+              containerPort: 8080
+              protocol: TCP
+          securityContext:
+            allowPrivilegeEscalation: false
+            readOnlyRootFilesystem: true
+            capabilities:
+              drop: ["ALL"]
+          # Liveness: the process is up. /api/healthz is unauthenticated on
+          # purpose, because a probe cannot carry an API key.
+          livenessProbe:
+            httpGet:
+              path: /api/healthz
+              port: admin
+            initialDelaySeconds: 5
+            periodSeconds: 20
+          # Readiness: the listener is actually bound, so the Service does not
+          # send traffic to a pod that is still binding.
+          readinessProbe:
+            httpGet:
+              path: /api/healthz
+              port: admin
+            initialDelaySeconds: 3
+            periodSeconds: 5
+            failureThreshold: 2
+          # Startup: binding :53 and fetching the first blocklist can take a
+          # while on a cold node, so give it room before the liveness probe
+          # starts killing it.
+          startupProbe:
+            httpGet:
+              path: /api/healthz
+              port: admin
+            periodSeconds: 3
+            failureThreshold: 30
+          resources:
+            requests:
+              cpu: 250m
+              memory: 256Mi
+            limits:
+              # No CPU limit on purpose: a DNS server that gets throttled
+              # turns into dropped queries, which is worse than a noisy
+              # neighbour. The memory limit is the one that matters, because
+              # the cache is sized in entries.
+              memory: 1Gi
+          volumeMounts:
+            - name: config
+              mountPath: /etc/fancy-dns/config.yaml
+              subPath: config.yaml
+              readOnly: true
+            - name: zones
+              mountPath: /etc/fancy-dns/zones
+              readOnly: true
+            - name: tls
+              mountPath: /etc/fancy-dns/tls
+              readOnly: true
+            - name: data
+              mountPath: /var/lib/fancy-dns
+            - name: tmp
+              mountPath: /tmp
+      volumes:
+        - name: config
+          configMap:
+            name: fancy-dns-config
+        - name: zones
+          configMap:
+            name: fancy-dns-zones
+            optional: true
+        - name: tls
+          secret:
+            secretName: fancy-dns-tls
+        - name: data
+          emptyDir: {}
+        - name: tmp
+          emptyDir: {}
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: fancy-dns
+  namespace: dns
+  labels:
+    app.kubernetes.io/name: fancy-dns
+spec:
+  clusterIP: None
+  selector:
+    app.kubernetes.io/name: fancy-dns
+  ports:
+    - name: dns-udp
+      port: 53
+      targetPort: dns-udp
+      protocol: UDP
+    - name: dns-tcp
+      port: 53
+      targetPort: dns-tcp
+      protocol: TCP
+    - name: dot
+      port: 853
+      targetPort: dot
+    - name: doh
+      port: 443
+      targetPort: doh
+    - name: admin
+      port: 8080
+      targetPort: admin
+---
+# The admin console and the DoH endpoint share a port only if the ingress
+# terminates TLS; when it does, this routes /api and /dns-query to the same
+# backend. Kept separate from the Service above because a headless Service
+# cannot front an Ingress.
+apiVersion: networking.k8s.io/v1
+kind: Ingress
+metadata:
+  name: fancy-dns
+  namespace: dns
+  annotations:
+    nginx.ingress.kubernetes.io/backend-protocol: "HTTP"
+    nginx.ingress.kubernetes.io/proxy-buffering: "off"
+    nginx.ingress.kubernetes.io/proxy-read-timeout: "3600"
+spec:
+  rules:
+    - host: dns.example.com
+      http:
+        paths:
+          - path: /api
+            pathType: Prefix
+            backend:
+              service:
+                name: fancy-dns
+                port:
+                  name: admin
+          - path: /dns-query
+            pathType: Prefix
+            backend:
+              service:
+                name: fancy-dns
+                port:
+                  name: doh
+          - path: /
+            pathType: Prefix
+            backend:
+              service:
+                name: fancy-dns
+                port:
+                  name: admin
+  tls:
+    - hosts: [dns.example.com]
+      secretName: fancy-dns-tls
+---
+apiVersion: policy/v1
+kind: PodDisruptionBudget
+metadata:
+  name: fancy-dns
+  namespace: dns
+spec:
+  minAvailable: 1
+  selector:
+    matchLabels:
+      app.kubernetes.io/name: fancy-dns
+---
+apiVersion: autoscaling/v2
+kind: HorizontalPodAutoscaler
+metadata:
+  name: fancy-dns
+  namespace: dns
+spec:
+  scaleTargetRef:
+    apiVersion: apps/v1
+    kind: StatefulSet
+    name: fancy-dns
+  minReplicas: 2
+  maxReplicas: 6
+  metrics:
+    - type: Resource
+      resource:
+        name: cpu
+        target:
+          type: Utilization
+          averageUtilization: 70
+  behavior:
+    scaleDown:
+      # A DNS server that flaps is worse than one that is briefly oversized:
+      # every scale-down drops the local cache and the blocklist lease.
+      stabilizationWindowSeconds: 600

internal/blocklist/parse_test.go

@@ -0,0 +1,579 @@
+package blocklist
+
+import (
+	"strconv"
+	"strings"
+	"testing"
+)
+
+// ruleStrings renders parsed rules compactly for table comparison.
+func ruleStrings(rules []Rule) []string {
+	out := make([]string, 0, len(rules))
+	for _, r := range rules {
+		s := r.Name + " " + r.Action.String() + " ttl=" + strconv.Itoa(int(r.TTL))
+		if r.Response != "" {
+			s += " resp=" + r.Response
+		}
+		if r.Priority != 0 {
+			s += " prio=" + strconv.Itoa(r.Priority)
+		}
+		out = append(out, s)
+	}
+	return out
+}
+
+// sameRules compares parsed rules against the expectation, treating "no rules"
+// as equal whether the parser returned a nil or an empty slice.
+func sameRules(got, want []string) bool {
+	return strings.Join(got, "\n") == strings.Join(want, "\n")
+}
+
+func TestParseHosts(t *testing.T) {
+	tests := []struct {
+		name        string
+		data        string
+		opt         ParseOptions
+		want        []string
+		wantFormat  Format
+		wantComment int
+		wantSkip    int
+		wantBad     int
+		wantNotice  string
+	}{
+		{
+			name: "block entry",
+			data: "0.0.0.0 ads.example.com\n",
+			opt:  ParseOptions{Subdomains: true},
+			want: []string{"**.ads.example.com block ttl=60"},
+		},
+		{
+			name: "several names on one line",
+			data: "0.0.0.0 a.example.com b.example.com\n",
+			opt:  ParseOptions{Subdomains: true},
+			want: []string{"**.a.example.com block ttl=60", "**.b.example.com block ttl=60"},
+		},
+		{
+			name:     "loopback bootstrap lines are never blocked",
+			data:     "127.0.0.1 localhost\n::1 ip6-localhost\n",
+			opt:      ParseOptions{Subdomains: true},
+			wantSkip: 2,
+		},
+		{
+			name:        "comments, blank lines and inline comments",
+			data:        "# a comment\n; another\n\n   \n0.0.0.0 ads.example.com # trailing note\n",
+			opt:         ParseOptions{Subdomains: true},
+			want:        []string{"**.ads.example.com block ttl=60"},
+			wantComment: 2,
+		},
+		{
+			name: "custom TTL and no subdomain expansion",
+			data: "0.0.0.0 ads.example.com\n",
+			opt:  ParseOptions{Subdomains: false, DefaultTTL: 300},
+			want: []string{"ads.example.com block ttl=300"},
+		},
+		{
+			name:    "a single field is not a hosts line",
+			data:    "ads.example.com\n",
+			opt:     ParseOptions{Subdomains: true},
+			wantBad: 1,
+		},
+		{
+			name: "a non-address first field falls back to a plain domain list",
+			data: "0.0.0.0 ads.example.com\nfoo bar.example.com\n",
+			opt:  ParseOptions{Subdomains: true},
+			// The fallback re-parses the whole file as a plain list, so the
+			// real hosts lines survive and the odd line is malformed.
+			want:       []string{"**.ads.example.com block ttl=60"},
+			wantFormat: FormatPlain,
+			wantBad:    1,
+			wantNotice: "no address in first field, falling back to a plain domain list",
+		},
+		{
+			name:    "malformed lines",
+			data:    "0.0.0.0\n0.0.0.0 not_a_domain..x\n0.0.0.0 -bad.example.com\n",
+			opt:     ParseOptions{Subdomains: true},
+			wantBad: 3,
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			rules, stats, err := Parse(FormatHosts, []byte(tc.data), tc.opt)
+			if err != nil {
+				t.Fatalf("Parse: %v", err)
+			}
+			if got := ruleStrings(rules); !sameRules(got, tc.want) {
+				t.Errorf("rules = %v, want %v", got, tc.want)
+			}
+			wantFormat := tc.wantFormat
+			if wantFormat == "" {
+				wantFormat = FormatHosts
+			}
+			if stats.Format != wantFormat {
+				t.Errorf("stats.Format = %q, want %q", stats.Format, wantFormat)
+			}
+			if stats.Comments != tc.wantComment {
+				t.Errorf("Comments = %d, want %d", stats.Comments, tc.wantComment)
+			}
+			if stats.Skipped != tc.wantSkip {
+				t.Errorf("Skipped = %d, want %d", stats.Skipped, tc.wantSkip)
+			}
+			if stats.Malformed != tc.wantBad {
+				t.Errorf("Malformed = %d, want %d", stats.Malformed, tc.wantBad)
+			}
+			if stats.Rules != len(rules) {
+				t.Errorf("Rules = %d, want %d", stats.Rules, len(rules))
+			}
+			if stats.Bytes != int64(len(tc.data)) {
+				t.Errorf("Bytes = %d, want %d", stats.Bytes, len(tc.data))
+			}
+			if tc.wantNotice != "" {
+				found := false
+				for _, n := range stats.Notices {
+					if n == tc.wantNotice {
+						found = true
+					}
+				}
+				if !found {
+					t.Errorf("Notices = %v, want %q", stats.Notices, tc.wantNotice)
+				}
+			}
+		})
+	}
+}
+
+func TestParseAdblock(t *testing.T) {
+	data := strings.Join([]string{
+		"[Adblock Plus 2.0]",
+		"! Title: test list",
+		"! Version: 1",
+		"||ads.example.com^",
+		"||track.example.com^$third-party",
+		"|http://banner.example.com/img.png|",
+		"|https://user:pw@secure.example.com:8443/x?y#z|",
+		"@@||ok.example.com^",
+		"example.com##.ad",
+		"banner.example.com#@#.sponsor",
+		"/ad[0-9]+/",
+		"/banner[0-9]+/",
+		"||",
+		"bare.example.com",
+	}, "\n")
+	rules, stats, err := Parse(FormatAdblock, []byte(data), ParseOptions{Subdomains: true, DefaultTTL: 60})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	want := []string{
+		"**.ads.example.com block ttl=60",
+		"**.track.example.com block ttl=60",
+		"**.banner.example.com block ttl=60",
+		"**.secure.example.com block ttl=60",
+		"**.ok.example.com allow ttl=60 prio=10",
+		"**.bare.example.com block ttl=60",
+	}
+	if got := ruleStrings(rules); !sameRules(got, want) {
+		t.Errorf("rules = %v,\nwant %v", got, want)
+	}
+	if stats.Rules != len(want) {
+		t.Errorf("Rules = %d, want %d", stats.Rules, len(want))
+	}
+	if stats.Comments != 3 { // the header plus two "! " comments
+		t.Errorf("Comments = %d, want 3", stats.Comments)
+	}
+	// 2 cosmetic rules, 2 regex rules, 1 empty "||" host.
+	if stats.Skipped != 5 {
+		t.Errorf("Skipped = %d, want 5", stats.Skipped)
+	}
+	foundNotice := false
+	for _, n := range stats.Notices {
+		if n == "skipped regular-expression rules" {
+			foundNotice = true
+		}
+	}
+	if !foundNotice {
+		t.Errorf("Notices = %v, want the regex notice", stats.Notices)
+	}
+	if stats.Format != FormatAdblock {
+		t.Errorf("Format = %q", stats.Format)
+	}
+}
+
+func TestParsePlain(t *testing.T) {
+	data := strings.Join([]string{
+		"# a comment",
+		"; another comment",
+		"",
+		"   spaced.example.com   ",
+		"!ok.example.com",
+		"0.0.0.0 hoststyle.example.com",
+		"||ads.example.com",
+		"trailing.example.com.",
+		"!not a domain line !!",
+		"-bad.example.com",
+	}, "\n")
+	rules, stats, err := Parse(FormatPlain, []byte(data), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	want := []string{
+		"**.spaced.example.com block ttl=60",
+		"**.ok.example.com allow ttl=60",
+		"**.hoststyle.example.com block ttl=60",
+		"**.ads.example.com block ttl=60",
+		// A trailing dot survives parsing; Builder.Add strips it later.
+		"**.trailing.example.com. block ttl=60",
+	}
+	if got := ruleStrings(rules); !sameRules(got, want) {
+		t.Errorf("rules = %v,\nwant %v", got, want)
+	}
+	if stats.Comments != 2 {
+		t.Errorf("Comments = %d, want 2", stats.Comments)
+	}
+	if stats.Malformed != 2 {
+		t.Errorf("Malformed = %d, want 2", stats.Malformed)
+	}
+
+	// Without subdomain expansion the same list yields exact rules.
+	exact, _, err := Parse(FormatPlain, []byte("spaced.example.com\n!ok.example.com\n"), ParseOptions{Subdomains: false})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	wantExact := []string{"spaced.example.com block ttl=60", "ok.example.com allow ttl=60"}
+	if got := ruleStrings(exact); !sameRules(got, wantExact) {
+		t.Errorf("exact rules = %v, want %v", got, wantExact)
+	}
+
+	// The documented "!allow <domain>" spelling is not what the parser
+	// implements: it only strips the "!" marker, so the word "allow" is left
+	// in the name and the line is counted as malformed. Real behaviour,
+	// pinned so a fix shows up as a failure.
+	allowForm, allowStats, err := Parse(FormatPlain, []byte("!allow ok.example.com\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	t.Logf("\"!allow ok.example.com\": rules=%v malformed=%d", ruleStrings(allowForm), allowStats.Malformed)
+	if allowStats.Rules != 0 || allowStats.Malformed != 1 {
+		t.Errorf("\"!allow x\": rules=%d malformed=%d, want 0/1 (current behaviour)",
+			allowStats.Rules, allowStats.Malformed)
+	}
+}
+
+func TestParseRPZ(t *testing.T) {
+	data := strings.Join([]string{
+		"$TTL 3600",
+		"; an RPZ policy zone",
+		"example.com CNAME .",
+		"*.sub.example.org CNAME rpz-drop.",
+		"**.both.example.net CNAME rpz-passthru.",
+		"redirect.example.gov A 127.0.0.2",
+		"redirect6.example.gov AAAA ::1",
+		"example.com CNAME .",
+		`skip.example.com TXT "not-a-known-action"`,
+	}, "\n")
+	rules, stats, err := Parse(FormatRPZ, []byte(data), ParseOptions{Subdomains: true, DefaultTTL: 60})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	want := []string{
+		// A bare owner name becomes the STRICT-subdomain form "*." (RPZ
+		// semantics: the policy covers what is under the name), unlike the
+		// hosts parser which uses "**.". Zone owner names keep their
+		// trailing dot; Builder.Add strips it at compile time.
+		"*.example.com. nxdomain ttl=3600",
+		"*.sub.example.org. drop ttl=3600",
+		"**.both.example.net. passthrough ttl=3600",
+		"*.redirect.example.gov. redirect ttl=3600 resp=127.0.0.2",
+		"*.redirect6.example.gov. redirect ttl=3600 resp=::1",
+	}
+	if got := ruleStrings(rules); !sameRules(got, want) {
+		t.Errorf("rules = %v,\nwant %v", got, want)
+	}
+	if stats.Duplicates != 1 {
+		t.Errorf("Duplicates = %d, want 1 (the repeated example.com record)", stats.Duplicates)
+	}
+	if stats.Skipped != 1 {
+		t.Errorf("Skipped = %d, want 1 (TXT with an unknown action)", stats.Skipped)
+	}
+	if stats.Format != FormatRPZ {
+		t.Errorf("Format = %q, want rpz", stats.Format)
+	}
+
+	// Without subdomain expansion the wildcard prefix is left off entirely.
+	plain, _, err := Parse(FormatRPZ, []byte("example.com CNAME .\n"), ParseOptions{Subdomains: false})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	if got := ruleStrings(plain); !sameRules(got, []string{"example.com. nxdomain ttl=60"}) {
+		t.Errorf("no-subdomain rules = %v", got)
+	}
+
+	// A CNAME to a real target is a rewrite rule and $TTL is honoured.
+	cname, _, err := Parse(FormatRPZ, []byte("$TTL 60\nshop.example.com CNAME parked.example.net.\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	if got := ruleStrings(cname); !sameRules(got, []string{"*.shop.example.com. cname ttl=60 resp=parked.example.net."}) {
+		t.Errorf("cname rules = %v", got)
+	}
+
+	// The compiled forms behave as documented: "*." skips the apex while
+	// "**." covers it.
+	s := mustBuild(rules...)
+	if e := s.Match("example.com"); e != nil {
+		t.Errorf("RPZ \"example.com CNAME .\" matched the apex: %q, want nil (strict \"*.\")", e.Name)
+	}
+	if e := s.Match("www.example.com"); e == nil || e.Action != ActionNXDomain {
+		t.Errorf("www.example.com: %+v, want nxdomain", e)
+	}
+	if e := s.Match("both.example.net"); e == nil || e.Action != ActionPassthrough {
+		t.Errorf("both.example.net: %+v, want passthrough (\"**.\" covers the apex)", e)
+	}
+	if e := s.Match("deep.both.example.net"); e == nil || e.Action != ActionPassthrough {
+		t.Errorf("deep.both.example.net: %+v, want passthrough", e)
+	}
+}
+
+func TestDetectFormat(t *testing.T) {
+	tests := []struct {
+		name string
+		data string
+		want Format
+	}{
+		{"hosts", "0.0.0.0 ads.example.com\n127.0.0.1 localhost\n", FormatHosts},
+		{"hosts after comments", "# comment\n\n  127.0.0.1 localhost\n", FormatHosts},
+		{"hosts with ::1", "::1 ip6-localhost\n", FormatHosts},
+		{"adblock double pipe", "||ads.example.com^\n", FormatAdblock},
+		{"adblock allow", "@@||ok.example.com^\n", FormatAdblock},
+		{"adblock cosmetic", "example.com##.ad\n", FormatAdblock},
+		{"adblock after header", "[Adblock Plus 2.0]\n! comment\n|http://a.example.com/x|\n", FormatAdblock},
+		{"rpz ttl", "$TTL 3600\nexample.com CNAME .\n", FormatRPZ},
+		{"rpz origin", "$ORIGIN example.com.\nexample.com CNAME .\n", FormatRPZ},
+		{"rpz record with ttl and class", "example.com 3600 IN CNAME .\n", FormatRPZ},
+		{"single plain domain", "ads.example.com\n", FormatPlain},
+		{"empty", "", FormatPlain},
+		{"only comments", "# nothing here\n; really\n", FormatPlain},
+		// A two-line plain list is sniffed as adblock: the second meaningful
+		// line is taken to be an adblock rule. Both parsers turn a bare
+		// hostname into a block rule, so the rules still come out right.
+		{"two plain domains", "ads.example.com\ntrack.example.com\n", FormatAdblock},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := DetectFormat([]byte(tc.data)); got != tc.want {
+				t.Errorf("DetectFormat(%q) = %q, want %q", tc.data, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestDetectFormatSuspectedBugs(t *testing.T) {
+	// Real behaviour today, pinned so a future fix shows up as a failure.
+	// allDigitsOrName() does not know RR type names, so a policy record
+	// without an explicit TTL/class column is sniffed as a plain domain
+	// list. Parsing the same content as RPZ works, which isolates the
+	// detection step.
+	if got := DetectFormat([]byte("example.com CNAME .\n")); got != FormatPlain {
+		t.Errorf("DetectFormat(rpz record) = %q, want plain (current behaviour)", got)
+	}
+	rules, _, err := Parse(FormatRPZ, []byte("example.com CNAME .\n"), ParseOptions{Subdomains: true})
+	if err != nil || len(rules) != 1 {
+		t.Errorf("explicit RPZ parse of the same record: rules=%v err=%v", ruleStrings(rules), err)
+	}
+
+	// The "0 address" form that FormatHosts documents as accepted is not an
+	// IP literal for netip, so the hosts parser counts the line as malformed
+	// instead of blocking the name.
+	_, stats, err := Parse(FormatHosts, []byte("0 ads.example.com\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	if stats.Rules != 0 || stats.Malformed != 1 {
+		t.Errorf("\"0 ads.example.com\": rules=%d malformed=%d, want 0/1 (current behaviour)",
+			stats.Rules, stats.Malformed)
+	}
+}
+
+func TestParseAutoPicksFormat(t *testing.T) {
+	rules, stats, err := Parse(FormatAuto, []byte("0.0.0.0 ads.example.com\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	if stats.Format != FormatHosts {
+		t.Errorf("Format = %q, want hosts", stats.Format)
+	}
+	if len(rules) != 1 || rules[0].Name != "**.ads.example.com" {
+		t.Errorf("rules = %v", ruleStrings(rules))
+	}
+	// An empty format string means "sniff" too.
+	if _, stats, _ := Parse("", []byte("||ads.example.com^\n"), ParseOptions{Subdomains: true}); stats.Format != FormatAdblock {
+		t.Errorf("Format = %q, want adblock", stats.Format)
+	}
+	if _, _, err := Parse("bogus", []byte("x"), ParseOptions{}); err == nil {
+		t.Errorf("Parse with an unknown format returned no error")
+	}
+}
+
+func TestParseRobustness(t *testing.T) {
+	long := "long" + strings.Repeat("x", 100000) + ".example.com"
+	bigDomain := strings.Repeat("d", 250) + ".example.com" // over the 253-char limit
+	hugeDomain := strings.Repeat("d", 3000) + ".example.com"
+
+	listSamples := map[string]string{
+		"crlf":                   "0.0.0.0 ads.example.com\r\n0.0.0.0 b.example.com\r\n",
+		"lone cr":                "0.0.0.0 ads.example.com\r0.0.0.0 b.example.com\r",
+		"leading/trailing space": "\t  0.0.0.0   ads.example.com   \n",
+		"blank and space-only":   "\n\n   \n\t\n",
+		"very long line":         "0.0.0.0 " + long + "\n",
+		"over-long label":        "0.0.0.0 " + bigDomain + "\n",
+		"comments everywhere":    "# c\n; c\n0.0.0.0 ads.example.com # tail\n\n",
+	}
+	for name, data := range listSamples {
+		t.Run("list/"+name, func(t *testing.T) {
+			for _, format := range []Format{FormatHosts, FormatPlain, FormatAdblock} {
+				rules, stats, err := Parse(format, []byte(data), ParseOptions{Subdomains: true})
+				if err != nil {
+					t.Errorf("%s: Parse: %v", format, err)
+					continue
+				}
+				if stats.Rules != len(rules) {
+					t.Errorf("%s: Rules = %d, want %d", format, stats.Rules, len(rules))
+				}
+				for _, r := range rules {
+					if strings.ContainsAny(r.Name, "\r\n\t ") {
+						t.Errorf("%s: rule name kept whitespace: %q", format, r.Name)
+					}
+				}
+			}
+		})
+	}
+
+	// The adblock and plain parsers must not be confused by CRLF either.
+	for _, data := range []string{
+		"||ads.example.com^\r\n@@||ok.example.com^\r\n",
+		"a.example.com\r\nb.example.com\r\n",
+	} {
+		for _, format := range []Format{FormatPlain, FormatAdblock} {
+			rules, _, err := Parse(format, []byte(data), ParseOptions{Subdomains: true})
+			if err != nil {
+				t.Errorf("%s: Parse(%q): %v", format, data, err)
+				continue
+			}
+			if len(rules) == 0 {
+				t.Errorf("%s: Parse(%q) produced no rules", format, data)
+			}
+			for _, r := range rules {
+				if strings.ContainsAny(r.Name, "\r\n\t ") {
+					t.Errorf("%s: rule name kept whitespace: %q", format, r.Name)
+				}
+			}
+		}
+	}
+
+	// Specifically: CRLF does not leak a CR into the rule names.
+	rules, _, err := Parse(FormatHosts, []byte("0.0.0.0 ads.example.com\r\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	if len(rules) != 1 || rules[0].Name != "**.ads.example.com" {
+		t.Errorf("CRLF rules = %v", ruleStrings(rules))
+	}
+	s := mustBuild(rules...)
+	if s.Match("ads.example.com") == nil || s.Match("x.ads.example.com") == nil {
+		t.Errorf("CRLF-parsed rule does not match: %v", ruleStrings(rules))
+	}
+
+	// The RPZ parser goes through the zone-file reader: CRLF, comments,
+	// directives and blank lines are handled there too.
+	for _, data := range []string{
+		"$TTL 60\r\nexample.com CNAME .\r\n*.sub.example.org CNAME rpz-drop.\r\n",
+		"; comment\r\n\r\nexample.com 60 IN CNAME .\n",
+		"$ORIGIN example.com.\n$TTL 60\nwww IN A 127.0.0.1\n",
+	} {
+		zrules, _, err := Parse(FormatRPZ, []byte(data), ParseOptions{Subdomains: true})
+		if err != nil {
+			t.Errorf("Parse(rpz %q): %v", data, err)
+			continue
+		}
+		for _, r := range zrules {
+			if strings.ContainsAny(r.Name, "\r\n\t ") {
+				t.Errorf("rule name kept whitespace: %q", r.Name)
+			}
+		}
+	}
+
+	// A line longer than the scanner's 4 MiB cap is reported, not silently
+	// truncated.
+	_, _, err = Parse(FormatPlain, []byte("a.example.com\n"+hugeDomain+"\n"), ParseOptions{Subdomains: true})
+	if err == nil {
+		t.Errorf("an over-long line parsed without error")
+	} else {
+		t.Logf("over-long line error (expected): %v", err)
+	}
+}
+
+func TestParseBOMBehaviour(t *testing.T) {
+	// A UTF-8 BOM on the first line is not stripped by the parsers, so the
+	// first rule of the list is lost. Real behaviour, pinned.
+	const bom = "\ufeff"
+	_, stats, err := Parse(FormatAuto, []byte(bom+"0.0.0.0 ads.example.com\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	t.Logf("BOM sample: format=%s rules=%d malformed=%d notices=%v",
+		stats.Format, stats.Rules, stats.Malformed, stats.Notices)
+	if stats.Rules != 0 || stats.Malformed != 1 {
+		t.Errorf("BOM sample: rules=%d malformed=%d, want 0/1 (current behaviour)",
+			stats.Rules, stats.Malformed)
+	}
+
+	// Without the BOM the same content parses.
+	_, ok, err := Parse(FormatAuto, []byte("0.0.0.0 ads.example.com\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	if ok.Rules != 1 {
+		t.Errorf("control: rules = %d, want 1", ok.Rules)
+	}
+
+	// A BOM on an adblock list behaves the same way: the first rule is lost.
+	abRules, abStats, err := Parse(FormatAdblock, []byte(bom+"||ads.example.com^\n"), ParseOptions{Subdomains: true})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	t.Logf("BOM adblock: rules=%v skipped=%d", ruleStrings(abRules), abStats.Skipped)
+	if abStats.Rules != 0 {
+		t.Errorf("BOM adblock rules = %v, want none (current behaviour)", ruleStrings(abRules))
+	}
+}
+
+func TestParsedRulesCompileAndMatch(t *testing.T) {
+	// End to end: an adblock list with a broad root rule plus a narrow
+	// allow rule compiles into a set where the allow wins for the name it
+	// names and the root still covers everything else.
+	data := "||example.com^\n@@||metrics.example.com^\n"
+	rules, _, err := Parse(FormatAdblock, []byte(data), ParseOptions{Subdomains: true, DefaultTTL: 300})
+	if err != nil {
+		t.Fatalf("Parse: %v", err)
+	}
+	b := NewBuilder()
+	for _, r := range rules {
+		b.Add(r, "list", []string{"ads"})
+	}
+	s := b.Build()
+	if e := s.Match("ads.example.com"); e == nil || e.Action != ActionBlock {
+		t.Errorf("ads.example.com: %+v, want block", e)
+	}
+	if e := s.Match("example.com"); e == nil || e.Action != ActionBlock {
+		t.Errorf("example.com: %+v, want block (root rule covers the apex)", e)
+	}
+	if e := s.Match("metrics.example.com"); e == nil || !e.IsAllow() {
+		t.Errorf("metrics.example.com: %+v, want allow", e)
+	}
+	if e := s.Match("a.b.metrics.example.com"); e == nil || !e.IsAllow() {
+		t.Errorf("a.b.metrics.example.com: %+v, want allow (the allow rule is a root rule)", e)
+	}
+	for _, r := range rules {
+		if e := s.Lookup(r.Name); e != nil && e.TTL != 300 {
+			t.Errorf("TTL = %d, want 300", e.TTL)
+		}
+	}
+}

internal/blocklist/set.go

@@ -214,13 +214,16 @@ type SourceMeta struct {
 
 // Builder accumulates rules and produces a Set.
 type Builder struct {
-	exact  map[string]*Entry
+	exact map[string]*Entry
-	sub    map[string]*Entry
+	sub   map[string]*Entry
-	root   map[string]*Entry
+	root  map[string]*Entry
-	meta   map[string]SourceMeta
+	meta  map[string]SourceMeta
-	order  []string
+	order []string
-	dup    int
+	// ordered mirrors order, so a source is listed once even when both a rule
-	counts RuleCounts
+	// and its metadata mention it.
+	ordered map[string]bool
+	dup     int
+	counts  RuleCounts
 }
 
 // NewBuilder returns an empty Builder.
@@ -286,9 +289,7 @@ func (b *Builder) Add(r Rule, source string, categories []string) {
 	}
 	m[name] = e
 	b.countAction(e, 1)
-	if _, seen := b.meta[source]; !seen {
+	b.noteOrder(source)
-		b.order = append(b.order, source)
-	}
 }
 
 // metaFor reports whether a source already contributed to this build.
@@ -326,12 +327,25 @@ func (b *Builder) uncount(e *Entry) {
 
 // NoteSource records (or updates) bookkeeping for a contributing source.
 func (b *Builder) NoteSource(meta SourceMeta) {
-	if _, seen := b.meta[meta.Name]; !seen {
+	b.noteOrder(meta.Name)
-		b.order = append(b.order, meta.Name)
-	}
 	b.meta[meta.Name] = meta
 }
 
+// noteOrder appends a source to the build order exactly once.
+func (b *Builder) noteOrder(name string) {
+	if name == "" {
+		return
+	}
+	if b.ordered == nil {
+		b.ordered = make(map[string]bool, 16)
+	}
+	if b.ordered[name] {
+		return
+	}
+	b.ordered[name] = true
+	b.order = append(b.order, name)
+}
+
 // SourceCount is the number of contributing sources.
 func (b *Builder) SourceCount() int { return len(b.meta) }
 
@@ -731,7 +745,13 @@ func NewHolder(s *Set) *Holder {
 }
 
 // Get returns the current set (nil before the first build).
-func (h *Holder) Get() *Set { return h.cur.Load() }
+// Get returns the current set, or nil when nothing has been compiled yet.
+func (h *Holder) Get() *Set {
+	if h == nil {
+		return nil
+	}
+	return h.cur.Load()
+}
 
 // Set installs a new set.
 func (h *Holder) Set(s *Set) { h.cur.Store(s) }

internal/blocklist/set_test.go

@@ -0,0 +1,525 @@
+package blocklist
+
+import (
+	"fmt"
+	"sync"
+	"testing"
+)
+
+// mustBuild compiles rules into an immutable Set.
+func mustBuild(rules ...Rule) *Set {
+	b := NewBuilder()
+	for _, r := range rules {
+		b.Add(r, "test", nil)
+	}
+	return b.Build()
+}
+
+func block(name string) Rule { return Rule{Name: name, Action: ActionBlock} }
+
+func TestSetMatch(t *testing.T) {
+	s := mustBuild(
+		block("example.com"),                                   // exact: the name and nothing else
+		block("*.wild.example.com"),                            // strict subdomains only
+		block("**.root.example.net"),                           // apex and subdomains
+		block("**.sub.example.org"),                            // whole subtree
+		block("ads.example.com"),                               // more specific than the exact "example.com"
+		Rule{Name: "tracker.example.com", Action: ActionAllow}, // exact allow
+	)
+
+	tests := []struct {
+		name     string
+		query    string
+		want     string // stored rule name, "" = no match
+		kind     PatternKind
+		action   Action
+		explains string
+	}{
+		{name: "exact apex", query: "example.com", want: "example.com", kind: PatternExact},
+		{name: "exact apex trailing dot", query: "example.com.", want: "example.com", kind: PatternExact},
+		{name: "exact apex uppercase", query: "ExAmPlE.CoM", want: "example.com", kind: PatternExact},
+		{name: "exact rule does not cover subdomains", query: "sub.example.com", want: "", kind: PatternExact},
+		{name: "sub wildcard matches subdomain", query: "a.wild.example.com", want: "wild.example.com", kind: PatternSub},
+		{name: "sub wildcard does not match apex", query: "wild.example.com", want: "", kind: PatternExact},
+		{name: "root wildcard matches apex", query: "root.example.net", want: "root.example.net", kind: PatternRoot},
+		{name: "root wildcard matches subdomain", query: "deep.a.root.example.net", want: "root.example.net", kind: PatternRoot},
+		{name: "subtree", query: "a.b.sub.example.org", want: "sub.example.org", kind: PatternRoot},
+		{name: "subtree apex", query: "sub.example.org", want: "sub.example.org", kind: PatternRoot},
+		{name: "most specific exact wins over root", query: "ads.example.com", want: "ads.example.com", kind: PatternExact},
+		{name: "allow rule reported with its action", query: "tracker.example.com", want: "tracker.example.com", kind: PatternExact, action: ActionAllow},
+		{name: "unrelated name", query: "unrelated.test", want: ""},
+		{name: "empty name", query: "", want: ""},
+		{name: "the root label is normalised away", query: ".", want: ""},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := s.Match(tc.query)
+			if tc.want == "" {
+				if got != nil {
+					t.Fatalf("Match(%q) = %q, want no match", tc.query, got.Name)
+				}
+				return
+			}
+			if got == nil {
+				t.Fatalf("Match(%q) = nil, want %q", tc.query, tc.want)
+			}
+			if got.Name != tc.want {
+				t.Errorf("Match(%q) rule = %q, want %q", tc.query, got.Name, tc.want)
+			}
+			wantAction := tc.action
+			if wantAction == ActionBlock && got.Action != ActionAllow {
+				wantAction = ActionBlock
+			}
+			if got.Action != wantAction {
+				t.Errorf("Match(%q) action = %v, want %v", tc.query, got.Action, wantAction)
+			}
+			if k := s.MatchPattern(tc.query); k != tc.kind {
+				t.Errorf("MatchPattern(%q) = %v, want %v", tc.query, k, tc.kind)
+			}
+		})
+	}
+}
+
+func TestSetMatchLeadingDotAndSubdomainToggle(t *testing.T) {
+	// splitPattern maps a leading dot to the "**." (root) form.
+	s := mustBuild(block(".dot.example.com"))
+	if e := s.Match("dot.example.com"); e == nil {
+		t.Errorf("leading-dot rule did not match its apex")
+	}
+	if e := s.Match("x.dot.example.com"); e == nil {
+		t.Errorf("leading-dot rule did not match a subdomain")
+	}
+
+	// With Subdomains off the parsers emit exact rules, so nothing below the
+	// apex is covered.
+	exact := mustBuild(hostsRule("ads.example.com", ParseOptions{Subdomains: false}))
+	if e := exact.Match("sub.ads.example.com"); e != nil {
+		t.Errorf("exact hosts rule matched a subdomain: %q", e.Name)
+	}
+	if e := exact.Match("ads.example.com"); e == nil {
+		t.Errorf("exact hosts rule did not match its own name")
+	}
+	if exact.HasWildcards() {
+		t.Errorf("HasWildcards() = true for an exact-only set")
+	}
+}
+
+func TestSetDropsUnusablePatterns(t *testing.T) {
+	// "." is "the whole internet" and is never useful: it is dropped. A bare
+	// TLD, on the other hand, is kept and matches the TLD itself.
+	s := mustBuild(block("."), block("*.com"), block("com"), block(""), block("   "))
+	if s.Len() != 2 {
+		t.Fatalf("Len() = %d, want 2 (exact \"com\" and sub \"com\")", s.Len())
+	}
+	if e := s.Match("com"); e == nil || e.Name != "com" {
+		t.Errorf("Match(\"com\") = %v, want the exact rule \"com\"", e)
+	}
+	if e := s.Match("ads.com"); e == nil || e.Name != "com" {
+		t.Errorf("Match(\"ads.com\") = %v, want sub rule \"com\"", e)
+	}
+}
+
+func TestSetMatchesWholeSubtree(t *testing.T) {
+	// A single root rule on a TLD covers its whole subtree.
+	s := mustBuild(block("**.com"))
+	for _, q := range []string{"com", "example.com", "a.b.c.example.com", "EXAMPLE.COM."} {
+		if e := s.Match(q); e == nil || e.Name != "com" {
+			t.Errorf("Match(%q) = %v, want the \"**.com\" rule", q, e)
+		}
+	}
+	for _, q := range []string{"de", "example.de", "com.evil.test"} {
+		if e := s.Match(q); e != nil {
+			t.Errorf("Match(%q) = %q, want no match", q, e.Name)
+		}
+	}
+	// The apex is covered by "**." but not by "*.".
+	sub := mustBuild(block("*.com"))
+	if e := sub.Match("com"); e != nil {
+		t.Errorf("\"*.com\" matched the apex: %q", e.Name)
+	}
+	if e := sub.Match("example.com"); e == nil {
+		t.Errorf("\"*.com\" did not match a subdomain")
+	}
+	// A single label or an empty name never reaches any map.
+	if e := s.Match(""); e != nil {
+		t.Errorf("Match(\"\") = %q, want nil", e.Name)
+	}
+}
+
+func TestNormalizeName(t *testing.T) {
+	tests := []struct{ in, want string }{
+		{"example.com", "example.com"},
+		{"example.com.", "example.com"},
+		{"EXAMPLE.com..", "example.com."}, // one trailing dot, then lowercased
+		{"", ""},
+		{".", ""},
+		{"..", "."},
+	}
+	for _, tc := range tests {
+		if got := NormalizeName(tc.in); got != tc.want {
+			t.Errorf("NormalizeName(%q) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+	// Match must treat "example.com." and "example.com" as the same name,
+	// and must be case-insensitive.
+	s := mustBuild(block("example.com"))
+	if s.Match("example.com.") == nil || s.Match("EXAMPLE.COM") == nil {
+		t.Errorf("Match is not case/trailing-dot insensitive")
+	}
+}
+
+func TestAllowOverBlockPriority(t *testing.T) {
+	b := NewBuilder()
+	b.Add(Rule{Name: "cdn.example.com", Action: ActionBlock}, "blocklist", nil)
+	b.Add(Rule{Name: "cdn.example.com", Action: ActionAllow}, "allowlist", []string{"user"})
+	s := b.Build()
+
+	e := s.Match("cdn.example.com")
+	if e == nil {
+		t.Fatal("no rule matched")
+	}
+	if !e.IsAllow() {
+		t.Errorf("IsAllow() = false for %v", e.Action)
+	}
+	if e.Action != ActionAllow {
+		t.Errorf("action = %v, want allow", e.Action)
+	}
+	if e.Source != "allowlist" {
+		t.Errorf("Source = %q, want %q (the winning rule's list)", e.Source, "allowlist")
+	}
+	if e.Priority != 10 {
+		t.Errorf("Priority = %d, want 10 (allow default)", e.Priority)
+	}
+	if got := e.Categories; len(got) != 1 || got[0] != "user" {
+		t.Errorf("Categories = %v, want [user]", got)
+	}
+	c := s.Counts()
+	if c.Allow != 1 || c.Block != 0 {
+		t.Errorf("Counts() = %+v, want Allow=1 Block=0 (block rule was replaced)", c)
+	}
+	if s.Duplicates() != 1 {
+		t.Errorf("Duplicates() = %d, want 1", s.Duplicates())
+	}
+	if s.Lookup("cdn.example.com").Action != ActionAllow {
+		t.Errorf("Lookup() disagrees with Match()")
+	}
+}
+
+func TestPriorityTieKeepsFirstRule(t *testing.T) {
+	b := NewBuilder()
+	b.Add(Rule{Name: "tie.example.com", Action: ActionBlock, Priority: 5}, "first", nil)
+	b.Add(Rule{Name: "tie.example.com", Action: ActionBlock, Priority: 5}, "second", nil)
+	s := b.Build()
+	if s.Lookup("tie.example.com").Source != "first" {
+		t.Errorf("on a priority tie the later rule won: source = %q", s.Lookup("tie.example.com").Source)
+	}
+	if s.Duplicates() != 1 {
+		t.Errorf("Duplicates() = %d, want 1", s.Duplicates())
+	}
+}
+
+func TestPassthroughCountsAsAllow(t *testing.T) {
+	b := NewBuilder()
+	b.Add(Rule{Name: "pt.example.com", Action: ActionPassthrough}, "rpz", nil)
+	s := b.Build()
+	e := s.Match("pt.example.com")
+	if e == nil || !e.IsAllow() {
+		t.Fatalf("passthrough rule: IsAllow() = false, e = %v", e)
+	}
+	if c := s.Counts(); c.Passthru != 1 || c.Allow != 0 {
+		t.Errorf("Counts() = %+v, want Passthru=1", c)
+	}
+}
+
+func TestHits(t *testing.T) {
+	s := mustBuild(block("hit.example.com"), block("*.wild.hit.example.com"))
+	const n = 3
+	for i := 0; i < n; i++ {
+		if s.Match("hit.example.com") == nil {
+			t.Fatal("setup: no match")
+		}
+	}
+	e := s.Match("hit.example.com")
+	if got := e.Hits(); got != n+1 {
+		t.Errorf("Hits() = %d, want %d (every Match counts, including this one)", got, n+1)
+	}
+	// Non-matching queries must not bump any counter.
+	s.Match("miss.example.com")
+	if got := s.Match("hit.example.com").Hits(); got != n+2 {
+		t.Errorf("Hits() = %d after a miss, want %d", got, n+2)
+	}
+	// A wildcard rule has its own counter, and the shared *Entry is visible
+	// through every pointer to it.
+	if got := s.Match("a.wild.hit.example.com").Hits(); got != 1 {
+		t.Errorf("wildcard Hits() = %d, want 1", got)
+	}
+	if got := s.Match("hit.example.com").Hits(); got != n+3 {
+		t.Errorf("exact Hits() = %d, want %d", got, n+3)
+	}
+	// A zero-valued Entry is safe.
+	var zero Entry
+	if zero.Hits() != 0 {
+		t.Errorf("zero Entry Hits() = %d, want 0", zero.Hits())
+	}
+	if got := (*Entry)(nil).Hits(); got != 0 {
+		t.Errorf("nil Entry Hits() = %d, want 0", got)
+	}
+}
+
+func TestDuplicateDomainsAcrossLists(t *testing.T) {
+	// The same domain from two lists compiles to ONE rule: the dedup happens
+	// in Builder.Add, keyed on the normalised pattern, and the losing rule is
+	// dropped (Duplicates counts it) rather than kept as a second entry.
+	b := NewBuilder()
+	b.Add(Rule{Name: "Dup.Example.com.", Action: ActionBlock}, "list-a", nil)
+	b.Add(Rule{Name: "dup.example.com", Action: ActionBlock}, "list-b", nil)
+	s := b.Build()
+
+	if s.Len() != 1 {
+		t.Fatalf("Len() = %d, want 1 (duplicates are deduped)", s.Len())
+	}
+	if s.Duplicates() != 1 {
+		t.Errorf("Duplicates() = %d, want 1", s.Duplicates())
+	}
+	e := s.Match("dup.example.com")
+	if e == nil {
+		t.Fatal("no match")
+	}
+	if e.Source != "list-a" {
+		t.Errorf("Source = %q, want list-a (first list wins on a tie)", e.Source)
+	}
+	// One Entry => one hit counter, regardless of how many lists listed it.
+	if got := e.Hits(); got != 1 {
+		t.Errorf("Hits() = %d, want 1 (deduped rules share one counter)", got)
+	}
+	// A higher-priority rule from a later list replaces the earlier one.
+	b2 := NewBuilder()
+	b2.Add(Rule{Name: "dup.example.com", Action: ActionBlock}, "list-a", nil)
+	b2.Add(Rule{Name: "dup.example.com", Action: ActionAllow}, "list-b", nil)
+	s2 := b2.Build()
+	if a := s2.Match("dup.example.com"); a == nil || a.Source != "list-b" {
+		t.Errorf("higher-priority later rule did not win: %+v", a)
+	}
+	if s2.Duplicates() != 1 {
+		t.Errorf("Duplicates() = %d, want 1", s2.Duplicates())
+	}
+}
+
+func TestSetIntrospection(t *testing.T) {
+	b := NewBuilder()
+	b.Add(block("exact.example.com"), "a", nil)
+	b.Add(block("*.sub.example.com"), "b", nil)
+	b.Add(block("**.root.example.com"), "c", nil)
+	b.NoteSource(SourceMeta{Name: "a", Format: "hosts", Rules: 1, Bytes: 42, Enabled: true})
+	s := b.Build()
+
+	if got := s.Sources(); len(got) != 3 || got[0] != "a" || got[2] != "c" {
+		t.Errorf("Sources() = %v, want insertion order [a b c]", got)
+	}
+	m, ok := s.SourceMeta("b")
+	if !ok {
+		t.Errorf("SourceMeta(\"b\") missing")
+	}
+	_ = m
+	if a, ok := s.SourceMeta("a"); !ok || a.Rules != 1 || a.Bytes != 42 || a.Format != "hosts" {
+		t.Errorf("SourceMeta(\"a\") = %+v, ok=%v", a, ok)
+	}
+	if _, ok := s.SourceMeta("nope"); ok {
+		t.Errorf("SourceMeta(\"nope\") reported ok")
+	}
+	if c := s.Counts(); c.Exact != 1 || c.Sub != 1 || c.Root != 1 || c.Block != 3 {
+		t.Errorf("Counts() = %+v, want 1/1/1 exact/sub/root and 3 block", c)
+	}
+	st := s.Stats()
+	if st.Total != 3 || st.Exact != 1 || st.Subdomains != 1 || st.RootRules != 1 {
+		t.Errorf("Stats() = %+v", st)
+	}
+	if st.BuiltAt.IsZero() {
+		t.Errorf("Stats().BuiltAt is zero")
+	}
+
+	// Each renders patterns back with their wildcard prefix.
+	seen := map[string]PatternKind{}
+	s.Each(func(pattern string, e *Entry, kind PatternKind) bool {
+		seen[pattern] = kind
+		if got := kind.Pattern(e.Name); got != pattern {
+			t.Errorf("Pattern(%q) = %q, want %q", e.Name, got, pattern)
+		}
+		return true
+	})
+	if seen["exact.example.com"] != PatternExact || seen["sub.example.com"] != PatternSub ||
+		seen["root.example.com"] != PatternRoot {
+		t.Errorf("Each() visited %v", seen)
+	}
+	if PatternSub.String() != "subdomain" || PatternRoot.String() != "domain" || PatternExact.String() != "exact" {
+		t.Errorf("PatternKind.String() values changed")
+	}
+
+	// Lookup is exact-only: it does not walk wildcards.
+	if s.Lookup("deep.sub.example.com") != nil {
+		t.Errorf("Lookup() walked wildcards")
+	}
+	if s.Lookup("EXACT.example.com.") == nil {
+		t.Errorf("Lookup() is not normalised")
+	}
+}
+
+func TestDigestStableAndContentSensitive(t *testing.T) {
+	a := mustBuild(block("x.example.com"), block("*.y.example.com"), block("**.z.example.com"))
+	b := mustBuild(block("**.z.example.com"), block("x.example.com"), block("*.y.example.com"))
+	if a.Digest() != b.Digest() {
+		t.Errorf("Digest() depends on insertion order: %q vs %q", a.Digest(), b.Digest())
+	}
+	c := mustBuild(block("x.example.com"), block("*.y.example.com"), block("**.different.example.com"))
+	if a.Digest() == c.Digest() {
+		t.Errorf("Digest() ignored a pattern change")
+	}
+	// The action is part of the digest.
+	d := mustBuild(Rule{Name: "x.example.com", Action: ActionAllow})
+	if a.Digest() == d.Digest() {
+		t.Errorf("Digest() ignored the action change")
+	}
+}
+
+func TestNilSetIsSafe(t *testing.T) {
+	var s *Set
+	if s.Match("example.com") != nil || s.Len() != 0 || s.HasWildcards() ||
+		s.DuplicatesN() != 0 || s.Sources() != nil || s.AllSourceMeta() != nil ||
+		s.Digest() != "" || !s.BuiltAt().IsZero() {
+		t.Errorf("nil *Set accessors are not nil-safe")
+	}
+	if _, ok := s.SourceMeta("x"); ok {
+		t.Errorf("nil Set SourceMeta reported ok")
+	}
+	if s.Lookup("x") != nil || s.MatchPattern("x") != PatternExact {
+		t.Errorf("nil Set lookup misbehaved")
+	}
+	if st := s.Stats(); st.Total != 0 || st.PerSource == nil {
+		t.Errorf("nil Stats() = %+v", st)
+	}
+	s.Each(func(string, *Entry, PatternKind) bool { t.Errorf("nil Each called fn"); return true })
+}
+
+func TestActionStringAndParseAction(t *testing.T) {
+	all := []struct {
+		a   Action
+		str string
+	}{
+		{ActionBlock, "block"},
+		{ActionAllow, "allow"},
+		{ActionPassthrough, "passthrough"},
+		{ActionDrop, "drop"},
+		{ActionNXDomain, "nxdomain"},
+		{ActionZero, "zero"},
+		{ActionCNAME, "cname"},
+		{ActionRedirect, "redirect"},
+		{Action(99), "unknown"},
+	}
+	for _, tc := range all {
+		if got := tc.a.String(); got != tc.str {
+			t.Errorf("Action(%d).String() = %q, want %q", tc.a, got, tc.str)
+		}
+	}
+	roundtrip := map[string]Action{
+		"":            ActionBlock,
+		"block":       ActionBlock,
+		"ALLOW":       ActionAllow,
+		" whitelist ": ActionAllow,
+		"pass":        ActionPassthrough,
+		"drop":        ActionDrop,
+		"nxdomain":    ActionNXDomain,
+		"null":        ActionZero,
+		"cname":       ActionCNAME,
+		"redirect":    ActionRedirect,
+	}
+	for in, want := range roundtrip {
+		got, ok := ParseAction(in)
+		if !ok || got != want {
+			t.Errorf("ParseAction(%q) = %v,%v; want %v,true", in, got, ok, want)
+		}
+	}
+	if got, ok := ParseAction("nonsense"); ok || got != ActionBlock {
+		t.Errorf("ParseAction(\"nonsense\") = %v,%v; want block,false", got, ok)
+	}
+}
+
+func TestHolderSwapIsAtomic(t *testing.T) {
+	old := mustBuild(block("a.example.com"))
+	newer := mustBuild(block("b.example.com"))
+	h := NewHolder(old)
+	if h.Get() != old {
+		t.Fatalf("NewHolder did not install the set")
+	}
+	var empty *Holder
+	if empty.Get() != nil {
+		t.Errorf("zero Holder is not empty")
+	}
+	// A reader either sees the whole old set or the whole new one.
+	var wg sync.WaitGroup
+	stop := make(chan struct{})
+	for i := 0; i < 4; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for {
+				select {
+				case <-stop:
+					return
+				default:
+				}
+				s := h.Get()
+				if s == nil {
+					t.Error("holder returned nil")
+					return
+				}
+				if e := s.Match("a.example.com"); e == nil {
+					t.Error("observed a half-applied set")
+					return
+				}
+			}
+		}()
+	}
+	for i := 0; i < 50; i++ {
+		h.Set(newer)
+		h.Set(old)
+	}
+	close(stop)
+	wg.Wait()
+	if h.Get() != old {
+		t.Errorf("holder lost the last swap")
+	}
+}
+
+func BenchmarkSetMatch(b *testing.B) {
+	// A mixed set, like a compiled blocklist: mostly exact rules plus some
+	// subdomain and root rules so the suffix walk is exercised.
+	bld := NewBuilder()
+	for i := 0; i < 2000; i++ {
+		bld.Add(block(fmt.Sprintf("host%d.ads-%d.example.com", i, i%7)), "bench", nil)
+	}
+	for i := 0; i < 200; i++ {
+		bld.Add(block(fmt.Sprintf("*.wild%d.example.net", i)), "bench", nil)
+		bld.Add(block(fmt.Sprintf("**.root%d.example.org", i)), "bench", nil)
+	}
+	s := bld.Build()
+
+	queries := []string{
+		"host0.ads-0.example.com",    // exact hit
+		"host1999.ads-5.example.com", // exact hit, different list
+		"a.b.wild199.example.net",    // wildcard suffix hit
+		"x.root42.example.org",       // root rule
+		"nothing.in.here.test",       // pure miss, walks every suffix
+	}
+	b.ReportAllocs()
+	b.ResetTimer()
+	var sink int
+	for i := 0; i < b.N; i++ {
+		if e := s.Match(queries[i%len(queries)]); e != nil {
+			sink++
+		}
+	}
+	if sink == 0 {
+		b.Fatal("no matches at all")
+	}
+}

internal/cache/cache.go

@@ -18,6 +18,7 @@ package cache
 import (
 	"encoding/binary"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/miekg/dns"
@@ -89,7 +90,9 @@ func canonical(name string) string {
 //
 //	varint len | wire name | uint16 type | uint16 class | uint8 flags
 func packKey(buf []byte, k Key) ([]byte, error) {
-	name := k.Name
+	// Keys are stored canonically (lowercase, no trailing dot) but the wire
+	// format needs a fully qualified name, so qualify it before packing.
+	name := dns.Fqdn(k.Name)
 	if n := len(name); n > 255*4 {
 		return nil, errKeyTooLong
 	}
@@ -160,8 +163,11 @@ type Entry struct {
 	StaleUntil time.Time
 	// Bytes is the accounted size, used by the LRU and the size gauges.
 	Bytes int
-	// Hits counts serves since it was stored.
+	// Hits counts serves since it was stored. It is atomic because an Entry is
-	Hits uint32
+	// shared by every goroutine that hits the key (the LRU hands out the same
+	// pointer on purpose, so a hit never copies the message), and the
+	// counter is bumped outside the shard lock on the hot path.
+	Hits atomic.Uint32
 	// Neg marks a cached NXDOMAIN or NODATA answer.
 	Neg bool
 	// Truncated records that the upstream answer did not fit in 512 bytes,

internal/cache/cache_test.go

@@ -0,0 +1,870 @@
+package cache
+
+import (
+	"errors"
+	"math/rand/v2"
+	"net"
+	"os"
+	"path/filepath"
+	"strconv"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/miekg/dns"
+)
+
+// fakeClock is a manually advanced clock so TTL, ageing and serve-stale
+// behaviour is tested without sleeping.
+type fakeClock struct {
+	mu sync.Mutex
+	t  time.Time
+}
+
+func newClock() *fakeClock { return &fakeClock{t: time.Date(2024, 5, 1, 12, 0, 0, 0, time.UTC)} }
+
+func (c *fakeClock) Now() time.Time {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.t
+}
+
+func (c *fakeClock) Advance(d time.Duration) {
+	c.mu.Lock()
+	c.t = c.t.Add(d)
+	c.mu.Unlock()
+}
+
+// answer builds a packed A response for name.
+func answer(t *testing.T, name string, ttl uint32, ip string) *dns.Msg {
+	t.Helper()
+	m := new(dns.Msg)
+	m.SetQuestion(dns.Fqdn(name), dns.TypeA)
+	m.Response = true
+	m.Answer = []dns.RR{&dns.A{
+		Hdr: dns.RR_Header{Name: dns.Fqdn(name), Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: ttl},
+		A:   net.ParseIP(ip).To4(),
+	}}
+	return m
+}
+
+func key(name string) Key { return QueryKey(name, dns.TypeA, dns.ClassINET, false, 0) }
+
+func TestLRUGetPutAndHitKinds(t *testing.T) {
+	clk := newClock()
+	l := NewLRU(100, 4)
+	m := answer(t, "example.com", 60, "1.2.3.4")
+	wire, err := m.Pack()
+	if err != nil {
+		t.Fatalf("pack: %v", err)
+	}
+	k := key("example.com")
+	now := clk.Now()
+	l.Put(k, &Entry{Wire: wire, TTL: 60, CreatedAt: now, ExpiresAt: now.Add(time.Minute), StaleUntil: now.Add(90 * time.Second), Bytes: len(wire)})
+
+	if e, kind := l.Get(k, now); e == nil || kind != Fresh {
+		t.Fatalf("fresh hit: got entry=%v kind=%v, want non-nil/Fresh", e, kind)
+	}
+	// One second later the entry is still fresh; after the TTL it is stale but
+	// answerable until StaleUntil, and past that it is a miss and dropped.
+	clk.Advance(59 * time.Second)
+	if _, kind := l.Get(k, clk.Now()); kind != Fresh {
+		t.Errorf("at TTL-1s: kind = %v, want Fresh", kind)
+	}
+	clk.Advance(2 * time.Second)
+	e, kind := l.Get(k, clk.Now())
+	if e == nil || kind != Stale {
+		t.Fatalf("just past TTL: got entry=%v kind=%v, want non-nil/Stale", e, kind)
+	}
+	if got := e.Hits.Load(); got != 3 {
+		t.Errorf("entry Hits = %d, want 3 (every Get counts, fresh or stale)", got)
+	}
+	clk.Advance(31 * time.Second) // now past StaleUntil (TTL+30s)
+	if e, kind := l.Get(k, clk.Now()); e != nil || kind != Miss {
+		t.Fatalf("past stale window: got entry=%v kind=%v, want nil/Miss", e, kind)
+	}
+	if got := l.Stats().Entries; got != 0 {
+		t.Errorf("entries after a dead read = %d, want 0 (the entry must be dropped)", got)
+	}
+	st := l.Stats()
+	if st.Hits != 2 || st.StaleHits != 1 || st.Misses != 1 {
+		t.Errorf("stats = %+v, want hits=2 staleHits=1 misses=1", st)
+	}
+}
+
+func TestLRUEvictionKeepsRecentlyUsed(t *testing.T) {
+	// One shard so the budget is exactly what we set: eviction order is then
+	// fully determined and the test can assert it.
+	l := NewLRU(4, 1)
+	clk := newClock()
+	now := clk.Now()
+	put := func(name string) {
+		l.Put(key(name), &Entry{TTL: 60, CreatedAt: now, ExpiresAt: now.Add(time.Minute), StaleUntil: now.Add(time.Minute), Bytes: 8})
+	}
+	put("a.example.com")
+	put("b.example.com")
+	put("c.example.com")
+	// Touch a so it is the most recent, then overflow the 4-entry budget.
+	if _, kind := l.Get(key("a.example.com"), now); kind != Fresh {
+		t.Fatalf("a: kind = %v, want Fresh", kind)
+	}
+	put("d.example.com")
+	put("e.example.com") // 5 inserts, budget 4
+
+	if got := l.Stats().Entries; got > 4 {
+		t.Fatalf("entries = %d, want <= 4 (per-shard budget)", got)
+	}
+	if _, kind := l.Get(key("a.example.com"), now); kind != Fresh {
+		t.Errorf("a was recently used, want it kept")
+	}
+	if _, kind := l.Get(key("b.example.com"), now); kind != Miss {
+		t.Errorf("b was the least recently used, want it evicted")
+	}
+	if l.Stats().Evictions == 0 {
+		t.Error("evictions = 0, want > 0")
+	}
+}
+
+func TestLRUDeleteWhereAndFlush(t *testing.T) {
+	l := NewLRU(100, 2)
+	now := newClock().Now()
+	names := []string{"x.example.com", "y.example.com", "z.other.net"}
+	for _, n := range names {
+		l.Put(key(n), &Entry{TTL: 30, CreatedAt: now, ExpiresAt: now.Add(30 * time.Second), StaleUntil: now.Add(time.Minute), Bytes: 4})
+	}
+	dropped := l.DeleteWhere(func(k Key, _ *Entry) bool {
+		return k.Name == "x.example.com" || k.Name == "y.example.com"
+	})
+	if dropped != 2 {
+		t.Errorf("DeleteWhere = %d, want 2", dropped)
+	}
+	if _, kind := l.Get(key("x.example.com"), now); kind != Miss {
+		t.Error("x.example.com still present after DeleteWhere")
+	}
+	if _, kind := l.Get(key("z.other.net"), now); kind != Fresh {
+		t.Error("z.other.net must survive an unrelated DeleteWhere")
+	}
+	l.Flush()
+	if got := l.Stats().Entries; got != 0 {
+		t.Errorf("entries after Flush = %d, want 0", got)
+	}
+	if l.Stats().Bytes != 0 {
+		t.Errorf("bytes after Flush = %d, want 0", l.Stats().Bytes)
+	}
+}
+
+func TestLRUWalkVisitsEveryEntryOnce(t *testing.T) {
+	l := NewLRU(400, 4) // budget is per shard: 100 entries per shard
+	now := newClock().Now()
+	for i := range 200 {
+		n := "n" + strconv.Itoa(i) + ".example.com"
+		l.Put(key(n), &Entry{TTL: 30, CreatedAt: now, ExpiresAt: now.Add(30 * time.Second), StaleUntil: now.Add(time.Minute), Bytes: 4})
+	}
+	seen := make(map[string]int, 200)
+	l.Walk(func(k Key, _ *Entry) bool {
+		seen[k.Name]++
+		return true
+	})
+	if len(seen) != 200 {
+		t.Fatalf("Walk visited %d names, want 200", len(seen))
+	}
+	for n, c := range seen {
+		if c != 1 {
+			t.Fatalf("%s visited %d times, want 1", n, c)
+		}
+	}
+	// Returning false stops the walk.
+	count := 0
+	l.Walk(func(Key, *Entry) bool { count++; return false })
+	if count != 1 {
+		t.Errorf("Walk with stop visited %d entries, want 1", count)
+	}
+}
+
+func TestSetCapacityEvictsDownToTheNewBudget(t *testing.T) {
+	l := NewLRU(100, 1)
+	now := newClock().Now()
+	for i := range 50 {
+		n := "n" + strconv.Itoa(i) + ".example.com"
+		l.Put(key(n), &Entry{TTL: 30, CreatedAt: now, ExpiresAt: now.Add(30 * time.Second), StaleUntil: now.Add(time.Minute), Bytes: 4})
+	}
+	l.SetCapacity(10, 1)
+	if got := l.Stats().Entries; got > 10 {
+		t.Errorf("entries after SetCapacity(10) = %d, want <= 10", got)
+	}
+	if got := l.Stats().Shards; got != 1 {
+		t.Errorf("shards after SetCapacity(10, 1) = %d, want 1", got)
+	}
+	// The cache must still work after a resize.
+	l.Put(key("after-resize.example.com"), &Entry{TTL: 30, CreatedAt: now, ExpiresAt: now.Add(30 * time.Second), StaleUntil: now.Add(time.Minute), Bytes: 4})
+	if _, kind := l.Get(key("after-resize.example.com"), now); kind != Fresh {
+		t.Error("a Put after a resize must be readable")
+	}
+}
+
+func TestLRUConcurrentPutGetResize(t *testing.T) {
+	// Exercises the resizing path: SetCapacity rewrites the per-shard budget
+	// while other goroutines insert and read. Run with -race.
+	l := NewLRU(512, 8)
+	now := newClock().Now()
+	var wg sync.WaitGroup
+	stop := make(chan struct{})
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		for {
+			select {
+			case <-stop:
+				return
+			default:
+				l.SetCapacity(200, 4)
+				l.SetCapacity(4096, 16)
+			}
+		}
+	}()
+	for g := range 8 {
+		wg.Add(1)
+		go func(g int) {
+			defer wg.Done()
+			for i := range 500 {
+				n := "g" + strconv.Itoa(g) + "-n" + strconv.Itoa(i) + ".example.com"
+				k := key(n)
+				l.Put(k, &Entry{TTL: 30, CreatedAt: now, ExpiresAt: now.Add(30 * time.Second), StaleUntil: now.Add(time.Minute), Bytes: 4})
+				e, _ := l.Get(k, now)
+				if e != nil {
+					_ = e.Hits.Load()
+				}
+			}
+		}(g)
+	}
+	// Readers keep walking while the resizer runs.
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		for {
+			select {
+			case <-stop:
+				return
+			default:
+				l.Walk(func(Key, *Entry) bool { return true })
+			}
+		}
+	}()
+	// Let the background loops run, then release them and wait for every
+	// goroutine: closing before the wait is what keeps this from deadlocking.
+	time.Sleep(20 * time.Millisecond)
+	close(stop)
+	wg.Wait()
+	if l.Stats().Bytes < 0 {
+		t.Errorf("byte accounting went negative: %d", l.Stats().Bytes)
+	}
+}
+
+func TestKeyPackUnpackRoundTrip(t *testing.T) {
+	cases := []Key{
+		{Name: "example.com", Type: dns.TypeA, Class: dns.ClassINET},
+		{Name: "a.b.c.d.example.com", Type: dns.TypeAAAA, Class: dns.ClassINET, DO: true},
+		{Name: "x", Type: dns.TypeMX, Class: dns.ClassCHAOS, Scope: 3},
+		{Name: "wildcard.example.com", Type: dns.TypeTXT, Class: dns.ClassINET, DO: true, Scope: 7},
+	}
+	for _, k := range cases {
+		buf, err := packKey(nil, k)
+		if err != nil {
+			t.Fatalf("packKey(%v): %v", k, err)
+		}
+		got, rest, err := unpackKey(buf)
+		if err != nil {
+			t.Fatalf("unpackKey(%v): %v", k, err)
+		}
+		if len(rest) != 0 {
+			t.Errorf("unpackKey(%v) left %d trailing bytes, want 0", k, len(rest))
+		}
+		if got != k {
+			t.Errorf("round trip: got %+v, want %+v", got, k)
+		}
+	}
+	// The wire name survives, so the canonical form comes back normalised.
+	buf, err := packKey(nil, Key{Name: "Example.COM", Type: dns.TypeA, Class: dns.ClassINET})
+	if err != nil {
+		t.Fatalf("packKey: %v", err)
+	}
+	got, _, err := unpackKey(buf)
+	if err != nil {
+		t.Fatalf("unpackKey: %v", err)
+	}
+	if got.Name != "example.com" {
+		t.Errorf("unpacked name = %q, want %q", got.Name, "example.com")
+	}
+}
+
+func TestKeyPackRejectsBadInput(t *testing.T) {
+	if _, err := packKey(nil, Key{Name: string(make([]byte, 256*4+1))}); !errors.Is(err, errKeyTooLong) {
+		t.Errorf("packKey(oversized name) error = %v, want errKeyTooLong", err)
+	}
+	if _, err := packKey(nil, Key{Name: "a..b", Type: dns.TypeA}); err == nil {
+		t.Error("packKey(name with an empty label) error = nil, want an error")
+	}
+	if _, _, err := unpackKey([]byte{0x05, 0x00}); !errors.Is(err, errBadKey) {
+		t.Errorf("unpackKey(truncated) error = %v, want errBadKey", err)
+	}
+	if _, _, err := unpackKey(nil); !errors.Is(err, errBadKey) {
+		t.Errorf("unpackKey(empty) error = %v, want errBadKey", err)
+	}
+}
+
+func TestQueryKeyCanonicalises(t *testing.T) {
+	a := QueryKey("Example.COM.", dns.TypeA, dns.ClassINET, false, 0)
+	b := QueryKey("example.com", dns.TypeA, dns.ClassINET, false, 0)
+	if a != b {
+		t.Errorf("QueryKey did not canonicalise: %+v != %+v", a, b)
+	}
+	if a.Name != "example.com" {
+		t.Errorf("canonical name = %q, want %q", a.Name, "example.com")
+	}
+}
+
+func TestEntryAgeTTL(t *testing.T) {
+	now := time.Date(2024, 1, 1, 0, 0, 0, 0, time.UTC)
+	cases := []struct {
+		name string
+		e    Entry
+		now  time.Time
+		want uint32
+	}{
+		{"fresh", Entry{TTL: 100, CreatedAt: now}, now.Add(30 * time.Second), 70},
+		{"expiring", Entry{TTL: 100, CreatedAt: now}, now.Add(100 * time.Second), 0},
+		{"past", Entry{TTL: 100, CreatedAt: now}, now.Add(500 * time.Second), 0},
+		{"zero created keeps the stored TTL", Entry{TTL: 100}, now, 100},
+		{"sub-second ageing truncates to 0 elapsed", Entry{TTL: 1, CreatedAt: now}, now.Add(400 * time.Millisecond), 1},
+	}
+	for i := range cases {
+		// Take the address: Entry now carries an atomic counter, so the
+		// struct must never be copied.
+		tc := &cases[i]
+		got := tc.e.AgeTTL(tc.now)
+		if got != tc.want {
+			t.Errorf("%s: AgeTTL = %d, want %d", tc.name, got, tc.want)
+		}
+	}
+}
+
+func TestManagerTTLFor(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{MinTTL: 0, MaxTTL: time.Hour, NegativeTTL: 30 * time.Second, Now: clk.Now}, nil)
+
+	soa := func(ttl uint32) *dns.SOA {
+		return &dns.SOA{Hdr: dns.RR_Header{Name: "example.com.", Rrtype: dns.TypeSOA, Class: dns.ClassINET, Ttl: ttl}, Minttl: ttl}
+	}
+
+	t.Run("answers use the lowest TTL and clamp", func(t *testing.T) {
+		m := NewManager(Options{MinTTL: 10 * time.Second, MaxTTL: time.Hour, Now: clk.Now}, nil)
+		msg := answer(t, "a.example.com", 300, "1.2.3.4")
+		msg.Answer = append(msg.Answer, &dns.A{
+			Hdr: dns.RR_Header{Name: "a.example.com.", Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 5},
+			A:   net.ParseIP("1.2.3.5").To4(),
+		})
+		ttl, neg := m.TTLFor(msg)
+		if neg {
+			t.Error("neg = true for a positive answer")
+		}
+		if ttl != 10 {
+			t.Errorf("ttl = %d, want 10 (5s clamped up to MinTTL)", ttl)
+		}
+	})
+
+	t.Run("clamp down to MaxTTL", func(t *testing.T) {
+		m := NewManager(Options{MaxTTL: 60 * time.Second, Now: clk.Now}, nil)
+		ttl, _ := m.TTLFor(answer(t, "a.example.com", 86400, "1.2.3.4"))
+		if ttl != 60 {
+			t.Errorf("ttl = %d, want 60 (clamped to MaxTTL)", ttl)
+		}
+	})
+
+	t.Run("NXDOMAIN takes the SOA TTL", func(t *testing.T) {
+		m := NewManager(Options{NegativeTTL: 30 * time.Second, Now: clk.Now}, nil)
+		msg := new(dns.Msg)
+		msg.SetQuestion("nope.example.com.", dns.TypeA)
+		msg.Response = true
+		msg.Rcode = dns.RcodeNameError
+		msg.Ns = []dns.RR{soa(45)}
+		ttl, neg := m.TTLFor(msg)
+		if !neg {
+			t.Error("neg = false for NXDOMAIN")
+		}
+		if ttl != 45 {
+			t.Errorf("ttl = %d, want 45", ttl)
+		}
+	})
+
+	t.Run("NODATA takes the SOA TTL and counts as negative", func(t *testing.T) {
+		m := NewManager(Options{NegativeTTL: 30 * time.Second, Now: clk.Now}, nil)
+		msg := new(dns.Msg)
+		msg.SetQuestion("nope.example.com.", dns.TypeA)
+		msg.Response = true
+		msg.Ns = []dns.RR{soa(12)}
+		ttl, neg := m.TTLFor(msg)
+		if !neg || ttl != 12 {
+			t.Errorf("NODATA: ttl = %d neg = %v, want 12/true", ttl, neg)
+		}
+	})
+
+	t.Run("NODATA without an SOA falls back to NegativeTTL", func(t *testing.T) {
+		m := NewManager(Options{NegativeTTL: 30 * time.Second, Now: clk.Now}, nil)
+		msg := new(dns.Msg)
+		msg.SetQuestion("nope.example.com.", dns.TypeA)
+		msg.Response = true
+		ttl, neg := m.TTLFor(msg)
+		if !neg || ttl != 30 {
+			t.Errorf("ttl = %d neg = %v, want 30/true", ttl, neg)
+		}
+	})
+
+	t.Run("truncated answers are never cached", func(t *testing.T) {
+		msg := answer(t, "big.example.com", 600, "1.2.3.4")
+		msg.Truncated = true
+		if ttl, _ := m.TTLFor(msg); ttl != 0 {
+			t.Errorf("ttl = %d for a truncated answer, want 0", ttl)
+		}
+	})
+
+	t.Run("transient rcodes are never cached", func(t *testing.T) {
+		for _, rc := range []int{dns.RcodeServerFailure, dns.RcodeRefused, dns.RcodeNotAuth} {
+			msg := answer(t, "a.example.com", 600, "1.2.3.4")
+			msg.Rcode = rc
+			if ttl, _ := m.TTLFor(msg); ttl != 0 {
+				t.Errorf("rcode %d: ttl = %d, want 0", rc, ttl)
+			}
+		}
+	})
+
+	t.Run("nil message", func(t *testing.T) {
+		if ttl, _ := m.TTLFor(nil); ttl != 0 {
+			t.Errorf("ttl = %d, want 0", ttl)
+		}
+	})
+}
+
+func TestManagerPutResponseGetAndAgeing(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, Now: clk.Now}, nil)
+	k := key("www.example.com")
+	m.PutResponse(k, answer(t, "www.example.com", 120, "93.184.216.34"), "127.0.0.53:53")
+
+	res := m.Get(k)
+	if res.Kind != Fresh {
+		t.Fatalf("kind = %v, want Fresh", res.Kind)
+	}
+	if res.TTL != 120 {
+		t.Errorf("TTL = %d, want 120", res.TTL)
+	}
+	if res.Upstream != "127.0.0.53:53" {
+		t.Errorf("upstream = %q, want the address it came from", res.Upstream)
+	}
+	if got := res.Msg.Answer[0].(*dns.A).A.String(); got != "93.184.216.34" {
+		t.Errorf("answer = %s, want 93.184.216.34", got)
+	}
+
+	// The stored copy is private: mutating the returned message must not
+	// change what the next reader sees.
+	res.Msg.Answer[0].(*dns.A).A = net.ParseIP("9.9.9.9").To4()
+	again := m.Get(k)
+	if got := again.Msg.Answer[0].(*dns.A).A.String(); got != "93.184.216.34" {
+		t.Errorf("answer after mutating the first result = %s, want an independent copy", got)
+	}
+
+	// TTLs age down monotonically.
+	clk.Advance(30 * time.Second)
+	if got := m.Get(k).TTL; got != 90 {
+		t.Errorf("TTL after 30s = %d, want 90", got)
+	}
+	if got := m.Get(k).Msg.Answer[0].Header().Ttl; got != 90 {
+		t.Errorf("record TTL after 30s = %d, want 90", got)
+	}
+}
+
+func TestManagerServeStale(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, ServeStale: time.Minute, Now: clk.Now}, nil)
+	k := key("stale.example.com")
+	m.PutResponse(k, answer(t, "stale.example.com", 60, "10.0.0.1"), "up")
+
+	clk.Advance(90 * time.Second) // 30s past the TTL, inside the stale window
+	res := m.Get(k)
+	if res.Kind != Stale {
+		t.Fatalf("kind = %v, want Stale", res.Kind)
+	}
+	if !res.ServedStale {
+		t.Error("ServedStale = false on a stale hit")
+	}
+	if res.TTL != 0 {
+		t.Errorf("TTL = %d on a stale hit, want 0 (the handler forces TTL 1)", res.TTL)
+	}
+	clk.Advance(61 * time.Second) // now past TTL+ServeStale
+	if res := m.Get(k); res.Kind != Miss {
+		t.Errorf("kind = %v past the stale window, want Miss", res.Kind)
+	}
+}
+
+func TestManagerServeStaleDisabled(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, Now: clk.Now}, nil)
+	k := key("nostale.example.com")
+	m.PutResponse(k, answer(t, "nostale.example.com", 60, "10.0.0.2"), "up")
+	clk.Advance(61 * time.Second)
+	if res := m.Get(k); res.Kind != Miss {
+		t.Errorf("kind = %v, want Miss when ServeStale is 0", res.Kind)
+	}
+}
+
+func TestManagerPutResponseRespectsZeroTTL(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, Now: clk.Now}, nil)
+	k := key("nocache.example.com")
+	m.PutResponse(k, answer(t, "nocache.example.com", 0, "10.0.0.3"), "up")
+	if res := m.Get(k); res.Kind != Miss {
+		t.Errorf("kind = %v, want Miss: a zero TTL means do not cache", res.Kind)
+	}
+}
+
+func TestManagerDeleteNameAndSubtree(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, Now: clk.Now}, nil)
+	names := []string{"blocked.example.com", "deep.sub.blocked.example.com", "keep.example.com", "notblocked.example.com"}
+	for _, n := range names {
+		m.PutResponse(key(n), answer(t, n, 600, "10.0.0.4"), "up")
+	}
+	// "sub.blocked.example.com" must not be caught by the "blocked.example.com"
+	// subtree walk, but "blocked.example.com" must not catch
+	// "notblocked.example.com" either.
+	if n := m.DeleteName("Blocked.Example.Com."); n != 1 {
+		t.Errorf("DeleteName = %d, want 1 (canonicalised, one name)", n)
+	}
+	if res := m.Get(key("blocked.example.com")); res.Kind != Miss {
+		t.Error("deleted name still served")
+	}
+	if n := m.DeleteSubtree("deep.sub.blocked.example.com"); n != 1 {
+		t.Errorf("DeleteSubtree = %d, want 1", n)
+	}
+	for _, n := range []string{"keep.example.com", "notblocked.example.com"} {
+		if res := m.Get(key(n)); res.Kind != Fresh {
+			t.Errorf("%s was dropped by an unrelated delete", n)
+		}
+	}
+}
+
+func TestManagerTopEntriesCountsAndFlush(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, Now: clk.Now}, nil)
+	for range 5 {
+		m.PutResponse(key("hot.example.com"), answer(t, "hot.example.com", 300, "10.0.0.5"), "up")
+	}
+	for range 2 {
+		m.PutResponse(key("warm.example.com"), answer(t, "warm.example.com", 300, "10.0.0.6"), "up")
+	}
+	top := m.TopEntries(10)
+	if len(top) < 2 {
+		t.Fatalf("TopEntries returned %d rows, want at least 2", len(top))
+	}
+	// Each Put counts too, so a name queried once still shows up.
+	var hot, warm uint64
+	for _, e := range top {
+		switch e.Name {
+		case "hot.example.com":
+			hot = e.Hits
+		case "warm.example.com":
+			warm = e.Hits
+		}
+	}
+	if hot < 5 || warm < 2 {
+		t.Errorf("top hits: hot=%d warm=%d, want >=5 and >=2", hot, warm)
+	}
+	if top[0].Name != "hot.example.com" {
+		t.Errorf("first row = %s, want the hottest name first", top[0].Name)
+	}
+	m.Flush()
+	if got := len(m.TopEntries(10)); got != 0 {
+		t.Errorf("TopEntries after Flush = %d rows, want 0", got)
+	}
+}
+
+func TestManagerStatsShape(t *testing.T) {
+	clk := newClock()
+	m := NewManager(Options{Size: 100, Shards: 4, MaxTTL: time.Hour, MinTTL: time.Second, Now: clk.Now}, nil)
+	m.PutResponse(key("a.example.com"), answer(t, "a.example.com", 60, "10.0.0.7"), "up")
+	m.Get(key("a.example.com"))
+	m.Get(key("miss.example.com"))
+	s := m.Stats()
+	if !s.Enabled {
+		t.Error("Enabled = false on a live manager")
+	}
+	if s.Entries != 1 {
+		t.Errorf("Entries = %d, want 1", s.Entries)
+	}
+	if s.Hits != 1 || s.Misses != 1 {
+		t.Errorf("hits = %d misses = %d, want 1/1", s.Hits, s.Misses)
+	}
+	if s.Size != 100 || s.Shards != 4 {
+		t.Errorf("size/shards = %d/%d, want 100/4", s.Size, s.Shards)
+	}
+	if s.MemoryHuman == "" || s.Bytes <= 0 {
+		t.Errorf("memory accounting: bytes = %d human = %q, want a positive byte count and a human string", s.Bytes, s.MemoryHuman)
+	}
+	if s.Persist.Enabled {
+		t.Error("Persist.Enabled = true without a store")
+	}
+}
+
+func TestHumanBytes(t *testing.T) {
+	cases := []struct {
+		in   int64
+		want string
+	}{
+		{0, "0 B"},
+		{512, "512 B"},
+		{1024, "1.0 KiB"},
+		{1536, "1.5 KiB"},
+		{1024 * 1024, "1.0 MiB"},
+		{3 * 1024 * 1024 * 1024 / 2, "1.50 GiB"},
+		{-1024, "-1.0 KiB"},
+	}
+	for _, tc := range cases {
+		if got := HumanBytes(tc.in); got != tc.want {
+			t.Errorf("HumanBytes(%d) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestStoreSnapshotAndLoad(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "cache.db")
+	st, err := OpenStore(StoreOptions{Path: path})
+	if err != nil {
+		t.Fatalf("OpenStore: %v", err)
+	}
+	defer st.Close()
+
+	clk := newClock()
+	now := clk.Now()
+	l := NewLRU(100, 2)
+	for i := range 25 {
+		n := "n" + strconv.Itoa(i) + ".example.com"
+		wire, err := answer(t, n, 600, "10.0.0.8").Pack()
+		if err != nil {
+			t.Fatalf("pack: %v", err)
+		}
+		l.Put(key(n), &Entry{
+			Wire: wire, TTL: 600, Origin: "upstream", Upstream: "1.1.1.1:53",
+			CreatedAt: now, ExpiresAt: now.Add(600 * time.Second), StaleUntil: now.Add(630 * time.Second), Bytes: len(wire),
+		})
+	}
+	written, err := st.Snapshot(l, now)
+	if err != nil {
+		t.Fatalf("Snapshot: %v", err)
+	}
+	if written != 25 {
+		t.Errorf("Snapshot wrote %d entries, want 25", written)
+	}
+	if fi, err := os.Stat(path); err != nil || fi.Size() == 0 {
+		t.Fatalf("snapshot file missing or empty: %v", err)
+	}
+
+	// A fresh LRU restored from the file keeps the answers and their metadata.
+	restored := NewLRU(100, 2)
+	loaded, skipped, err := st.Load(restored, now)
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	if loaded != 25 || skipped != 0 {
+		t.Errorf("Load = %d loaded, %d skipped, want 25/0", loaded, skipped)
+	}
+	e, kind := restored.Get(key("n7.example.com"), now)
+	if e == nil || kind != Fresh {
+		t.Fatalf("restored entry: e = %v kind = %v, want non-nil/Fresh", e, kind)
+	}
+	if e.Upstream != "1.1.1.1:53" || e.Origin != "upstream" {
+		t.Errorf("restored metadata: origin = %q upstream = %q, want upstream/1.1.1.1:53", e.Origin, e.Upstream)
+	}
+	// The wire is intact.
+	msg := new(dns.Msg)
+	if err := msg.Unpack(e.Wire); err != nil {
+		t.Fatalf("unpack restored wire: %v", err)
+	}
+	if len(msg.Answer) != 1 {
+		t.Errorf("restored answer has %d records, want 1", len(msg.Answer))
+	}
+}
+
+func TestStoreLoadSkipsExpiredEntries(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "cache.db")
+	st, err := OpenStore(StoreOptions{Path: path})
+	if err != nil {
+		t.Fatalf("OpenStore: %v", err)
+	}
+	defer st.Close()
+
+	clk := newClock()
+	now := clk.Now()
+	l := NewLRU(100, 1)
+	for i, ttl := range []uint32{600, 600, 5} {
+		n := "n" + strconv.Itoa(i) + ".example.com"
+		wire, _ := answer(t, n, ttl, "10.0.0.9").Pack()
+		l.Put(key(n), &Entry{
+			Wire: wire, TTL: ttl, CreatedAt: now, ExpiresAt: now.Add(time.Duration(ttl) * time.Second),
+			StaleUntil: now.Add(time.Duration(ttl)*time.Second + 30*time.Second), Bytes: len(wire),
+		})
+	}
+	if _, err := st.Snapshot(l, now); err != nil {
+		t.Fatalf("Snapshot: %v", err)
+	}
+	// Load 10s later: the 5s entry is dead, the two 600s entries are fresh.
+	restored := NewLRU(100, 1)
+	loaded, skipped, err := st.Load(restored, now.Add(10*time.Second))
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	if loaded != 2 {
+		t.Errorf("loaded = %d, want 2", loaded)
+	}
+	if skipped != 1 {
+		t.Errorf("skipped = %d, want 1", skipped)
+	}
+	if _, kind := restored.Get(key("n2.example.com"), now.Add(10*time.Second)); kind != Miss {
+		t.Errorf("expired entry came back after a reload (kind = %v)", kind)
+	}
+}
+
+func TestStoreTruncateAndStats(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "cache.db")
+	st, err := OpenStore(StoreOptions{Path: path})
+	if err != nil {
+		t.Fatalf("OpenStore: %v", err)
+	}
+	defer st.Close()
+	clk := newClock()
+	now := clk.Now()
+	l := NewLRU(100, 1)
+	wire, _ := answer(t, "a.example.com", 600, "10.0.0.10").Pack()
+	l.Put(key("a.example.com"), &Entry{Wire: wire, TTL: 600, CreatedAt: now, ExpiresAt: now.Add(600 * time.Second), StaleUntil: now.Add(630 * time.Second), Bytes: len(wire)})
+	if _, err := st.Snapshot(l, now); err != nil {
+		t.Fatalf("Snapshot: %v", err)
+	}
+	s := st.Stats()
+	if !s.Enabled || s.Path != path || s.Entries != 1 || s.Writes == 0 || s.Bytes <= 0 {
+		t.Errorf("Stats = %+v, want enabled with 1 entry and at least one write", s)
+	}
+	if !s.LastSnapshot.IsZero() {
+		t.Logf("LastSnapshot = %v", s.LastSnapshot)
+	}
+	if err := st.Truncate(); err != nil {
+		t.Fatalf("Truncate: %v", err)
+	}
+	if s := st.Stats(); s.Entries != 0 {
+		t.Errorf("entries after Truncate = %d, want 0", s.Entries)
+	}
+	restored := NewLRU(100, 1)
+	loaded, _, err := st.Load(restored, now)
+	if err != nil {
+		t.Fatalf("Load after Truncate: %v", err)
+	}
+	if loaded != 0 {
+		t.Errorf("loaded = %d after Truncate, want 0", loaded)
+	}
+}
+
+// TestManagerFlushDropsSnapshot checks the interaction the blocklist API
+// depends on: flushing the cache must also clear the persisted snapshot, or a
+// restart would resurrect answers an operator just invalidated.
+func TestManagerFlushDropsSnapshot(t *testing.T) {
+	dir := t.TempDir()
+	st, err := OpenStore(StoreOptions{Path: filepath.Join(dir, "cache.db")})
+	if err != nil {
+		t.Fatalf("OpenStore: %v", err)
+	}
+	defer st.Close()
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, Now: clk.Now}, st)
+	k := key("flushed.example.com")
+	m.PutResponse(k, answer(t, "flushed.example.com", 600, "10.0.0.11"), "up")
+	if _, err := st.Snapshot(m.LRU(), clk.Now()); err != nil {
+		t.Fatalf("Snapshot: %v", err)
+	}
+	m.Flush()
+	restored := NewLRU(10, 1)
+	loaded, _, err := st.Load(restored, clk.Now())
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	if loaded != 0 {
+		t.Errorf("loaded = %d entries after Flush, want 0", loaded)
+	}
+	if res := m.Get(k); res.Kind != Miss {
+		t.Errorf("kind = %v after Flush, want Miss", res.Kind)
+	}
+}
+
+func TestLRUStatsHitRate(t *testing.T) {
+	l := NewLRU(100, 1)
+	now := newClock().Now()
+	l.Put(key("a.example.com"), &Entry{TTL: 60, CreatedAt: now, ExpiresAt: now.Add(time.Minute), StaleUntil: now.Add(90 * time.Second), Bytes: 4})
+	for range 3 {
+		l.Get(key("a.example.com"), now)
+	}
+	for range 1 {
+		l.Get(key("nothere.example.com"), now)
+	}
+	s := l.Stats()
+	if s.Hits != 3 || s.Misses != 1 {
+		t.Fatalf("hits/misses = %d/%d, want 3/1", s.Hits, s.Misses)
+	}
+	if s.HitRate < 0.74 || s.HitRate > 0.76 {
+		t.Errorf("HitRate = %v, want ~0.75", s.HitRate)
+	}
+}
+
+func BenchmarkLRUGet(b *testing.B) {
+	l := NewLRU(1<<16, 16)
+	now := time.Now()
+	names := make([]string, 4096)
+	for i := range names {
+		names[i] = "n" + strconv.Itoa(i) + ".example.com"
+		wire, _ := answer(&testing.T{}, names[i], 600, "10.0.0.12").Pack()
+		l.Put(key(names[i]), &Entry{Wire: wire, TTL: 600, CreatedAt: now, ExpiresAt: now.Add(600 * time.Second), StaleUntil: now.Add(630 * time.Second), Bytes: len(wire)})
+	}
+	b.ResetTimer()
+	b.RunParallel(func(pb *testing.PB) {
+		i := rand.Int()
+		for pb.Next() {
+			l.Get(key(names[i%len(names)]), now)
+			i += 7
+		}
+	})
+}
+
+func BenchmarkLRUPut(b *testing.B) {
+	l := NewLRU(1<<16, 16)
+	now := time.Now()
+	b.ResetTimer()
+	for i := 0; b.Loop(); i++ {
+		l.Put(key("n"+strconv.Itoa(i)+".example.com"), &Entry{TTL: 600, CreatedAt: now, ExpiresAt: now.Add(600 * time.Second), StaleUntil: now.Add(630 * time.Second), Bytes: 64})
+	}
+}
+
+func BenchmarkManagerGetUnpack(b *testing.B) {
+	clk := newClock()
+	m := NewManager(Options{MaxTTL: time.Hour, Now: clk.Now}, nil)
+	names := make([]string, 4096)
+	for i := range names {
+		names[i] = "n" + strconv.Itoa(i) + ".example.com"
+		m.PutResponse(key(names[i]), answer(&testing.T{}, names[i], 600, "10.0.0.13"), "up")
+	}
+	b.ResetTimer()
+	b.RunParallel(func(pb *testing.PB) {
+		i := rand.Int()
+		for pb.Next() {
+			m.Get(key(names[i%len(names)]))
+			i += 7
+		}
+	})
+}

internal/cache/lru.go

@@ -22,8 +22,10 @@ type LRU struct {
 	shards []*shard
 	mask   uint64
 
-	// perShard is the per-shard entry budget.
+	// perShard is the per-shard entry budget. It is atomic because a config
-	perShard int
+	// reload resizes the cache from the API goroutine while queries are
+	// inserting in every shard.
+	perShard atomic.Int64
 
 	// counters are the hot-path stats, kept as atomics so Get does not need
 	// the lock just to record a hit.
@@ -69,13 +71,10 @@ func NewLRU(size, shards int) *LRU {
 	l := &LRU{
 		shards:       make([]*shard, shards),
 		mask:         uint64(shards - 1),
-		perShard:     size / shards,
 		seed:         maphash.MakeSeed(),
 		prefetchGate: make(chan struct{}, 64),
 	}
-	if l.perShard < 1 {
+	l.perShard.Store(int64(max(1, size/shards)))
-		l.perShard = 1
-	}
 	for i := range l.shards {
 		l.shards[i] = &shard{
 			items: make(map[Key]*list.Element, 1024),
@@ -140,7 +139,7 @@ func (l *LRU) Get(k Key, now time.Time) (e *Entry, kind HitKind) {
 	}
 	s.mu.Unlock()
 
-	e.Hits++
+	e.Hits.Add(1)
 	if kind == Fresh {
 		l.hits.Add(1)
 	} else {
@@ -171,7 +170,8 @@ func (l *LRU) Put(k Key, e *Entry) {
 	s.bytes += e.Bytes
 	l.bytes.Add(int64(e.Bytes))
 	var dropped int
-	for s.ll.Len() > l.perShard {
+	budget := int(l.perShard.Load())
+	for s.ll.Len() > budget {
 		oldest := s.ll.Back()
 		if oldest == nil {
 			break
@@ -347,7 +347,7 @@ func (l *LRU) Stats() LRUStats {
 		Expired:   l.expired.Load(),
 		HitRate:   rate,
 		Shards:    len(l.shards),
-		Capacity:  l.perShard * len(l.shards),
+		Capacity:  int(l.perShard.Load()) * len(l.shards),
 	}
 }
 
@@ -358,5 +358,31 @@ func (l *LRU) SetCapacity(size, shards int) {
 		shards = len(l.shards)
 	}
 	shards = roundPow2(shards)
-	l.perShard = max(1, size/shards)
+	budget := int64(max(1, size/shards))
+	l.perShard.Store(budget)
+	// A shrink has to reclaim memory now, not on the next insert: after a
+	// config reload lowers the cache size, the old entries would otherwise
+	// sit there until they happened to be overwritten.
+	var evicted int
+	for _, s := range l.shards[:min(shards, len(l.shards))] {
+		s.mu.Lock()
+		for s.ll.Len() > int(budget) {
+			oldest := s.ll.Back()
+			if oldest == nil {
+				break
+			}
+			it, ok := oldest.Value.(*lruItem)
+			if !ok {
+				break
+			}
+			s.ll.Remove(oldest)
+			delete(s.items, it.key)
+			s.bytes -= it.entry.Bytes
+			l.bytes.Add(-int64(it.entry.Bytes))
+			l.entries.Add(-1)
+			evicted++
+		}
+		s.mu.Unlock()
+	}
+	l.evictions.Add(uint64(evicted))
 }

internal/qlog/qlog.go

@@ -90,8 +90,11 @@ type Logger struct {
 	// seq assigns monotonic ids so the UI can detect drops.
 	seq atomic.Uint64
 
-	// live is a fixed-size ring of the most recent entries.
+	// live is a fixed-size ring of the most recent entries. Each slot is a
-	live     []liveEntry
+	// pointer, so a writer on the query path never races a reader walking the
+	// ring: publishing is one atomic word, and a reader sees either the whole
+	// entry or the one it replaced.
+	live     []atomic.Pointer[liveEntry]
 	livePos  atomic.Uint32
 	liveMu   sync.RWMutex
 	liveCond *sync.Cond
@@ -184,7 +187,7 @@ func New(opt Options) (*Logger, error) {
 	}
 	l := &Logger{
 		opts:    opt,
-		live:    make([]liveEntry, opt.LiveBuffer),
+		live:    make([]atomic.Pointer[liveEntry], opt.LiveBuffer),
 		counts:  newDomainCounts(opt.TopDomains, opt.TopTTL),
 		subs:    make(map[int]chan Entry),
 		stop:    make(chan struct{}),
@@ -258,9 +261,11 @@ func (l *Logger) Log(e Entry) {
 	e.ID = l.seq.Add(1)
 	l.counts.observe(e)
 
-	// Live ring.
+	// Live ring. The slot is published with a single atomic store; a reader
+	// walking the ring concurrently gets either the old or the new entry,
+	// never a half-written one.
 	pos := l.livePos.Add(1) - 1
-	l.live[pos%uint32(len(l.live))] = liveEntry{entry: e}
+	l.live[pos%uint32(len(l.live))].Store(&liveEntry{entry: e})
 
 	// Fan out to SSE subscribers (non-blocking: a slow consumer drops entries
 	// rather than stalling the query path).
@@ -435,7 +440,12 @@ func (l *Logger) Recent(limit int) []Entry {
 	}
 	for i := uint32(0); i < total; i++ {
 		idx := (pos - 1 - i) % uint32(len(l.live))
-		out = append(out, l.live[idx].entry)
+		le := l.live[idx].Load()
+		if le == nil {
+			// Never written (or already aged out of the window).
+			continue
+		}
+		out = append(out, le.entry)
 		if len(out) >= limit {
 			break
 		}

internal/resolver/pool.go

@@ -26,6 +26,10 @@ var ErrNoUpstream = errors.New("resolver: no healthy upstream available")
 // ErrCircuitOpen is returned by a single upstream whose breaker is open.
 var ErrCircuitOpen = errors.New("resolver: upstream circuit breaker open")
 
+// lookupTimeout bounds a name resolution when building the pool, so a broken
+// system resolver cannot stall a config reload.
+const lookupTimeout = 2 * time.Second
+
 // Health is an upstream's current state.
 type Health string
 
@@ -40,28 +44,46 @@ const (
 	HealthUnknown Health = "unknown"
 )
 
-// Upstream is one configured server plus its live health.
+// upstreamStats is everything mutable about an upstream. It lives behind a
+// pointer shared by successive Upstream values for the same server, so a
+// reload keeps the live health state while the immutable Upstream itself is
+// replaced under the readers' feet.
+type upstreamStats struct {
+	queries     atomic.Uint64
+	failures    atomic.Uint64
+	consecutive atomic.Uint32
+	latencySum  atomic.Uint64
+	latencyN    atomic.Uint64
+
+	// ring keeps the last 512 latencies in ms for the UI's p95.
+	mu   sync.Mutex
+	ring []float64
+
+	breakerOpenUntil atomic.Int64
+	health           atomic.Pointer[Health]
+	lastErr          atomic.Pointer[string]
+}
+
+func newUpstreamStats() *upstreamStats {
+	s := &upstreamStats{}
+	s.health.Store(ptr(HealthUnknown))
+	return s
+}
+
+// Upstream is one configured server: its address, its dial-time
+// configuration and a pointer to the live stats it shares across reloads.
+//
+// An Upstream is immutable after construction. Rebind builds a fresh Upstream
+// for every server and swaps the whole pool state with one atomic store, so a
+// query in flight keeps using a consistent, unchanging view.
 type Upstream struct {
 	cfg config.UpstreamServer
 	// addr is the address actually dialled: a bootstrap IP for TLS (SNI keeps
 	// the hostname) and the literal address otherwise.
 	addr string
-	// resolved holds the current bootstrap addresses when the configured
+	// resolved holds the bootstrap addresses behind a named upstream.
-	// address is a name.
 	resolved []string
-
+	stats    *upstreamStats
-	queries     atomic.Uint64
-	failures    atomic.Uint64
-	consecutive atomic.Uint32
-	latencySum  atomic.Uint64 // microseconds
-	latencyN    atomic.Uint64
-	p95         []float64 // recent latencies in ms, for the UI
-	p95mu       sync.Mutex
-	p95ring     []float64
-
-	breakerOpenUntil atomic.Int64
-	health           atomic.Pointer[Health]
-	lastErr          atomic.Pointer[string]
 }
 
 // Addr is the address the resolver dials.
@@ -70,29 +92,35 @@ func (u *Upstream) Addr() string { return u.addr }
 // Config returns the upstream's configuration entry.
 func (u *Upstream) Config() config.UpstreamServer { return u.cfg }
 
+// Resolved returns the addresses a named upstream resolved to, if any.
+func (u *Upstream) Resolved() []string { return u.resolved }
+
 // Health returns the current health state.
 func (u *Upstream) Health() Health {
-	if h := u.health.Load(); h != nil {
+	if u == nil || u.stats == nil {
+		return HealthUnknown
+	}
+	if h := u.stats.health.Load(); h != nil {
 		return *h
 	}
 	return HealthUnknown
 }
 
-func (u *Upstream) setHealth(h Health) { u.health.Store(&h) }
+func (u *Upstream) setHealth(h Health) { u.stats.health.Store(&h) }
 
 // SetError records the last error message for the UI.
 func (u *Upstream) SetError(err error) {
 	if err == nil {
-		u.lastErr.Store(nil)
+		u.stats.lastErr.Store(nil)
 		return
 	}
 	s := err.Error()
-	u.lastErr.Store(&s)
+	u.stats.lastErr.Store(&s)
 }
 
 // LastError returns the last error message, or "".
 func (u *Upstream) LastError() string {
-	if p := u.lastErr.Load(); p != nil {
+	if p := u.stats.lastErr.Load(); p != nil {
 		return *p
 	}
 	return ""
@@ -100,10 +128,11 @@ func (u *Upstream) LastError() string {
 
 // Success records a successful query.
 func (u *Upstream) Success(d time.Duration) {
-	u.queries.Add(1)
+	s := u.stats
-	u.consecutive.Store(0)
+	s.queries.Add(1)
-	u.latencySum.Add(uint64(d.Microseconds()))
+	s.consecutive.Store(0)
-	u.latencyN.Add(1)
+	s.latencySum.Add(uint64(d.Microseconds()))
+	s.latencyN.Add(1)
 	u.recordLatency(d)
 	u.SetError(nil)
 	if u.Health() != HealthUp {
@@ -114,12 +143,13 @@ func (u *Upstream) Success(d time.Duration) {
 // Failure records a failed query and trips the breaker once the threshold is
 // crossed.
 func (u *Upstream) Failure(err error, threshold int, cooldown time.Duration) {
-	u.queries.Add(1)
+	s := u.stats
-	u.failures.Add(1)
+	s.queries.Add(1)
-	n := u.consecutive.Add(1)
+	s.failures.Add(1)
+	n := s.consecutive.Add(1)
 	u.SetError(err)
 	if threshold > 0 && int(n) >= threshold {
-		u.breakerOpenUntil.Store(time.Now().Add(cooldown).UnixNano())
+		s.breakerOpenUntil.Store(time.Now().Add(cooldown).UnixNano())
 		u.setHealth(HealthDown)
 		return
 	}
@@ -130,32 +160,33 @@ func (u *Upstream) Failure(err error, threshold int, cooldown time.Duration) {
 
 // BreakerOpen reports whether the upstream's breaker is open.
 func (u *Upstream) BreakerOpen() bool {
-	until := u.breakerOpenUntil.Load()
+	return u.stats.breakerOpenUntil.Load() > time.Now().UnixNano()
-	return until > time.Now().UnixNano()
 }
 
 func (u *Upstream) recordLatency(d time.Duration) {
 	ms := float64(d.Microseconds()) / 1000
-	u.p95mu.Lock()
+	s := u.stats
-	if len(u.p95ring) < 512 {
+	s.mu.Lock()
-		u.p95ring = append(u.p95ring, ms)
+	if len(s.ring) < 512 {
+		s.ring = append(s.ring, ms)
 	} else {
-		copy(u.p95ring, u.p95ring[1:])
+		copy(s.ring, s.ring[1:])
-		u.p95ring[511] = ms
+		s.ring[511] = ms
 	}
-	u.p95mu.Unlock()
+	s.mu.Unlock()
 }
 
 // LatencyStats returns average and p95 latency in milliseconds.
 func (u *Upstream) LatencyStats() (avg, p95 float64) {
-	n := u.latencyN.Load()
+	s := u.stats
+	n := s.latencyN.Load()
 	if n == 0 {
 		return 0, 0
 	}
-	avg = float64(u.latencySum.Load()) / float64(n) / 1000
+	avg = float64(s.latencySum.Load()) / float64(n) / 1000
-	u.p95mu.Lock()
+	s.mu.Lock()
-	samples := append([]float64(nil), u.p95ring...)
+	samples := append([]float64(nil), s.ring...)
-	u.p95mu.Unlock()
+	s.mu.Unlock()
 	if len(samples) == 0 {
 		return avg, 0
 	}
@@ -187,8 +218,11 @@ type Stats struct {
 
 // Stats snapshots one upstream.
 func (u *Upstream) Stats() Stats {
-	q := u.queries.Load()
+	if u == nil {
-	f := u.failures.Load()
+		return Stats{}
+	}
+	q := u.stats.queries.Load()
+	f := u.stats.failures.Load()
 	var rate float64
 	if q > 0 {
 		rate = float64(f) / float64(q)
@@ -212,83 +246,127 @@ func (u *Upstream) Stats() Stats {
 	}
 }
 
-// Pool is the set of upstreams plus the policy that selects among them.
+// poolState is an immutable snapshot of the pool: the config it was built
-type Pool struct {
+// from, the upstreams, and the address index used for per-client upstream
-	cfg   config.Upstream
+// overrides. Readers take the pointer once and are then immune to reloads.
-	mu    sync.RWMutex
+type poolState struct {
-	ups   []*Upstream
+	cfg config.Upstream
-	byTag map[string][]*Upstream
+	ups []*Upstream
-	// clients are reused per upstream: one UDP socket, one TCP dialer, one TLS
+	// byAddr indexes the upstreams by every address they answer to, so an ACL
-	// session pool. Reusing them is what keeps the per-query cost low.
+	// group naming either the configured Addr or the resolved one works.
-	clients map[*Upstream]*dns.Client
+	byAddr map[string][]*Upstream
-	// bootstrap resolves names to addresses.
-	bootstrap map[string][]string
-
-	rng *rand.Rand
-	mu2 sync.Mutex
 }
 
-// NewPool builds a pool from config. bootstrapIPs may be empty; then upstream
+func (st *poolState) all() []*Upstream {
-// hostnames must be literal IPs (or resolved by the OS).
+	if st == nil {
+		return nil
+	}
+	return st.ups
+}
+
+// Pool is the set of upstreams plus the policy that selects among them.
+type Pool struct {
+	state atomic.Pointer[poolState]
+	// clients are memoised per upstream: one UDP socket, one TCP dialer, one
+	// TLS session pool. Reusing them is what keeps the per-query cost low; a
+	// dns.Client is safe for concurrent use.
+	clients sync.Map // *Upstream -> *dns.Client
+
+	// mu serialises Rebind so two overlapping reloads cannot interleave their
+	// health hand-over. It is never taken on the query path.
+	mu sync.Mutex
+	// rng backs the "random" policy.
+	rngMu sync.Mutex
+	rng   *rand.Rand
+}
+
+// NewPool builds a pool from config. Bootstraps are used for named upstreams;
+// anything unresolved falls back to the system resolver.
 func NewPool(cfg config.Upstream) *Pool {
 	p := &Pool{
-		cfg:       cfg,
+		rng: rand.New(rand.NewPCG(uint64(time.Now().UnixNano()), 0x9e3779b9)),
-		byTag:     make(map[string][]*Upstream),
-		clients:   make(map[*Upstream]*dns.Client),
-		bootstrap: make(map[string][]string),
-		rng:       rand.New(rand.NewPCG(uint64(time.Now().UnixNano()), 0x9e3779b9)),
 	}
-	for _, b := range cfg.Bootstraps {
+	p.state.Store(buildState(cfg, nil, false))
-		p.bootstrap[b.Name] = append([]string(nil), b.Addresses...)
-	}
-	for _, s := range cfg.Servers {
-		up := &Upstream{cfg: s}
-		up.setHealth(HealthUnknown)
-		p.ups = append(p.ups, up)
-	}
-	p.resolveAll()
 	return p
 }
 
-func (p *Pool) resolveAll() {
+// buildState turns an upstream config into an immutable pool snapshot. When
-	for _, up := range p.ups {
+// carry is non-nil, servers that keep the same address and protocol also keep
-		host, port, err := net.SplitHostPort(up.cfg.Addr)
+// their live health, latency and breaker state.
-		if err != nil {
+func buildState(cfg config.Upstream, carry *poolState, lookup bool) *poolState {
-			continue
+	bootstraps := make(map[string][]string, len(cfg.Bootstraps))
-		}
+	for _, b := range cfg.Bootstraps {
-		if net.ParseIP(host) != nil {
+		bootstraps[b.Name] = append([]string(nil), b.Addresses...)
-			up.addr = up.cfg.Addr
-			continue
-		}
-		if addrs := p.bootstrap[host]; len(addrs) > 0 {
-			up.resolved = addrs
-			up.addr = net.JoinHostPort(addrs[0], port)
-			continue
-		}
-		// Fall back to the system resolver.
-		ips, err := net.LookupIP(host)
-		if err != nil || len(ips) == 0 {
-			up.addr = up.cfg.Addr
-			continue
-		}
-		strs := make([]string, 0, len(ips))
-		for _, ip := range ips {
-			strs = append(strs, ip.String())
-		}
-		up.resolved = strs
-		up.addr = net.JoinHostPort(strs[0], port)
 	}
+	prev := make(map[string]*Upstream)
+	if carry != nil {
+		for _, u := range carry.ups {
+			prev[u.cfg.Addr+"/"+u.cfg.Proto] = u
+		}
+	}
+
+	st := &poolState{cfg: cfg, ups: make([]*Upstream, 0, len(cfg.Servers))}
+	for _, s := range cfg.Servers {
+		addr, resolved := resolveAddr(s, bootstraps, lookup)
+		up := &Upstream{cfg: s, addr: addr, resolved: resolved}
+		if old, ok := prev[s.Addr+"/"+s.Proto]; ok {
+			up.stats = old.stats
+		} else {
+			up.stats = newUpstreamStats()
+		}
+		st.ups = append(st.ups, up)
+	}
+	st.byAddr = make(map[string][]*Upstream, len(st.ups)*2)
+	for _, up := range st.ups {
+		for _, a := range append([]string{up.cfg.Addr, up.addr}, up.resolved...) {
+			if a == "" {
+				continue
+			}
+			st.byAddr[a] = append(st.byAddr[a], up)
+		}
+	}
+	return st
+}
+
+// resolveAddr returns the address to dial for a server plus the bootstrap
+// addresses behind it, if the configured host is a name.
+func resolveAddr(s config.UpstreamServer, bootstraps map[string][]string, lookup bool) (addr string, resolved []string) {
+	host, port, err := net.SplitHostPort(s.Addr)
+	if err != nil {
+		// Not host:port — hand it to the dialer verbatim.
+		return s.Addr, nil
+	}
+	if net.ParseIP(host) != nil {
+		return s.Addr, nil
+	}
+	if addrs := bootstraps[host]; len(addrs) > 0 {
+		return net.JoinHostPort(addrs[0], port), addrs
+	}
+	if lookup {
+		ctx, cancel := context.WithTimeout(context.Background(), lookupTimeout)
+		defer cancel()
+		addrs, err := net.DefaultResolver.LookupHost(ctx, host)
+		if err == nil && len(addrs) > 0 {
+			return net.JoinHostPort(addrs[0], port), addrs
+		}
+	}
+	return s.Addr, nil
 }
 
 // Config returns the pool's upstream config.
-func (p *Pool) Config() config.Upstream { return p.cfg }
+func (p *Pool) Config() config.Upstream {
+	if st := p.state.Load(); st != nil {
+		return st.cfg
+	}
+	return config.Upstream{}
+}
 
 // All returns every upstream.
-func (p *Pool) All() []*Upstream { return p.ups }
+func (p *Pool) All() []*Upstream { return p.state.Load().all() }
 
-// Get returns an upstream by address, or nil.
+// Get returns an upstream by configured or resolved address, or nil.
 func (p *Pool) Get(addr string) *Upstream {
-	for _, u := range p.ups {
+	for _, u := range p.All() {
 		if u.addr == addr || u.cfg.Addr == addr {
 			return u
 		}
@@ -298,8 +376,9 @@ func (p *Pool) Get(addr string) *Upstream {
 
 // Healthy returns upstreams whose breaker is closed, in configured order.
 func (p *Pool) Healthy() []*Upstream {
-	out := make([]*Upstream, 0, len(p.ups))
+	ups := p.All()
-	for _, u := range p.ups {
+	out := make([]*Upstream, 0, len(ups))
+	for _, u := range ups {
 		if !u.BreakerOpen() {
 			out = append(out, u)
 		}
@@ -309,8 +388,9 @@ func (p *Pool) Healthy() []*Upstream {
 
 // Stats snapshots every upstream.
 func (p *Pool) Stats() []Stats {
-	out := make([]Stats, 0, len(p.ups))
+	ups := p.All()
-	for _, u := range p.ups {
+	out := make([]Stats, 0, len(ups))
+	for _, u := range ups {
 		out = append(out, u.Stats())
 	}
 	return out
@@ -318,95 +398,54 @@ func (p *Pool) Stats() []Stats {
 
 // clientFor returns the memoised dns.Client for an upstream.
 func (p *Pool) clientFor(u *Upstream) *dns.Client {
-	p.mu.RLock()
+	if c, ok := p.clients.Load(u); ok {
-	c, ok := p.clients[u]
+		return c.(*dns.Client)
-	p.mu.RUnlock()
-	if ok {
-		return c
 	}
-	var nc *dns.Client
+	nc := &dns.Client{Net: "udp", UDPSize: 1232, Timeout: 3 * time.Second}
 	switch u.cfg.Proto {
 	case "tls":
-		tlsCfg := &tls.Config{
-			ServerName:         u.cfg.ServerName,
-			InsecureSkipVerify: u.cfg.InsecureSkipVerify,
-			MinVersion:         tls.VersionTLS12,
-			NextProtos:         []string{"dot"},
-		}
 		nc = &dns.Client{
-			Net:       "tcp-tls",
+			Net: "tcp-tls",
-			TLSConfig: tlsCfg,
+			TLSConfig: &tls.Config{
-			Dialer:    &net.Dialer{Timeout: 3 * time.Second},
+				ServerName:         u.cfg.ServerName,
-			Timeout:   5 * time.Second,
+				InsecureSkipVerify: u.cfg.InsecureSkipVerify,
+				MinVersion:         tls.VersionTLS12,
+				NextProtos:         []string{"dot"},
+			},
+			Dialer:  &net.Dialer{Timeout: 3 * time.Second},
+			Timeout: 5 * time.Second,
 		}
 	case "tcp":
 		nc = &dns.Client{Net: "tcp", Timeout: 5 * time.Second}
-	default:
-		nc = &dns.Client{Net: "udp", UDPSize: 1232, Timeout: 3 * time.Second}
-	}
-	p.mu.Lock()
-	p.clients[u] = nc
-	p.mu.Unlock()
-	return nc
-}
-
-// selectFor returns the upstreams to try, honouring the policy. tags, if
-// non-empty, restricts the pool to a group's own upstreams.
-func (p *Pool) selectFor(policy string, maxAttempts int) []*Upstream {
-	healthy := p.Healthy()
-	if len(healthy) == 0 {
-		// Every breaker is open: try them all anyway rather than failing the
-		// query outright. A blip should not become an outage.
-		healthy = append(healthy, p.ups...)
-	}
-	if len(healthy) == 0 {
-		return nil
-	}
-	if maxAttempts <= 0 || maxAttempts > len(healthy) {
-		maxAttempts = len(healthy)
-	}
-	switch policy {
-	case "failover":
-		// Configured order, and rotate past the unhealthy ones.
-		return healthy[:maxAttempts]
-	case "random":
-		p.mu2.Lock()
-		p.rng.Shuffle(len(healthy), func(i, j int) { healthy[i], healthy[j] = healthy[j], healthy[i] })
-		p.mu2.Unlock()
-		return healthy[:maxAttempts]
-	default: // parallel
-		return healthy
 	}
+	actual, _ := p.clients.LoadOrStore(u, nc)
+	return actual.(*dns.Client)
 }
 
 // Exchange forwards q to the pool and returns the first acceptable answer.
 //
-// policy is the pool's configured policy; tags restricts the pool to a
+// want restricts the query to the upstreams an ACL group named, by configured
-// per-client upstream override. edns0Size is the client's advertised buffer,
+// or resolved address; an empty list means the whole pool. timeout bounds one
-// used to decide TCP fallback on truncation.
+// attempt.
-func (p *Pool) Exchange(ctx context.Context, q *dns.Msg, tags []string, timeout time.Duration) (*dns.Msg, string, error) {
+func (p *Pool) Exchange(ctx context.Context, q *dns.Msg, want []string, timeout time.Duration) (*dns.Msg, string, error) {
-	candidates := p.ups
+	st := p.state.Load()
-	if len(tags) > 0 {
+	candidates := st.all()
-		if tagged := p.byTag[strings.Join(tags, ",")]; len(tagged) > 0 {
+	if len(want) > 0 {
-			candidates = tagged
+		candidates = p.restrict(st, want)
-		}
 	}
 	if len(candidates) == 0 {
 		return nil, "", ErrNoUpstream
 	}
-	attempts := p.cfg.MaxAttempts
+	chosen := p.pick(st, candidates, st.cfg.MaxAttempts)
-	chosen := selectFrom(candidates, p.cfg.Policy, attempts)
 	if len(chosen) == 0 {
 		return nil, "", ErrNoUpstream
 	}
-
 	if len(chosen) == 1 {
-		msg, up, err := p.exchangeOne(ctx, chosen[0], q, timeout)
+		return p.exchangeOne(ctx, st.cfg, chosen[0], q, timeout)
-		return msg, up, err
 	}
 
-	// Parallel: send to all at once, take the first answer that is not a
+	// Parallel: send to all at once and take the first real answer. The
-	// transport failure. Cancellable so we do not leak requests to a slow
+	// context is cancelled on return so we do not leak a request to a slow
 	// upstream after another one has answered.
 	type res struct {
 		msg *dns.Msg
@@ -418,7 +457,7 @@ func (p *Pool) Exchange(ctx context.Context, q *dns.Msg, tags []string, timeout
 	defer cancel()
 	for _, up := range chosen {
 		go func(up *Upstream) {
-			msg, addr, err := p.exchangeOne(sub, up, q, timeout)
+			msg, addr, err := p.exchangeOne(sub, st.cfg, up, q, timeout)
 			results <- res{msg: msg, up: addr, err: err}
 		}(up)
 	}
@@ -426,13 +465,13 @@ func (p *Pool) Exchange(ctx context.Context, q *dns.Msg, tags []string, timeout
 	for i := 0; i < len(chosen); i++ {
 		select {
 		case r := <-results:
-			if r.err == nil && r.msg != nil && !isTransportFailure(r.msg.Rcode) {
+			switch {
-				return r.msg, r.up, nil
+			case r.err != nil:
-			}
-			if r.err != nil {
 				lastErr = r.err
-			} else if r.msg != nil {
+			case r.msg != nil && r.msg.Rcode != dns.RcodeServerFailure:
-				lastErr = fmt.Errorf("upstream %s returned %s", r.up, dns.RcodeToString[r.msg.Rcode])
+				return r.msg, r.up, nil
+			default:
+				lastErr = fmt.Errorf("upstream %s returned SERVFAIL", r.up)
 			}
 		case <-ctx.Done():
 			return nil, "", ctx.Err()
@@ -444,8 +483,29 @@ func (p *Pool) Exchange(ctx context.Context, q *dns.Msg, tags []string, timeout
 	return nil, "", lastErr
 }
 
-func selectFrom(ups []*Upstream, policy string, maxAttempts int) []*Upstream {
+// restrict resolves an ACL group's upstream list to upstreams, keeping the
-	// Filter out open breakers but keep everything if that empties the list.
+// configured order and dropping names that match nothing. When nothing in the
+// list resolves the full pool is used, so a typo cannot black-hole a client.
+func (p *Pool) restrict(st *poolState, want []string) []*Upstream {
+	seen := make(map[*Upstream]bool, len(want))
+	out := make([]*Upstream, 0, len(want))
+	for _, name := range want {
+		for _, up := range st.byAddr[name] {
+			if !seen[up] {
+				seen[up] = true
+				out = append(out, up)
+			}
+		}
+	}
+	if len(out) == 0 {
+		return st.all()
+	}
+	return out
+}
+
+// pick filters out open breakers (keeping everything if that empties the
+// list) and applies the pool policy.
+func (p *Pool) pick(st *poolState, ups []*Upstream, maxAttempts int) []*Upstream {
 	open := make([]*Upstream, 0, len(ups))
 	for _, u := range ups {
 		if !u.BreakerOpen() {
@@ -458,11 +518,13 @@ func selectFrom(ups []*Upstream, policy string, maxAttempts int) []*Upstream {
 	if maxAttempts <= 0 || maxAttempts > len(open) {
 		maxAttempts = len(open)
 	}
-	switch policy {
+	switch st.cfg.Policy {
 	case "failover":
 		return open[:maxAttempts]
 	case "random":
-		rand.Shuffle(len(open), func(i, j int) { open[i], open[j] = open[j], open[i] })
+		p.rngMu.Lock()
+		defer p.rngMu.Unlock()
+		p.rng.Shuffle(len(open), func(i, j int) { open[i], open[j] = open[j], open[i] })
 		return open[:maxAttempts]
 	default:
 		return open
@@ -471,63 +533,54 @@ func selectFrom(ups []*Upstream, policy string, maxAttempts int) []*Upstream {
 
 // exchangeOne sends q to a single upstream, retrying over TCP when the UDP
 // answer comes back truncated (RFC 1035 §4.2.1).
-func (p *Pool) exchangeOne(ctx context.Context, up *Upstream, q *dns.Msg, timeout time.Duration) (*dns.Msg, string, error) {
+func (p *Pool) exchangeOne(ctx context.Context, cfg config.Upstream, up *Upstream, q *dns.Msg, timeout time.Duration) (*dns.Msg, string, error) {
 	if timeout <= 0 {
 		timeout = 2 * time.Second
 	}
 	client := p.clientFor(up)
-	msg := q.Copy()
+	resp, rtt, err := client.ExchangeContext(ctx, q.Copy(), up.addr)
-	start := time.Now()
-	resp, rtt, err := client.ExchangeContext(ctx, msg, up.addr)
-	up.Success(rtt)
 	if err != nil {
+		up.Failure(err, thresholdFor(up, cfg.BreakerThreshold), cfg.BreakerCooldown.D())
 		return nil, up.addr, err
 	}
-	_ = start
+	up.Success(rtt)
 	if !resp.Truncated {
 		return resp, up.addr, nil
 	}
-	// Retry over TCP to get the full answer. Some upstreams are configured
+	// Retry over TCP for the full answer. An upstream that is UDP-only just
-	// UDP-only in which case this simply fails and we use the TC answer.
+	// fails here and we return the truncated answer; the client retries over
+	// TCP itself.
 	tcpClient := &dns.Client{Net: "tcp", Timeout: timeout}
 	resp2, rtt2, err2 := tcpClient.ExchangeContext(ctx, q.Copy(), up.addr)
-	if err2 == nil && resp2 != nil && !resp2.Truncated {
+	if err2 != nil || resp2 == nil || resp2.Truncated {
-		up.Success(rtt2)
+		if err2 == nil {
-		return resp2, up.addr, nil
+			err2 = errors.New("still truncated over TCP")
+		}
+		up.Failure(fmt.Errorf("truncated answer, TCP retry failed: %w", err2),
+			thresholdFor(up, cfg.BreakerThreshold), cfg.BreakerCooldown.D())
+		return resp, up.addr, nil
 	}
-	up.Failure(fmt.Errorf("truncated answer, TCP retry failed: %w", err2), up.breakerThreshold(), p.cfg.BreakerCooldown.D())
+	up.Success(rtt2)
-	// Return the truncated answer; the client will retry over TCP itself.
+	return resp2, up.addr, nil
-	return resp, up.addr, nil
-}
-
-func isTransportFailure(rcode int) bool {
-	return rcode == dns.RcodeServerFailure
-}
-
-// breakerThreshold is the effective failure threshold for an upstream.
-func (u *Upstream) breakerThreshold() int {
-	if u.cfg.MaxFails > 0 {
-		return u.cfg.MaxFails
-	}
-	return 0
 }
 
 // HealthCheck probes every upstream with a cheap query, refreshing health
 // state even when no client traffic is flowing. It returns the number of
 // upstreams that answered.
 func (p *Pool) HealthCheck(ctx context.Context, timeout time.Duration) int {
+	st := p.state.Load()
 	probe := new(dns.Msg)
 	probe.SetQuestion(".", dns.TypeNS)
 	probe.RecursionDesired = true
 	ok := 0
-	for _, up := range p.ups {
+	for _, up := range st.all() {
 		if up.BreakerOpen() {
 			continue
 		}
 		c := p.clientFor(up)
 		_, rtt, err := c.ExchangeContext(ctx, probe, up.addr)
 		if err != nil {
-			up.Failure(err, p.breakerThresholdFor(up), p.cfg.BreakerCooldown.D())
+			up.Failure(err, thresholdFor(up, st.cfg.BreakerThreshold), st.cfg.BreakerCooldown.D())
 			continue
 		}
 		up.Success(rtt)
@@ -536,63 +589,47 @@ func (p *Pool) HealthCheck(ctx context.Context, timeout time.Duration) int {
 	return ok
 }
 
-func (p *Pool) breakerThresholdFor(u *Upstream) int {
+func thresholdFor(up *Upstream, def int) int {
-	if t := u.breakerThreshold(); t > 0 {
+	if up.cfg.MaxFails > 0 {
-		return t
+		return up.cfg.MaxFails
 	}
-	return p.cfg.BreakerThreshold
+	return def
 }
 
-// Rebind applies a new upstream configuration in place, keeping the live
+// Rebind applies a new upstream configuration, keeping the live health state
-// health state of upstreams whose address is unchanged.
+// of upstreams whose address and protocol are unchanged.
+//
+// It builds a whole new immutable state and swaps it in with a single atomic
+// store: queries already in flight finish against the state they started
+// with, and the next query sees the new one. There is no window in which half
+// the upstreams are old and half new.
 func (p *Pool) Rebind(cfg config.Upstream) {
 	p.mu.Lock()
 	defer p.mu.Unlock()
-	p.cfg = cfg
+	old := p.state.Load()
-	old := make(map[string]*Upstream, len(p.ups))
+	next := buildState(cfg, old, true)
-	for _, u := range p.ups {
+	p.state.Store(next)
-		old[u.cfg.Addr+"/"+u.cfg.Proto] = u
+	// Drop the memoised clients of upstreams that are gone; their shells are
+	// only reachable from in-flight queries, which let them go on their own.
+	live := make(map[*Upstream]bool, len(next.all()))
+	for _, u := range next.all() {
+		live[u] = true
 	}
-	next := make([]*Upstream, 0, len(cfg.Servers))
+	for _, u := range old.all() {
-	for _, s := range cfg.Servers {
+		if !live[u] {
-		if u, ok := old[s.Addr+"/"+s.Proto]; ok {
+			p.clients.Delete(u)
-			u.cfg = s
-			next = append(next, u)
-			delete(old, s.Addr+"/"+s.Proto)
-			continue
-		}
-		u := &Upstream{cfg: s}
-		u.setHealth(HealthUnknown)
-		next = append(next, u)
-	}
-	p.ups = next
-	// Drop clients for removed upstreams.
-	for u := range p.clients {
-		if old[u.cfg.Addr+"/"+u.cfg.Proto] != nil {
-			delete(p.clients, u)
 		}
 	}
-	p.resolveAllLocked()
 }
 
-func (p *Pool) resolveAllLocked() {
+// String renders the pool for logs.
-	for _, b := range p.cfg.Bootstraps {
+func (p *Pool) String() string {
-		p.bootstrap[b.Name] = append([]string(nil), b.Addresses...)
+	ups := p.All()
-	}
+	addrs := make([]string, 0, len(ups))
-	for _, up := range p.ups {
+	for _, u := range ups {
-		host, port, err := net.SplitHostPort(up.cfg.Addr)
+		addrs = append(addrs, u.addr)
-		if err != nil {
-			continue
-		}
-		if net.ParseIP(host) != nil {
-			up.addr = up.cfg.Addr
-			continue
-		}
-		if addrs := p.bootstrap[host]; len(addrs) > 0 {
-			up.resolved = addrs
-			up.addr = net.JoinHostPort(addrs[0], port)
-			continue
-		}
-		up.addr = up.cfg.Addr
 	}
+	return "pool[" + strings.Join(addrs, " ") + "]"
 }
+
+func ptr[T any](v T) *T { return &v }

internal/web/dist/index.html

@@ -1,3 +1,14 @@
-<!doctype html><html><head><meta charset="utf-8"><title>fancy-dns</title></head>
+<!doctype html>
-<body style="font:16px system-ui;padding:2rem"><h1>fancy-dns</h1>
+<html lang="en">
-<p>Web console assets are not built in this binary. Run <code>make web</code> and rebuild.</p></body></html>
+  <head>
+    <meta charset="UTF-8" />
+    <link rel="icon" type="image/svg+xml" href="./favicon.svg" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>web</title>
+    <script type="module" crossorigin src="./assets/index-BVuSgjZf.js"></script>
+    <link rel="stylesheet" crossorigin href="./assets/index-CZ8p4VMZ.css">
+  </head>
+  <body>
+    <div id="root"></div>
+  </body>
+</html>

internal/zonefile/zonefile_test.go

@@ -0,0 +1,918 @@
+package zonefile
+
+import (
+	"bufio"
+	"fmt"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+	"testing"
+
+	"github.com/miekg/dns"
+)
+
+func mustParse(t *testing.T, src string, opt Options) *Zone {
+	t.Helper()
+	z, err := Parse(strings.NewReader(src), opt)
+	if err != nil {
+		t.Fatalf("Parse: unexpected error: %v\nsource:\n%s", err, src)
+	}
+	return z
+}
+
+func parseErr(t *testing.T, src string, opt Options) error {
+	t.Helper()
+	z, err := Parse(strings.NewReader(src), opt)
+	if err == nil {
+		t.Fatalf("Parse: expected an error, got %d records: %v", len(z.Records), z.Records)
+	}
+	return err
+}
+
+// summary renders every record as "line name ttl class type rdata" so a single
+// comparison covers folding, owner/TTL/class inheritance and origin resolution.
+func summary(t *testing.T, z *Zone) []string {
+	t.Helper()
+	out := make([]string, 0, len(z.Records))
+	for _, rec := range z.Records {
+		out = append(out, renderRecord(t, rec, true))
+	}
+	return out
+}
+
+// summaryNoTTL is summary without the TTL column, for fixtures whose records
+// carry no explicit TTL (the TTL then comes from the RR header).
+func summaryNoTTL(t *testing.T, z *Zone) []string {
+	t.Helper()
+	out := make([]string, 0, len(z.Records))
+	for _, rec := range z.Records {
+		out = append(out, renderRecord(t, rec, false))
+	}
+	return out
+}
+
+func renderRecord(t *testing.T, rec Record, withTTL bool) string {
+	t.Helper()
+	parts := strings.SplitN(rec.RR.String(), "\t", 5)
+	if len(parts) != 5 {
+		t.Fatalf("record line %d does not render as 5 tab-separated fields: %q", rec.Line, rec.Text())
+	}
+	if !withTTL {
+		return fmt.Sprintf("%d %s %s %s %s", rec.Line, parts[0], parts[2], parts[3], parts[4])
+	}
+	return fmt.Sprintf("%d %s %d %s %s %s", rec.Line, parts[0], rec.TTL, parts[2], parts[3], parts[4])
+}
+
+func checkSummary(t *testing.T, z *Zone, want []string) {
+	t.Helper()
+	if got := summary(t, z); !slices.Equal(got, want) {
+		t.Errorf("records mismatch\n got: %q\nwant: %q", got, want)
+	}
+}
+
+func recordByName(t *testing.T, z *Zone, name string) Record {
+	t.Helper()
+	for _, rec := range z.Records {
+		if rec.Name == name {
+			return rec
+		}
+	}
+	t.Fatalf("no record for %q in %v", name, summary(t, z))
+	return Record{}
+}
+
+func recordByLine(t *testing.T, z *Zone, line int) Record {
+	t.Helper()
+	for _, rec := range z.Records {
+		if rec.Line == line {
+			return rec
+		}
+	}
+	t.Fatalf("no record starting on line %d in %v", line, summary(t, z))
+	return Record{}
+}
+
+func TestParseTTL(t *testing.T) {
+	tests := []struct {
+		in      string
+		want    uint32
+		wantErr bool
+	}{
+		{in: "", wantErr: true},
+		{in: "   ", wantErr: true},
+		{in: "300", want: 300},
+		{in: "  60  ", want: 60},
+		{in: "0", want: 0},
+		{in: "300s", want: 300},
+		{in: "300sec", want: 300},
+		{in: "300seconds", want: 300},
+		{in: "5m", want: 300},
+		{in: "5min", want: 300},
+		{in: "5minutes", want: 300},
+		{in: "1h", want: 3600},
+		{in: "1hour", want: 3600},
+		{in: "1hours", want: 3600},
+		{in: "1d", want: 86400},
+		{in: "1days", want: 86400},
+		{in: "1w", want: 604800},
+		{in: "1weeks", want: 604800},
+		{in: "1h30m", want: 5400},
+		{in: "1w2d3h4m5s", want: 788645},
+		{in: "1h30", want: 3630},
+		{in: "4294967295", want: 4294967295},
+		{in: "4294967296", wantErr: true},
+		{in: "-1", wantErr: true},
+		{in: "-5m", wantErr: true},
+		{in: "abc", wantErr: true},
+		{in: "m", wantErr: true},
+		{in: "1x", wantErr: true},
+		{in: "1.5h", wantErr: true},
+		{in: "1h-", wantErr: true},
+		{in: "0x10", wantErr: true},
+	}
+	for _, tc := range tests {
+		got, err := ParseTTL(tc.in)
+		if tc.wantErr {
+			if err == nil {
+				t.Errorf("ParseTTL(%q) = %d, want an error", tc.in, got)
+			}
+			continue
+		}
+		if err != nil {
+			t.Errorf("ParseTTL(%q) error: %v", tc.in, err)
+			continue
+		}
+		if got != tc.want {
+			t.Errorf("ParseTTL(%q) = %d, want %d", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestFormatTTL(t *testing.T) {
+	tests := []struct {
+		in   uint32
+		want string
+	}{
+		{0, "0"},
+		{1, "1"},
+		{59, "59"},
+		{60, "1m"},
+		{300, "5m"},
+		{3600, "1h"},
+		{86400, "1d"},
+		{604800, "1w"},
+		{3601, "3601"},
+		{1209600, "2w"},
+	}
+	for _, tc := range tests {
+		if got := FormatTTL(tc.in); got != tc.want {
+			t.Errorf("FormatTTL(%d) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestFormatTTLRoundTrip(t *testing.T) {
+	for _, ttl := range []uint32{0, 1, 59, 60, 300, 3600, 3601, 86400, 604800, 1209600, 4294967295} {
+		s := FormatTTL(ttl)
+		got, err := ParseTTL(s)
+		if err != nil {
+			t.Errorf("ParseTTL(FormatTTL(%d)=%q) error: %v", ttl, s, err)
+			continue
+		}
+		if got != ttl {
+			t.Errorf("round trip of %d via %q = %d", ttl, s, got)
+		}
+	}
+}
+
+func TestSerialNext(t *testing.T) {
+	tests := []struct {
+		name    string
+		current uint32
+		spec    string
+		want    uint32
+		wantErr bool
+	}{
+		{name: "increment", current: 1, spec: "+1", want: 2},
+		{name: "increment large", current: 2024010101, spec: "+100", want: 2024010201},
+		{name: "increment from zero", current: 0, spec: "+7", want: 7},
+		{name: "decrement", current: 10, spec: "-1", want: 9},
+		{name: "decrement below zero wraps", current: 0, spec: "-1", want: 4294967295},
+		{name: "absolute", current: 5, spec: "42", want: 42},
+		{name: "absolute max", current: 5, spec: "4294967295", want: 4294967295},
+		{name: "surrounding space", current: 1, spec: "  +1  ", want: 2},
+		{name: "empty", current: 1, spec: "", wantErr: true},
+		{name: "only plus", current: 1, spec: "+", wantErr: true},
+		{name: "not a number", current: 1, spec: "+x", wantErr: true},
+		{name: "word", current: 1, spec: "next", wantErr: true},
+		{name: "bare minus", current: 1, spec: "-", wantErr: true},
+		{name: "overflow", current: 1, spec: "+4294967296", wantErr: true},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got, err := SerialNext(tc.current, tc.spec)
+			if tc.wantErr {
+				if err == nil {
+					t.Fatalf("SerialNext(%d, %q) = %d, want an error", tc.current, tc.spec, got)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("SerialNext(%d, %q): %v", tc.current, tc.spec, err)
+			}
+			if got != tc.want {
+				t.Errorf("SerialNext(%d, %q) = %d, want %d", tc.current, tc.spec, got, tc.want)
+			}
+		})
+	}
+}
+
+// addSerial is plain modular (RFC 1982 §2.1) arithmetic: 0xFFFFFFFF + 1 is 0,
+// not 1. The high-half special case only stops two large operands from being
+// mistaken for a wrap.
+func TestAddSerial(t *testing.T) {
+	tests := []struct{ a, b, want uint32 }{
+		{1, 1, 2},
+		{0, 0, 0},
+		{4294967295, 1, 0},
+		{4294967295, 2, 1},
+		{4294967294, 3, 1},
+		{0, 4294967295, 4294967295},
+		{0x7FFFFFFF, 1, 0x80000000},
+		{0x80000000, 0x80000000, 0},
+		{0x80000000, 0x7FFFFFFF, 0xFFFFFFFF},
+	}
+	for _, tc := range tests {
+		if got := addSerial(tc.a, tc.b); got != tc.want {
+			t.Errorf("addSerial(%d, %d) = %d, want %d", tc.a, tc.b, got, tc.want)
+		}
+	}
+}
+
+const fullZone = `$ORIGIN example.com.
+$TTL 1h
+@       300 IN SOA ns1.example.com. hostmaster.example.com. (
+                        2024010101 ; serial
+                        7200       ; refresh
+                        3600       ; retry
+                        1209600    ; expire
+                        300 )      ; minimum
+@       IN NS   ns1.example.com.
+www     3600 IN A     192.0.2.1
+www     600 IN A 192.0.2.2
+        AAAA 2001:db8::1
+alias   3600 IN CNAME www.example.com.
+mail    3600 IN MX    10 mx1.example.com.
+note    3600 IN TXT   "hello ; world @ example"
+*.wild  3600 IN A     192.0.2.3
+198.51.100.5. IN A 198.51.100.5
+low     in  a 192.0.2.7
+mixed   600 IN A 192.0.2.8
+cls     in 3600 A 192.0.2.9
+at      3600 IN TXT   "@"
+
+; a whole-line comment
+	; an indented comment
+
+`
+
+func TestParseFullZone(t *testing.T) {
+	z := mustParse(t, fullZone, Options{File: "full.zone", Origin: "ignored.example."})
+	if z.Origin != "example.com." {
+		t.Errorf("Origin = %q, want %q", z.Origin, "example.com.")
+	}
+	if z.File != "full.zone" {
+		t.Errorf("File = %q, want %q", z.File, "full.zone")
+	}
+	// The SOA and NS lines are folded, so every record reports the line its
+	// statement started on. TTLs are left out here: only some records carry an
+	// explicit one, see TestParseTTLDirective for the $TTL handling.
+	if got, want := summaryNoTTL(t, z), []string{
+		`3 example.com. IN SOA ns1.example.com. hostmaster.example.com. 2024010101 7200 3600 1209600 300`,
+		`9 example.com. IN NS ns1.example.com.`,
+		`10 www.example.com. IN A 192.0.2.1`,
+		`11 www.example.com. IN A 192.0.2.2`,
+		`12 www.example.com. IN AAAA 2001:db8::1`,
+		`13 alias.example.com. IN CNAME www.example.com.`,
+		`14 mail.example.com. IN MX 10 mx1.example.com.`,
+		`15 note.example.com. IN TXT "hello ; world @ example"`,
+		`16 *.wild.example.com. IN A 192.0.2.3`,
+		`17 198.51.100.5. IN A 198.51.100.5`,
+		`18 low.example.com. IN A 192.0.2.7`,
+		`19 mixed.example.com. IN A 192.0.2.8`,
+		`20 cls.example.com. IN A 192.0.2.9`,
+		`21 at.example.com. IN TXT "@"`,
+	}; !slices.Equal(got, want) {
+		t.Errorf("records mismatch\n got: %q\nwant: %q", got, want)
+	}
+	if z.SOA == nil {
+		t.Fatal("SOA = nil, want the parsed start of authority")
+	}
+	if z.SOA.Serial != 2024010101 {
+		t.Errorf("SOA.Serial = %d, want 2024010101", z.SOA.Serial)
+	}
+	if got := recordByLine(t, z, 11).TTL; got != 600 {
+		t.Errorf("explicit TTL = %d, want 600", got)
+	}
+	if got := recordByLine(t, z, 10).RR.Header().Ttl; got != 3600 {
+		t.Errorf("record TTL of the first www = %d, want the explicit 3600", got)
+	}
+	if got := recordByLine(t, z, 20).TTL; got != 3600 {
+		t.Errorf("TTL before the class = %d, want 3600", got)
+	}
+	if z.MinTTL != 300 {
+		t.Errorf("MinTTL = %d, want 300 (the smallest explicit TTL)", z.MinTTL)
+	}
+	for _, rec := range z.Records {
+		if rec.RR.Header().Class != dns.ClassINET {
+			t.Errorf("record %q has class %s, want IN", rec.Text(), dns.ClassToString[rec.RR.Header().Class])
+		}
+		if rec.Name != strings.ToLower(rec.Name) {
+			t.Errorf("Record.Name = %q, want lowercase", rec.Name)
+		}
+		if !strings.HasSuffix(rec.Name, ".") {
+			t.Errorf("Record.Name = %q, want a fully qualified name", rec.Name)
+		}
+		if rec.File != "full.zone" {
+			t.Errorf("record %q has File %q, want full.zone", rec.Text(), rec.File)
+		}
+	}
+}
+
+func TestParseTTLDirective(t *testing.T) {
+	z := mustParse(t, `$ORIGIN example.com.
+$TTL 1h
+one 3600 IN A 192.0.2.1
+two 300 IN A 192.0.2.2
+three 300 IN A 192.0.2.3
+$TTL 2h
+four 7200 IN A 192.0.2.4
+`, Options{})
+	checkSummary(t, z, []string{
+		`3 one.example.com. 3600 IN A 192.0.2.1`,
+		`4 two.example.com. 300 IN A 192.0.2.2`,
+		`5 three.example.com. 300 IN A 192.0.2.3`,
+		`7 four.example.com. 7200 IN A 192.0.2.4`,
+	})
+	if z.DefaultTTL != 7200 {
+		t.Errorf("DefaultTTL = %d, want 7200 after \"$TTL 2h\"", z.DefaultTTL)
+	}
+	if z.MinTTL != 300 {
+		t.Errorf("MinTTL = %d, want 300", z.MinTTL)
+	}
+}
+
+func TestParseDefaultTTLSources(t *testing.T) {
+	t.Run("option default ttl", func(t *testing.T) {
+		z := mustParse(t, "host IN A 192.0.2.1\n", Options{Origin: "example.com.", DefaultTTL: 42})
+		if z.DefaultTTL != 42 {
+			t.Errorf("DefaultTTL = %d, want 42", z.DefaultTTL)
+		}
+		if len(z.Records) != 1 {
+			t.Fatalf("got %d records, want 1", len(z.Records))
+		}
+	})
+
+	t.Run("no ttl source at all", func(t *testing.T) {
+		z := mustParse(t, "host IN A 192.0.2.1\n", Options{Origin: "example.com."})
+		if z.DefaultTTL != 0 {
+			t.Errorf("DefaultTTL = %d, want 0", z.DefaultTTL)
+		}
+		// The RR header already carries the library default, so MinTTL is
+		// that value and the parser's "no TTL at all" fallback never runs.
+		if z.MinTTL != z.Records[0].RR.Header().Ttl {
+			t.Errorf("MinTTL = %d, want the record TTL %d", z.MinTTL, z.Records[0].RR.Header().Ttl)
+		}
+	})
+
+	t.Run("soa ttl becomes the default", func(t *testing.T) {
+		z := mustParse(t, "@ 300 IN SOA ns1.example.com. hostmaster.example.com. 1 2 3 4 5\nhost IN A 192.0.2.1\n",
+			Options{Origin: "example.com."})
+		if z.DefaultTTL != 300 {
+			t.Errorf("DefaultTTL = %d, want 300 from the SOA", z.DefaultTTL)
+		}
+	})
+
+	t.Run("soa minimum is the last resort", func(t *testing.T) {
+		z := mustParse(t, "@ IN SOA ns1.example.com. hostmaster.example.com. 1 2 3 4 7\nhost IN A 192.0.2.1\n",
+			Options{Origin: "example.com."})
+		// The SOA carries no explicit TTL, so its header TTL wins over Minttl.
+		if z.SOA == nil || z.SOA.Minttl != 7 {
+			t.Fatalf("SOA = %v, want Minttl 7", z.SOA)
+		}
+	})
+
+	t.Run("a file without ttl, soa or option keeps the header ttl", func(t *testing.T) {
+		z := mustParse(t, "host IN A 192.0.2.1\n", Options{Origin: "example.com."})
+		if got := z.Records[0].RR.Header().Ttl; got == 0 {
+			t.Errorf("record header TTL = 0, want the parser default")
+		}
+	})
+}
+
+func TestParseOwnerForms(t *testing.T) {
+	tests := []struct {
+		name string
+		src  string
+		want []string
+	}{
+		{
+			name: "at sign is the origin",
+			src:  "@ IN A 192.0.2.1\n@ IN A 192.0.2.2\n",
+			want: []string{"origin.example. 192.0.2.1", "origin.example. 192.0.2.2"},
+		},
+		{
+			name: "at sign follows $ORIGIN",
+			src:  "$ORIGIN sub.example.com.\n@ IN A 192.0.2.1\n",
+			want: []string{"sub.example.com. 192.0.2.1"},
+		},
+		{
+			// An omitted owner is recognised by the line starting with a type
+			// token; a leading class or TTL is taken as an owner name instead.
+			name: "omitted owner repeats the previous one",
+			src:  "one IN A 192.0.2.1\n  A 192.0.2.2\n  AAAA 2001:db8::1\n  A 192.0.2.3\n",
+			want: []string{
+				"one.origin.example. 192.0.2.1",
+				"one.origin.example. 192.0.2.2",
+				"one.origin.example. 2001:db8::1",
+				"one.origin.example. 192.0.2.3",
+			},
+		},
+		{
+			name: "omitted owner after a parenthesised record",
+			src:  "one IN TXT ( \"one\" ; comment\n            \"two\" )\n  A 192.0.2.1\n",
+			want: []string{"one.origin.example. \"one\" \"two\"", "one.origin.example. 192.0.2.1"},
+		},
+		{
+			name: "omitted owner is the origin after a directive",
+			src:  "$ORIGIN example.com.\nA 192.0.2.1\n",
+			want: []string{"example.com. 192.0.2.1"},
+		},
+		{
+			name: "wildcard owner keeps the star",
+			src:  "*.sub IN A 192.0.2.1\n",
+			want: []string{"*.sub.origin.example. 192.0.2.1"},
+		},
+		{
+			name: "at sign in a value is not expanded",
+			src:  "note IN TXT \"user @ example.com\"\n",
+			want: []string{"note.origin.example. \"user @ example.com\""},
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			z := mustParse(t, tc.src, Options{Origin: "origin.example."})
+			got := make([]string, 0, len(z.Records))
+			for _, rec := range z.Records {
+				parts := strings.SplitN(rec.RR.String(), "\t", 5)
+				got = append(got, parts[0]+" "+parts[4])
+			}
+			if !slices.Equal(got, tc.want) {
+				t.Errorf("got %q, want %q", got, tc.want)
+			}
+			// Record.Name is the resolved, lowercased, fully qualified owner:
+			// "@" always resolves to the origin, never to a literal "@" label.
+			for _, rec := range z.Records {
+				if strings.ContainsAny(rec.Name, "@") {
+					t.Errorf("Record.Name = %q, want an expanded owner", rec.Name)
+				}
+				if !strings.HasSuffix(rec.Name, ".") {
+					t.Errorf("Record.Name = %q, want a fully qualified name", rec.Name)
+				}
+				if rec.Name != strings.ToLower(rec.Name) {
+					t.Errorf("Record.Name = %q, want lowercase", rec.Name)
+				}
+			}
+		})
+	}
+}
+
+func TestParseOwnerNameCaseIsPreservedInTheRecord(t *testing.T) {
+	// The RR keeps the case the file used; only Record.Name is normalised.
+	z := mustParse(t, "WWW IN A 192.0.2.1\n", Options{Origin: "example.com."})
+	if got := z.Records[0].Name; got != "www.example.com." {
+		t.Errorf("Record.Name = %q, want www.example.com.", got)
+	}
+	if !strings.HasPrefix(z.Records[0].Text(), "WWW.example.com.") {
+		t.Errorf("Text() = %q, want the owner as written", z.Records[0].Text())
+	}
+}
+
+func TestParseParenthesised(t *testing.T) {
+	t.Run("multi line soa", func(t *testing.T) {
+		z := mustParse(t, `@ SOA ns1.example.com. hostmaster.example.com. (
+   2024010101
+   7200
+   3600
+   1209600
+   300 )
+`, Options{Origin: "example.com."})
+		if z.SOA == nil {
+			t.Fatal("SOA = nil")
+		}
+		if z.SOA.Serial != 2024010101 || z.SOA.Refresh != 7200 || z.SOA.Minttl != 300 {
+			t.Errorf("SOA = %v, want serial 2024010101, refresh 7200, minimum 300", z.SOA)
+		}
+		if z.Records[0].Line != 1 {
+			t.Errorf("Line = %d, want 1 (the start of the folded record)", z.Records[0].Line)
+		}
+	})
+
+	t.Run("comment inside parentheses", func(t *testing.T) {
+		z := mustParse(t, "host TXT ( \"one\" ; first\n        \"two\" ; second\n        )\n", Options{Origin: "example.com."})
+		if len(z.Records) != 1 {
+			t.Fatalf("got %d records, want 1", len(z.Records))
+		}
+		if got := summaryNoTTL(t, z)[0]; !strings.HasSuffix(got, `IN TXT "one" "two"`) {
+			t.Errorf("record = %q, want a two-string TXT", got)
+		}
+	})
+
+	t.Run("parentheses on a single line", func(t *testing.T) {
+		z := mustParse(t, "host MX ( 10 mx1.example.com. )\n", Options{Origin: "example.com."})
+		if len(z.Records) != 1 {
+			t.Fatalf("got %d records, want 1", len(z.Records))
+		}
+		if got := summaryNoTTL(t, z)[0]; !strings.HasSuffix(got, "IN MX 10 mx1.example.com.") {
+			t.Errorf("record = %q, want one MX rdata field", got)
+		}
+	})
+
+	t.Run("parentheses are stripped from the assembled line", func(t *testing.T) {
+		lineno := 0
+		lines := bufio.NewScanner(strings.NewReader("host MX 10 mx1.example.com.\n"))
+		got, err := fold(lines, "host MX ( 10 mx1.example.com. )", &lineno, Options{})
+		if err != nil {
+			t.Fatalf("fold: %v", err)
+		}
+		if strings.ContainsAny(got, "()") {
+			t.Errorf("fold = %q, want no parentheses", got)
+		}
+		// The parens and the whitespace they wrapped are gone; the token
+		// stream is what the record parser cares about, not the exact spacing.
+		if want := []string{"host", "MX", "10", "mx1.example.com."}; !slices.Equal(strings.Fields(got), want) {
+			t.Errorf("fold = %q, tokens = %v, want %v", got, strings.Fields(got), want)
+		}
+	})
+}
+
+func TestParseOriginDirective(t *testing.T) {
+	t.Run("relative names follow $ORIGIN", func(t *testing.T) {
+		z := mustParse(t, "$ORIGIN sub.example.com.\nwww IN A 192.0.2.1\n", Options{Origin: "example.com."})
+		if z.Origin != "sub.example.com." {
+			t.Errorf("Origin = %q, want sub.example.com.", z.Origin)
+		}
+		if z.Records[0].Name != "www.sub.example.com." {
+			t.Errorf("name = %q, want www.sub.example.com.", z.Records[0].Name)
+		}
+	})
+
+	t.Run("absolute names are left alone", func(t *testing.T) {
+		z := mustParse(t, "$ORIGIN sub.example.com.\nwww.other.example. IN A 192.0.2.1\n", Options{})
+		if z.Records[0].Name != "www.other.example." {
+			t.Errorf("name = %q, want www.other.example.", z.Records[0].Name)
+		}
+	})
+
+	t.Run("no origin at all is the root", func(t *testing.T) {
+		z := mustParse(t, "host IN A 192.0.2.1\n", Options{})
+		if z.Origin != "." {
+			t.Errorf("Origin = %q, want .", z.Origin)
+		}
+		if z.Records[0].Name != "host." {
+			t.Errorf("name = %q, want host.", z.Records[0].Name)
+		}
+	})
+
+	t.Run("relative names at the root get no suffix", func(t *testing.T) {
+		z := mustParse(t, "host IN A 192.0.2.1\n", Options{Origin: "."})
+		if z.Records[0].Name != "host." {
+			t.Errorf("name = %q, want host.", z.Records[0].Name)
+		}
+	})
+
+	t.Run("$ORIGIN without a name is rejected", func(t *testing.T) {
+		// BIND resets to the zone origin when $ORIGIN carries no argument;
+		// this parser treats it as an error instead.
+		err := parseErr(t, "$ORIGIN example.com.\n$ORIGIN\n", Options{})
+		if !strings.Contains(err.Error(), "$ORIGIN needs a name") {
+			t.Errorf("error = %q, want it to mention \"$ORIGIN needs a name\"", err)
+		}
+	})
+
+	t.Run("origin option applies when the file has no $ORIGIN", func(t *testing.T) {
+		z := mustParse(t, "www IN A 192.0.2.1\n", Options{Origin: "example.com"})
+		if z.Origin != "example.com." || z.Records[0].Name != "www.example.com." {
+			t.Errorf("origin = %q, name = %q", z.Origin, z.Records[0].Name)
+		}
+	})
+}
+
+func TestParseClassHandling(t *testing.T) {
+	z := mustParse(t, `one IN A 192.0.2.1
+two in A 192.0.2.2
+three A 192.0.2.3
+four 300 in A 192.0.2.4
+five ch A 192.0.2.5
+six hs A 192.0.2.6
+seven 300 A 192.0.2.7
+`, Options{Origin: "example.com."})
+	if len(z.Records) != 7 {
+		t.Fatalf("got %d records, want 7", len(z.Records))
+	}
+	// An omitted class defaults to IN; "in" and "IN" are the same token.
+	want := []uint16{
+		dns.ClassINET, dns.ClassINET, dns.ClassINET, dns.ClassINET,
+		dns.ClassCHAOS, dns.ClassHESIOD, dns.ClassINET,
+	}
+	for i, rec := range z.Records {
+		if rec.RR.Header().Class != want[i] {
+			t.Errorf("%q has class %s, want %s", rec.Text(),
+				dns.ClassToString[rec.RR.Header().Class], dns.ClassToString[want[i]])
+		}
+	}
+}
+
+func TestParseCommentsAndBlankLines(t *testing.T) {
+	z := mustParse(t, `; leading comment
+$TTL 1h
+
+   ; indented comment
+
+one 300 IN A 192.0.2.1 ; trailing comment
+
+
+two 300 IN A 192.0.2.2 ;comment without space
+three 300 IN TXT "semi ; colon" ; real comment
+`, Options{Origin: "example.com."})
+	if got := summary(t, z); !slices.Equal(got, []string{
+		`6 one.example.com. 300 IN A 192.0.2.1`,
+		`9 two.example.com. 300 IN A 192.0.2.2`,
+		`10 three.example.com. 300 IN TXT "semi ; colon"`,
+	}) {
+		t.Errorf("records = %q, want three records on lines 6, 9 and 10", got)
+	}
+}
+
+func TestParseErrors(t *testing.T) {
+	tests := []struct {
+		name    string
+		src     string
+		opt     Options
+		wantSub string
+	}{
+		{name: "no type", src: "host 3600\n", wantSub: "has no type"},
+		{name: "bad token before type", src: "host BOGUS 1.2.3.4\n", wantSub: `unexpected token "BOGUS"`},
+		{name: "bad rdata", src: "host 3600 IN A not-an-ip\n"},
+		{name: "unterminated parenthesis", src: "host 3600 IN A ( 192.0.2.1\n", wantSub: "unterminated parenthesis"},
+		{name: "unterminated parenthesis with a blank line", src: "host 3600 IN A (\n\n  192.0.2.1\n", wantSub: "unterminated parenthesis"},
+		{name: "unterminated quote", src: "host 3600 IN TXT \"unterminated\n"},
+		{name: "no owner and no previous owner", src: "A 192.0.2.1\n", wantSub: "no owner name"},
+		{name: "bad ttl directive", src: "$TTL forever\n", wantSub: "bad TTL"},
+		{name: "ttl directive without a value", src: "$TTL\n", wantSub: "$TTL needs a value"},
+		{name: "origin directive without a name", src: "$ORIGIN\n", wantSub: "$ORIGIN needs a name"},
+		{name: "include disabled", src: "$INCLUDE other.zone\n", wantSub: "disabled"},
+		{name: "include unsupported", src: "$INCLUDE other.zone\n", opt: Options{MaxIncludeDepth: 3}, wantSub: "not supported"},
+		{name: "include without a name", src: "$INCLUDE\n", wantSub: "$INCLUDE needs a file name"},
+		{name: "error carries file and line", src: "host 3600 IN A 192.0.2.1\nother 3600 IN A nope\n", opt: Options{File: "bad.zone"}, wantSub: "bad.zone:2:"},
+		{name: "directive error carries the line", src: "$TTL 1h\n$TTL\n", opt: Options{File: "bad.zone"}, wantSub: "bad.zone:2:"},
+		{name: "read error is reported", src: "host 3600 IN A 192.0.2.1\n", opt: Options{File: "bad.zone"}},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if tc.name == "read error is reported" {
+				// A reader that fails after the first line exercises the
+				// lines.Err() path in Parse.
+				if _, err := Parse(&failingReader{}, Options{File: "bad.zone"}); err == nil {
+					t.Fatal("Parse with a failing reader: want an error")
+				}
+				return
+			}
+			err := parseErr(t, tc.src, tc.opt)
+			if tc.wantSub != "" && !strings.Contains(err.Error(), tc.wantSub) {
+				t.Errorf("error = %q, want it to contain %q", err, tc.wantSub)
+			}
+		})
+	}
+}
+
+// failingReader fails on the first Read so Parse must surface the error.
+type failingReader struct{}
+
+func (failingReader) Read([]byte) (int, error) { return 0, errRead }
+
+var errRead = fmt.Errorf("synthetic read failure")
+
+func TestParseEmptyFile(t *testing.T) {
+	for _, src := range []string{"", "\n\n", "   \n\t\n", "; only a comment\n"} {
+		z := mustParse(t, src, Options{Origin: "example.com."})
+		if len(z.Records) != 0 {
+			t.Errorf("Parse(%q) returned %d records, want 0", src, len(z.Records))
+		}
+		if z.SOA != nil {
+			t.Errorf("Parse(%q) returned an SOA, want nil", src)
+		}
+		if z.Origin != "example.com." {
+			t.Errorf("Parse(%q) origin = %q, want example.com.", src, z.Origin)
+		}
+	}
+}
+
+func TestParseFileAndParseFiles(t *testing.T) {
+	dir := t.TempDir()
+	a := filepath.Join(dir, "a.zone")
+	b := filepath.Join(dir, "b.zone")
+	if err := os.WriteFile(a, []byte("$TTL 1h\nwww IN A 192.0.2.1\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(b, []byte("$TTL 2h\nmail IN A 192.0.2.2\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	z, err := ParseFile(a, "example.com.")
+	if err != nil {
+		t.Fatalf("ParseFile: %v", err)
+	}
+	if z.File != a || z.Origin != "example.com." || len(z.Records) != 1 {
+		t.Fatalf("ParseFile = %+v", z)
+	}
+	if z.Records[0].Name != "www.example.com." {
+		t.Errorf("name = %q, want www.example.com.", z.Records[0].Name)
+	}
+	if z.Records[0].File != a {
+		t.Errorf("record File = %q, want %q", z.Records[0].File, a)
+	}
+	if _, err := ParseFile(filepath.Join(dir, "missing.zone"), "example.com."); err == nil {
+		t.Error("ParseFile on a missing file: want an error")
+	}
+
+	merged, err := ParseFiles([]string{a, b}, OriginFromPath)
+	if err != nil {
+		t.Fatalf("ParseFiles: %v", err)
+	}
+	if len(merged.Records) != 2 {
+		t.Fatalf("ParseFiles returned %d records, want 2", len(merged.Records))
+	}
+	if merged.DefaultTTL != 7200 {
+		t.Errorf("merged DefaultTTL = %d, want 7200 (the last file wins)", merged.DefaultTTL)
+	}
+	if _, err := ParseFiles([]string{a, filepath.Join(dir, "missing.zone")}, OriginFromPath); err == nil {
+		t.Error("ParseFiles with a missing file: want an error")
+	}
+}
+
+func TestOriginFromPath(t *testing.T) {
+	tests := []struct{ in, want string }{
+		{"ads.zone", "ads."},
+		{"rpz/ads.zone", "ads."},
+		{"plain", "plain."},
+		{".hidden", ".hidden."},
+		{"/var/lib/fancy-dns/ads.zone", "ads."},
+	}
+	for _, tc := range tests {
+		if got := OriginFromPath(tc.in); got != tc.want {
+			t.Errorf("OriginFromPath(%q) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestEnsureDot(t *testing.T) {
+	tests := []struct{ in, want string }{
+		{"", ""},
+		{".", "."},
+		{"example.com", "example.com."},
+		{"example.com.", "example.com."},
+	}
+	for _, tc := range tests {
+		if got := ensureDot(tc.in); got != tc.want {
+			t.Errorf("ensureDot(%q) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestRelative(t *testing.T) {
+	tests := []struct{ name, origin, want string }{
+		{"@", "example.com.", "example.com."},
+		{"www", "example.com.", "www.example.com."},
+		{"www.other.", "example.com.", "www.other."},
+		{"www", ".", "www."},
+		{"*", "example.com.", "*.example.com."},
+		{"www", "", "www."},
+	}
+	for _, tc := range tests {
+		if got := relative(tc.name, tc.origin); got != tc.want {
+			t.Errorf("relative(%q, %q) = %q, want %q", tc.name, tc.origin, got, tc.want)
+		}
+	}
+}
+
+func TestStripComment(t *testing.T) {
+	tests := []struct{ in, want string }{
+		{"a ; comment", "a "},
+		{"a;comment", "a"},
+		{`a "has ; semicolon" b`, `a "has ; semicolon" b`},
+		{`a \; escaped`, `a \; escaped`},
+		{"no comment here", "no comment here"},
+		{`"unterminated ; quote`, `"unterminated ; quote`},
+	}
+	for _, tc := range tests {
+		if got := stripComment(tc.in); got != tc.want {
+			t.Errorf("stripComment(%q) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestSplitFields(t *testing.T) {
+	tests := []struct {
+		in   string
+		want []string
+	}{
+		{"a b  c", []string{"a", "b", "c"}},
+		{`a "b c" d`, []string{"a", `"b c"`, "d"}},
+		{"  ", nil},
+		{`a\ b c`, []string{`a\ b`, "c"}},
+		{"a\tb", []string{"a", "b"}},
+	}
+	for _, tc := range tests {
+		if got := splitFields(tc.in); !slices.Equal(got, tc.want) {
+			t.Errorf("splitFields(%q) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestDeparen(t *testing.T) {
+	tests := []struct{ in, want string }{
+		{"a ( b", "a   b"},
+		{"a ) b", "a   b"},
+		{"a b", "a b"},
+		{"()", ""},
+	}
+	for _, tc := range tests {
+		if got := deparen(tc.in); got != tc.want {
+			t.Errorf("deparen(%q) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestTokenClassification(t *testing.T) {
+	for _, s := range []string{"0", "300", "1h30m", "300s"} {
+		if !isTTLToken(s) {
+			t.Errorf("isTTLToken(%q) = false, want true", s)
+		}
+	}
+	for _, s := range []string{"", "-1", "IN", "in", "CH", "HS", "1x", "forever"} {
+		if isTTLToken(s) {
+			t.Errorf("isTTLToken(%q) = true, want false", s)
+		}
+	}
+	for _, s := range []string{"IN", "in", "In", "CH", "ch", "HS", "hs"} {
+		if !isClassToken(s) {
+			t.Errorf("isClassToken(%q) = false, want true", s)
+		}
+	}
+	for _, s := range []string{"", "300", "ANY", "cs", "in4"} {
+		if isClassToken(s) {
+			t.Errorf("isClassToken(%q) = true, want false", s)
+		}
+	}
+	for _, s := range []string{"A", "a", "AAAA", "ns", "CNAME", "SOA", "ANY", "SVCB", "HTTPS", "nsec3"} {
+		if !isRRTYPE(s) {
+			t.Errorf("isRRTYPE(%q) = false, want true", s)
+		}
+	}
+	for _, s := range []string{"", "www", "IN", "300", "BOGUS", "www.example.com."} {
+		if isRRTYPE(s) {
+			t.Errorf("isRRTYPE(%q) = true, want false", s)
+		}
+	}
+}
+
+func TestCloneSOA(t *testing.T) {
+	if cloneSOA(nil) != nil {
+		t.Error("cloneSOA(nil) != nil")
+	}
+	in := &dns.SOA{Hdr: dns.RR_Header{Name: "example.com."}, Serial: 7, Minttl: 60}
+	out := cloneSOA(in)
+	if out == in {
+		t.Fatal("cloneSOA returned the same pointer")
+	}
+	if out.Serial != 7 || out.Minttl != 60 {
+		t.Errorf("cloneSOA copied %+v", out)
+	}
+}
+
+func TestWhereAndDirOf(t *testing.T) {
+	if got := where(""); got != "zone" {
+		t.Errorf("where(\"\") = %q, want zone", got)
+	}
+	if got := where("a.zone"); got != "a.zone" {
+		t.Errorf("where(a.zone) = %q", got)
+	}
+	if got := dirOf("a.zone"); got != "." {
+		t.Errorf("dirOf(a.zone) = %q, want .", got)
+	}
+	if got := dirOf("/x/y/a.zone"); got != "/x/y" {
+		t.Errorf("dirOf(/x/y/a.zone) = %q", got)
+	}
+}

web/src/App.tsx

@@ -1,122 +1,84 @@
-import { useState } from 'react'
+/**
-import heroImg from './assets/hero.png'
+ * App — routes, the shell and the theme bootstrap.
-import reactLogo from './assets/react.svg'
+ *
-import viteLogo from './assets/vite.svg'
+ * The daemon serves the console and the API from one origin, so a real
-import './App.css'
+ * BrowserRouter works as long as the Go handler falls back to index.html (it
+ * does). `useThemeInit` runs the stored theme onto `<html>` before the first
+ * paint; `ThemeToggle` inside AppShell keeps it in sync afterwards.
+ */
 
-function App() {
+import { useEffect } from 'react'
-  const [count, setCount] = useState(0)
+import { BrowserRouter, Route, Routes } from 'react-router-dom'
 
+import { ToastProvider } from './components/ui'
+import { AppShell } from './components/layout'
+import Blocklists from './pages/Blocklists.tsx'
+import Cache from './pages/Cache.tsx'
+import Clients from './pages/Clients.tsx'
+import Dashboard from './pages/Dashboard.tsx'
+import Domains from './pages/Domains.tsx'
+import Login from './pages/Login.tsx'
+import Logs from './pages/Logs.tsx'
+import NotFound from './pages/NotFound.tsx'
+import Settings from './pages/Settings.tsx'
+import Upstreams from './pages/Upstreams.tsx'
+
+const THEME_KEY = 'fancydns.theme'
+
+/** Applies the stored theme to `<html>` before anything paints. */
+function useThemeInit(): void {
+  useEffect(() => {
+    if (typeof document === 'undefined') return
+    if (document.documentElement.dataset.theme !== undefined) return
+    let stored: string | null = null
+    try {
+      stored = localStorage.getItem(THEME_KEY)
+    } catch {
+      // Storage disabled: fall back to the system preference.
+      stored = null
+    }
+    if (stored === 'light' || stored === 'dark') {
+      document.documentElement.dataset.theme = stored
+    }
+  }, [])
+}
+
+/** Everything that lives inside the shell. */
+function AppRoutes() {
   return (
-    <>
+    <Routes>
-      <section id="center">
+      <Route path="/" element={<Dashboard />} />
-        <div className="hero">
+      <Route path="/blocklists" element={<Blocklists />} />
-          <img src={heroImg} className="base" width="170" height="179" alt="" />
+      <Route path="/domains" element={<Domains />} />
-          <img src={reactLogo} className="framework" alt="React logo" />
+      <Route path="/clients" element={<Clients />} />
-          <img src={viteLogo} className="vite" alt="Vite logo" />
+      <Route path="/upstreams" element={<Upstreams />} />
-        </div>
+      <Route path="/logs" element={<Logs />} />
-        <div>
+      <Route path="/cache" element={<Cache />} />
-          <h1>Get started</h1>
+      <Route path="/settings" element={<Settings />} />
-          <p>
+      <Route path="*" element={<NotFound />} />
-            Edit <code>src/App.tsx</code> and save to test <code>HMR</code>
+    </Routes>
-          </p>
-        </div>
-        <button
-          type="button"
-          className="counter"
-          onClick={() => setCount((count) => count + 1)}
-        >
-          Count is {count}
-        </button>
-      </section>
-
-      <div className="ticks"></div>
-
-      <section id="next-steps">
-        <div id="docs">
-          <svg className="icon" role="presentation" aria-hidden="true">
-            <use href="/icons.svg#documentation-icon"></use>
-          </svg>
-          <h2>Documentation</h2>
-          <p>Your questions, answered</p>
-          <ul>
-            <li>
-              <a href="https://vite.dev/" target="_blank">
-                <img className="logo" src={viteLogo} alt="" />
-                Explore Vite
-              </a>
-            </li>
-            <li>
-              <a href="https://react.dev/" target="_blank">
-                <img className="button-icon" src={reactLogo} alt="" />
-                Learn more
-              </a>
-            </li>
-          </ul>
-        </div>
-        <div id="social">
-          <svg className="icon" role="presentation" aria-hidden="true">
-            <use href="/icons.svg#social-icon"></use>
-          </svg>
-          <h2>Connect with us</h2>
-          <p>Join the Vite community</p>
-          <ul>
-            <li>
-              <a href="https://github.com/vitejs/vite" target="_blank">
-                <svg
-                  className="button-icon"
-                  role="presentation"
-                  aria-hidden="true"
-                >
-                  <use href="/icons.svg#github-icon"></use>
-                </svg>
-                GitHub
-              </a>
-            </li>
-            <li>
-              <a href="https://chat.vite.dev/" target="_blank">
-                <svg
-                  className="button-icon"
-                  role="presentation"
-                  aria-hidden="true"
-                >
-                  <use href="/icons.svg#discord-icon"></use>
-                </svg>
-                Discord
-              </a>
-            </li>
-            <li>
-              <a href="https://x.com/vite_js" target="_blank">
-                <svg
-                  className="button-icon"
-                  role="presentation"
-                  aria-hidden="true"
-                >
-                  <use href="/icons.svg#x-icon"></use>
-                </svg>
-                X.com
-              </a>
-            </li>
-            <li>
-              <a href="https://bsky.app/profile/vite.dev" target="_blank">
-                <svg
-                  className="button-icon"
-                  role="presentation"
-                  aria-hidden="true"
-                >
-                  <use href="/icons.svg#bluesky-icon"></use>
-                </svg>
-                Bluesky
-              </a>
-            </li>
-          </ul>
-        </div>
-      </section>
-
-      <div className="ticks"></div>
-      <section id="spacer"></section>
-    </>
   )
 }
 
-export default App
+export default function App() {
+  useThemeInit()
+
+  return (
+    <BrowserRouter>
+      <ToastProvider>
+        <Routes>
+          {/* Bare: no shell, no header, no nav. */}
+          <Route path="/auth/login" element={<Login />} />
+          <Route
+            path="*"
+            element={
+              <AppShell>
+                <AppRoutes />
+              </AppShell>
+            }
+          />
+        </Routes>
+      </ToastProvider>
+    </BrowserRouter>
+  )
+}

web/src/hooks/useApi.ts

@@ -27,7 +27,9 @@ export function useApi<T>(fn: () => Promise<T>, deps: unknown[] = []): ApiState<
   fnRef.current = fn
 
   useEffect(() => {
-    const controller = new AbortController()
+    // The fetcher takes no AbortSignal, so cancellation is a guard: a response
+    // that arrives after an unmount or a newer request is dropped on the floor
+    // instead of overwriting fresher state.
     let live = true
     setLoading(true)
 
@@ -50,7 +52,6 @@ export function useApi<T>(fn: () => Promise<T>, deps: unknown[] = []): ApiState<
 
     return () => {
       live = false
-      controller.abort()
     }
     // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [...deps, nonce])

web/src/index.css

@@ -198,12 +198,6 @@
     color: var(--fg);
   }
 
-  /* Numeric columns must not jitter as digits change. */
-  .tabular {
-    font-variant-numeric: tabular-nums;
-    font-feature-settings: "tnum" 1;
-  }
-
   * {
     scrollbar-width: thin;
     scrollbar-color: var(--scroll-thumb) transparent;
@@ -244,3 +238,9 @@
     }
   }
 }
+
+/** Numeric columns must not jitter as their digits change. */
+@utility tabular {
+  font-variant-numeric: tabular-nums;
+  font-feature-settings: "tnum" 1;
+}

web/src/lib/api.ts

@@ -417,6 +417,8 @@ export function streamLogs(handlers: LogStreamHandlers): () => void {
   let backoff = RECONNECT_MIN_MS
   let timer: ReturnType<typeof setTimeout> | undefined
   let attempt = 0
+  // One error report per outage streak: a flaky link must not toast every 15s.
+  let reported = false
 
   const open = (): void => {
     if (stopped) return
@@ -447,11 +449,15 @@ export function streamLogs(handlers: LogStreamHandlers): () => void {
         const message = res.ok
           ? 'the live log stream is not supported by this browser'
           : `the live log stream failed with HTTP ${res.status}`
-        handlers.onError(message)
+        if (!reported) {
+          reported = true
+          handlers.onError(message)
+        }
         schedule()
         return
       }
 
+      reported = false
       attempt = 0
       backoff = RECONNECT_MIN_MS
       handlers.onOpen()
@@ -463,7 +469,10 @@ export function streamLogs(handlers: LogStreamHandlers): () => void {
       if (stopped) return
       if (controller.signal.aborted) return
       const message = err instanceof Error ? err.message : 'the live log stream dropped'
-      handlers.onError(message)
+      if (!reported) {
+        reported = true
+        handlers.onError(message)
+      }
       schedule()
     }
   }

web/src/main.tsx

@@ -1,9 +1,19 @@
+/**
+ * Entry point. The stylesheet is imported first so Tailwind's layers are in
+ * place before the first React render commits anything.
+ */
+
 import { StrictMode } from 'react'
 import { createRoot } from 'react-dom/client'
 import './index.css'
 import App from './App.tsx'
 
-createRoot(document.getElementById('root')!).render(
+const container = document.getElementById('root')
+if (container === null) {
+  throw new Error('fancy-dns: #root is missing from index.html')
+}
+
+createRoot(container).render(
   <StrictMode>
     <App />
   </StrictMode>,

web/src/pages/Blocklists.tsx

@@ -10,12 +10,12 @@ import { useMemo, useState } from 'react'
 import type { ReactNode } from 'react'
 import { AlertTriangle, Clock, Layers, RefreshCw, ShieldCheck, Split, Trash2 } from 'lucide-react'
 
-import { checkDomain, getBlocklists, refreshBlocklists, toggleBlocklist } from '../../lib/api'
+import { checkDomain, getBlocklists, refreshBlocklists, toggleBlocklist } from '../lib/api'
-import type { BlocklistItem, BlocklistStats, DomainCheck } from '../../lib/types'
+import type { BlocklistItem, BlocklistStats, DomainCheck } from '../lib/types'
-import { useApi, useAsyncAction } from '../../hooks/useApi'
+import { useApi, useAsyncAction } from '../hooks/useApi'
-import { useAutoRefresh } from '../../hooks/useAutoRefresh'
+import { useAutoRefresh } from '../hooks/useAutoRefresh'
-import { useDebounced } from '../../hooks/useDebounced'
+import { useDebounced } from '../hooks/useDebounced'
-import { formatBytes, formatMs, formatNumber, formatRelative, toDate } from '../../lib/format'
+import { formatBytes, formatMs, formatNumber, formatRelative, toDate } from '../lib/format'
 import {
   Badge,
   Button,
@@ -29,21 +29,21 @@ import {
   Table,
   useToast,
 } from '../components/ui'
+import type { ToastOptions } from '../components/ui'
 import { PageError, PageHeader } from '../components/layout'
 
 type Row = Record<string, ReactNode>
 
-type ToastLike = {
+/**
-  success?: (message: string) => void
+ * One line for every mutation on this page: the message plus the tone it
-  error?: (message: string) => void
+ * earned. `fail` is the danger tone, `ok` the success one.
-  toast?: (message: string) => void
+ */
-  push?: (message: string) => void
+function notify(
-}
+  toast: (options: ToastOptions) => void,
-
+  kind: 'ok' | 'fail',
-function notify(toast: unknown, kind: 'ok' | 'fail', message: string): void {
+  message: string,
-  const t = toast as ToastLike
+): void {
-  const fn = kind === 'ok' ? (t.success ?? t.toast ?? t.push) : (t.error ?? t.toast ?? t.push)
+  toast({ title: message, tone: kind === 'ok' ? 'ok' : 'danger' })
-  fn?.call(t, message)
 }
 
 const EMPTY_STATS: BlocklistStats = {
@@ -79,7 +79,7 @@ function truncate(value: string, max = 42): string {
 }
 
 export default function Blocklists() {
-  const toast = useToast()
+  const { toast } = useToast()
   const { data, error, loading, refetch } = useApi(() => getBlocklists(), [])
   useAutoRefresh(refetch, 15000)
 
@@ -200,7 +200,7 @@ export default function Blocklists() {
       ),
       refresh: (
         <IconButton
-          icon={<RefreshCw size={14} />}
+          icon={RefreshCw}
           label={`refresh ${item.name}`}
           variant="ghost"
           disabled={busy}
@@ -221,7 +221,7 @@ export default function Blocklists() {
         description="Compiled rule sources, their freshness and their health."
         actions={
           <Button
-            icon={<RefreshCw size={14} />}
+            icon={RefreshCw}
             onClick={() => void onRefresh()}
             loading={refresh.pending}
           >
@@ -231,24 +231,24 @@ export default function Blocklists() {
       />
 
       <div className="grid grid-cols-2 gap-3 lg:grid-cols-3 xl:grid-cols-5">
-        <StatCard label="Total rules" value={formatNumber(stats.total)} icon={<Layers size={16} />} />
+        <StatCard label="Total rules" value={formatNumber(stats.total)} icon={Layers} />
         <StatCard label="Sources" value={formatNumber(stats.sources)} hint={`${items.filter((i) => i.enabled).length} enabled`} />
         <StatCard
           label="Allow rules"
           value={formatNumber(stats.allow)}
-          icon={<ShieldCheck size={16} />}
+          icon={ShieldCheck}
           tone={stats.allow > 0 ? 'default' : 'warn'}
         />
         <StatCard
           label="Duplicates"
           value={formatNumber(stats.duplicates)}
-          icon={<Split size={16} />}
+          icon={Split}
           tone={stats.duplicates > 0 ? 'warn' : 'ok'}
         />
         <StatCard
           label="Last compile"
           value={compiledAgo}
-          icon={<Clock size={16} />}
+          icon={Clock}
           hint={stats.built_at ? undefined : 'no build recorded'}
         />
       </div>
@@ -265,7 +265,7 @@ export default function Blocklists() {
                 placeholder="ads.example.com"
                 value={probe}
                 onChange={(e) => setProbe(e.target.value)}
-                icon={<ShieldCheck size={14} />}
+                icon={ShieldCheck}
                 autoComplete="off"
                 spellCheck={false}
               />

web/src/pages/Cache.tsx

@@ -0,0 +1,393 @@
+/**
+ * Cache — what the resolver keeps in memory between queries.
+ *
+ * Everything here is a read of `/api/cache` plus the top-50 ranking, refreshed
+ * on a slow 15s tick. The only mutation is the flush, which is destructive
+ * enough to sit behind a confirmation: it costs a cold cache, not a restart.
+ */
+
+import { useMemo, useState } from 'react'
+import type { ReactNode } from 'react'
+import {
+  Cell,
+  Pie,
+  PieChart,
+  ResponsiveContainer,
+  Tooltip as ChartTooltip,
+} from 'recharts'
+import { Database, FileClock, Gauge, HardDrive, Layers, RefreshCw, Timer } from 'lucide-react'
+
+import { flushCache, getCache, getCacheTop } from '../lib/api'
+import type { CachePersisted, CacheTopEntry } from '../lib/types'
+import { useApi, useAsyncAction } from '../hooks/useApi'
+import { useAutoRefresh } from '../hooks/useAutoRefresh'
+import { formatBytes, formatMs, formatNumber, formatRate, formatRelative } from '../lib/format'
+import {
+  Badge,
+  Button,
+  Card,
+  CardBody,
+  ConfirmDialog,
+  EmptyState,
+  StatCard,
+  Table,
+  useToast,
+} from '../components/ui'
+import { PageError, PageHeader } from '../components/layout'
+
+type Row = Record<string, ReactNode>
+
+const REFRESH_MS = 15000
+const TOP_LIMIT = 50
+
+/** Fixed hex so the donut reads on both themes, like the dashboard charts. */
+const HITS = '#34d399'
+const MISSES = '#f87171'
+
+/** `persisted` gained fields over time; read defensively. */
+type PersistedView = Partial<CachePersisted> & {
+  errors?: number
+  error?: string
+  last_error?: string
+}
+
+type PieSlice = { name: string; value: number; fill: string }
+
+type TipProps = {
+  active?: boolean
+  payload?: { name?: string; value?: number | string; payload?: PieSlice }[]
+}
+
+function ChartTip({ active, payload }: TipProps) {
+  if (!active || payload === undefined || payload.length === 0) return null
+  return (
+    <div className="rounded-md border border-slate-700 bg-slate-900 px-3 py-2 text-xs shadow-lg">
+      <ul className="space-y-0.5">
+        {payload.map((item) => (
+          <li key={item.name} className="flex items-center gap-2 text-slate-100">
+            <span
+              aria-hidden="true"
+              className="inline-block h-2 w-2 shrink-0 rounded-full"
+              style={{ background: item.payload?.fill }}
+            />
+            <span className="text-slate-400">{item.name}</span>
+            <span className="ml-auto font-mono">{formatNumber(Number(item.value ?? 0))}</span>
+          </li>
+        ))}
+      </ul>
+    </div>
+  )
+}
+
+export default function Cache() {
+  const { toast } = useToast()
+  const notify = (title: string, description?: string, tone: 'ok' | 'fail' = 'fail'): void =>
+    toast({ title, description, tone: tone === 'ok' ? 'ok' : 'danger' })
+
+  const { data, error, loading, refetch } = useApi(() => getCache(), [])
+  const {
+    data: top,
+    error: topError,
+    loading: topLoading,
+    refetch: refetchTop,
+  } = useApi(() => getCacheTop(TOP_LIMIT), [])
+
+  useAutoRefresh(() => {
+    refetch()
+    refetchTop()
+  }, REFRESH_MS)
+
+  const [confirmFlush, setConfirmFlush] = useState(false)
+  const flush = useAsyncAction(async () => flushCache())
+
+  const hits = data?.hits ?? 0
+  const misses = data?.misses ?? 0
+  const lookups = hits + misses
+  const hitRate = data?.hit_rate ?? 0
+
+  const pie: PieSlice[] = useMemo(
+    () => [
+      { name: 'hits', value: hits, fill: HITS },
+      { name: 'misses', value: misses, fill: MISSES },
+    ],
+    [hits, misses],
+  )
+
+  const persisted: PersistedView = (data?.persisted ?? {}) as PersistedView
+  const persistErrors = persisted.errors ?? 0
+  const persistMessage = persisted.last_error ?? persisted.error ?? ''
+
+  const entries: CacheTopEntry[] = useMemo(() => top?.items ?? [], [top])
+
+  const onFlush = async (): Promise<void> => {
+    const res = await flush.run()
+    if (res === null) {
+      notify('Could not flush the cache', flush.error ?? 'the server rejected the request')
+      return
+    }
+    setConfirmFlush(false)
+    notify(
+      'Cache flushed',
+      `${formatNumber(res.entries ?? 0)} ${res.entries === 1 ? 'entry' : 'entries'} dropped`,
+      'ok',
+    )
+    refetch()
+    refetchTop()
+  }
+
+  if (error) return <PageError error={error} onRetry={refetch} />
+
+  const topColumns = [
+    { key: 'name', header: 'Name', width: '48%' },
+    { key: 'qtype', header: 'Type', width: '10%' },
+    { key: 'count', header: 'Lookups', width: '14%', align: 'right' as const },
+    { key: 'hits', header: 'Served from cache', width: '14%', align: 'right' as const },
+    { key: 'last_seen', header: 'Last seen', width: '14%', align: 'right' as const },
+  ]
+
+  const topRows: Row[] = entries.map((entry) => ({
+    __key: `${entry.name}-${entry.qtype}`,
+    name: (
+      <span className="block max-w-[420px] truncate font-mono text-xs" title={entry.name}>
+        {entry.name}
+      </span>
+    ),
+    qtype: <span className="font-mono text-xs">{entry.qtype || '—'}</span>,
+    count: <span className="font-mono text-xs">{formatNumber(entry.count)}</span>,
+    hits: <span className="font-mono text-xs">{formatNumber(entry.hits)}</span>,
+    last_seen: <span className="text-xs text-fg-muted">{formatRelative(entry.last_seen)}</span>,
+  }))
+
+  return (
+    <div className="space-y-4">
+      <PageHeader
+        title="Cache"
+        description="Answer reuse, its lifetime, and what survives a restart."
+        actions={
+          <Button
+            variant="danger"
+            icon={RefreshCw}
+            onClick={() => setConfirmFlush(true)}
+            loading={flush.pending}
+          >
+            Flush cache
+          </Button>
+        }
+      />
+
+      <div className="grid grid-cols-2 gap-3 lg:grid-cols-3 xl:grid-cols-6">
+        <StatCard
+          label="Entries"
+          value={formatNumber(data?.entries ?? 0)}
+          icon={Database}
+          hint={data?.enabled === false ? 'cache disabled' : undefined}
+        loading={loading && data === null}
+        />
+        <StatCard
+          label="Size"
+          value={formatBytes(data?.size ?? 0)}
+          icon={HardDrive}
+          hint={`${formatNumber(lookups)} lookups`}
+        />
+        <StatCard
+          label="Shards"
+          value={formatNumber(data?.shards ?? 0)}
+          icon={Layers}
+        />
+        <StatCard
+          label="Max TTL"
+          value={formatMs(data?.max_ttl_ms ?? 0)}
+          icon={Timer}
+          hint={`min ${formatMs(data?.min_ttl_ms ?? 0)}`}
+        />
+        <StatCard
+          label="Serve stale"
+          value={formatMs(data?.serve_stale_ms ?? 0)}
+          icon={FileClock}
+          hint={data?.serve_stale_ms ? 'upstream timeouts tolerated' : 'never serve stale'}
+          tone={data?.serve_stale_ms ? 'default' : 'warn'}
+        />
+        <StatCard
+          label="Evictions"
+          value={formatNumber(data?.evictions ?? 0)}
+          icon={Gauge}
+          hint={`${formatNumber(data?.stale_hits ?? 0)} stale hits`}
+        />
+      </div>
+
+      <div className="grid grid-cols-1 gap-3 xl:grid-cols-2">
+        <Card
+          title="Hit rate"
+          subtitle={`${formatRate(hitRate)} of ${formatNumber(lookups)} lookups served from memory`}
+        >
+          <CardBody>
+            <div className="flex flex-col items-center gap-3 sm:flex-row">
+              <div className="relative h-48 w-48 shrink-0">
+                <ResponsiveContainer width="100%" height="100%">
+                  <PieChart>
+                    <Pie
+                      data={pie}
+                      dataKey="value"
+                      nameKey="name"
+                      innerRadius="62%"
+                      outerRadius="92%"
+                      paddingAngle={1}
+                      stroke="none"
+                      isAnimationActive={false}
+                    >
+                      {pie.map((slice) => (
+                        <Cell key={slice.name} fill={slice.fill} />
+                      ))}
+                    </Pie>
+                    <ChartTooltip content={<ChartTip />} />
+                  </PieChart>
+                </ResponsiveContainer>
+                <div className="pointer-events-none absolute inset-0 grid place-items-center">
+                  <div className="text-center">
+                    <p className="tabular text-xl font-semibold text-fg">{formatRate(hitRate, 0)}</p>
+                    <p className="text-xs text-fg-muted">hit rate</p>
+                  </div>
+                </div>
+              </div>
+              <ul className="w-full min-w-0 space-y-2 text-sm">
+                <li className="flex items-center gap-2">
+                  <span
+                    aria-hidden="true"
+                    className="size-2.5 shrink-0 rounded-full"
+                    style={{ background: HITS }}
+                  />
+                  <span className="text-fg-muted">Hits</span>
+                  <span className="tabular ml-auto font-mono text-xs">
+                    {formatNumber(hits)}
+                  </span>
+                </li>
+                <li className="flex items-center gap-2">
+                  <span
+                    aria-hidden="true"
+                    className="size-2.5 shrink-0 rounded-full"
+                    style={{ background: MISSES }}
+                  />
+                  <span className="text-fg-muted">Misses</span>
+                  <span className="tabular ml-auto font-mono text-xs">
+                    {formatNumber(misses)}
+                  </span>
+                </li>
+                <li className="flex items-center gap-2 border-t border-line pt-2 text-xs text-fg-muted">
+                  <span>Negative TTL</span>
+                  <span className="tabular ml-auto font-mono">
+                    {formatMs(data?.negative_ttl_ms ?? 0)}
+                  </span>
+                </li>
+                {lookups === 0 && (
+                  <li className="text-xs text-fg-subtle">
+                    Nothing has been looked up since the daemon started.
+                  </li>
+                )}
+              </ul>
+            </div>
+          </CardBody>
+        </Card>
+
+        <Card
+          title="Persisted file"
+          subtitle={persisted.enabled ? 'Snapshots survive a restart' : 'Snapshots are turned off'}
+          actions={
+            <Badge tone={persisted.enabled ? 'ok' : 'muted'}>
+              {persisted.enabled ? 'enabled' : 'disabled'}
+            </Badge>
+          }
+        >
+          <CardBody>
+            <dl className="grid grid-cols-2 gap-x-4 gap-y-3 sm:grid-cols-3">
+              <div className="col-span-2 min-w-0 sm:col-span-3">
+                <dt className="text-xs tracking-wide text-fg-muted uppercase">Path</dt>
+                <dd className="mt-0.5 truncate font-mono text-xs" title={persisted.path}>
+                  {persisted.path || '—'}
+                </dd>
+              </div>
+              <div>
+                <dt className="text-xs tracking-wide text-fg-muted uppercase">Size</dt>
+                <dd className="tabular mt-0.5 font-mono text-sm">
+                  {formatBytes(persisted.bytes ?? 0)}
+                </dd>
+              </div>
+              <div>
+                <dt className="text-xs tracking-wide text-fg-muted uppercase">Entries</dt>
+                <dd className="tabular mt-0.5 font-mono text-sm">
+                  {formatNumber(persisted.entries ?? 0)}
+                </dd>
+              </div>
+              <div>
+                <dt className="text-xs tracking-wide text-fg-muted uppercase">Writes</dt>
+                <dd className="tabular mt-0.5 font-mono text-sm">
+                  {formatNumber(persisted.writes ?? 0)}
+                </dd>
+              </div>
+              <div>
+                <dt className="text-xs tracking-wide text-fg-muted uppercase">Errors</dt>
+                <dd
+                  className={
+                    persistErrors > 0
+                      ? 'tabular mt-0.5 font-mono text-sm text-danger'
+                      : 'tabular mt-0.5 font-mono text-sm'
+                  }
+                >
+                  {formatNumber(persistErrors)}
+                </dd>
+              </div>
+              <div className="col-span-2 sm:col-span-1">
+                <dt className="text-xs tracking-wide text-fg-muted uppercase">Last snapshot</dt>
+                <dd className="mt-0.5 text-sm">{formatRelative(persisted.last_snapshot)}</dd>
+              </div>
+            </dl>
+            {persistMessage !== '' && (
+              <p className="mt-3 rounded-lg border border-danger/30 bg-danger-soft px-3 py-2 text-xs text-danger">
+                {persistMessage}
+              </p>
+            )}
+          </CardBody>
+        </Card>
+      </div>
+
+      <Card
+        title="Most requested"
+        subtitle={`Top ${TOP_LIMIT} names by lookups`}
+        padded={false}
+        actions={<Badge tone="muted">auto-refresh every 15s</Badge>}
+      >
+        {topError !== null ? (
+          <CardBody>
+            <p className="text-sm text-danger">{topError}</p>
+          </CardBody>
+        ) : (
+          <div className="overflow-x-auto">
+            <Table
+              columns={topColumns}
+              rows={topRows}
+              rowKey={(row) => String(row.__key)}
+              loading={topLoading && entries.length === 0}
+              empty={
+                <EmptyState
+                  icon={Database}
+                  title="The cache is empty"
+                  description="Nothing has been cached yet, so there is no ranking to show."
+                />
+              }
+            />
+          </div>
+        )}
+      </Card>
+
+      <ConfirmDialog
+        open={confirmFlush}
+        onClose={() => setConfirmFlush(false)}
+        onConfirm={() => void onFlush()}
+        title="Flush the cache?"
+        message="Every cached answer is dropped. The next query for each name goes back to the upstreams, so expect a burst of upstream traffic."
+        confirmLabel="Flush cache"
+        destructive
+        pending={flush.pending}
+      />
+    </div>
+  )
+}

web/src/pages/Clients.tsx

@@ -10,10 +10,10 @@
 import { useMemo, useState } from 'react'
 import { Copy, Pencil, Plus, Trash2, Users } from 'lucide-react'
 
-import { ApiError, getClients, putClients } from '../../lib/api'
+import { ApiError, getClients, putClients } from '../lib/api'
-import type { ClientGroup, ClientMatch } from '../../lib/types'
+import type { ClientGroup, ClientMatch } from '../lib/types'
-import { useApi, useAsyncAction } from '../../hooks/useApi'
+import { useApi, useAsyncAction } from '../hooks/useApi'
-import { fromLines, toLines } from '../../lib/format'
+import { fromLines, toLines } from '../lib/format'
 import {
   Badge,
   Button,
@@ -28,19 +28,19 @@ import {
   Textarea,
   useToast,
 } from '../components/ui'
+import type { ToastOptions } from '../components/ui'
 import { PageError, PageHeader } from '../components/layout'
 
-type ToastLike = {
+/**
-  success?: (message: string) => void
+ * One line for every mutation on this page: the message plus the tone it
-  error?: (message: string) => void
+ * earned. `fail` is the danger tone, `ok` the success one.
-  toast?: (message: string) => void
+ */
-  push?: (message: string) => void
+function notify(
-}
+  toast: (options: ToastOptions) => void,
-
+  kind: 'ok' | 'fail',
-function notify(toast: unknown, kind: 'ok' | 'fail', message: string): void {
+  message: string,
-  const t = toast as ToastLike
+): void {
-  const fn = kind === 'ok' ? (t.success ?? t.toast ?? t.push) : (t.error ?? t.toast ?? t.push)
+  toast({ title: message, tone: kind === 'ok' ? 'ok' : 'danger' })
-  fn?.call(t, message)
 }
 
 const POLICIES = [
@@ -173,11 +173,11 @@ function fieldLeaf(path: string): string {
 }
 
 export default function Clients() {
-  const toast = useToast()
+  const { toast } = useToast()
   const { data, error, loading, refetch } = useApi(() => getClients(), [])
 
   const groups = useMemo<ClientGroup[]>(() => data?.groups ?? [], [data])
-  const [defaultPolicy, setDefaultPolicy] = useState<string>('')
+  const [defaultPolicy] = useState<string>('')
   const [defaultDraft, setDefaultDraft] = useState<string | null>(null)
   const [editorOpen, setEditorOpen] = useState(false)
   const [draft, setDraft] = useState<GroupDraft>(EMPTY_DRAFT)
@@ -302,7 +302,7 @@ export default function Clients() {
         title="Clients"
         description="Group clients by network, reverse-DNS name or tag and give each group its own policy."
         actions={
-          <Button icon={<Plus size={14} />} onClick={openNew}>
+          <Button icon={Plus} onClick={openNew}>
             Add group
           </Button>
         }
@@ -322,11 +322,11 @@ export default function Clients() {
 
       {groups.length === 0 ? (
         <EmptyState
-          icon={<Users size={20} />}
+          icon={Users}
           title="No client groups"
           description="Without a group every client gets the default policy."
           action={
-            <Button icon={<Plus size={14} />} onClick={openNew}>
+            <Button icon={Plus} onClick={openNew}>
               Add the first group
             </Button>
           }
@@ -362,10 +362,10 @@ export default function Clients() {
                   </div>
                 </div>
                 <div className="flex shrink-0 gap-1">
-                  <IconButton icon={<Pencil size={14} />} label={`edit ${group.name}`} variant="ghost" onClick={() => openEdit(i)} />
+                  <IconButton icon={Pencil} label={`edit ${group.name}`} variant="ghost" onClick={() => openEdit(i)} />
-                  <IconButton icon={<Copy size={14} />} label={`duplicate ${group.name}`} variant="ghost" onClick={() => duplicate(i)} />
+                  <IconButton icon={Copy} label={`duplicate ${group.name}`} variant="ghost" onClick={() => duplicate(i)} />
                   <IconButton
-                    icon={<Trash2 size={14} />}
+                    icon={Trash2}
                     label={`delete ${group.name}`}
                     variant="ghost"
                     onClick={() => setPendingDelete(i)}

web/src/pages/Dashboard.tsx

@@ -23,11 +23,11 @@ import {
 } from 'recharts'
 import { Activity, Ban, Gauge, RefreshCw, Timer } from 'lucide-react'
 
-import { getDashboard } from '../../lib/api'
+import { getDashboard } from '../lib/api'
-import type { Dashboard as DashboardData, LogItem, UpstreamStat } from '../../lib/types'
+import type { Dashboard as DashboardData, LogItem, UpstreamStat } from '../lib/types'
-import { useApi } from '../../hooks/useApi'
+import { useApi } from '../hooks/useApi'
-import { useAutoRefresh } from '../../hooks/useAutoRefresh'
+import { useAutoRefresh } from '../hooks/useAutoRefresh'
-import { TIME_RANGES, useTimeRange } from '../../hooks/useTimeRange'
+import { TIME_RANGES, useTimeRange } from '../hooks/useTimeRange'
 import {
   formatClock,
   formatCompact,
@@ -36,7 +36,7 @@ import {
   formatPercent,
   formatRate,
   formatUptime,
-} from '../../lib/format'
+} from '../lib/format'
 import { Badge, Button, Card, CardBody, EmptyState, Progress, StatCard, Table } from '../components/ui'
 import { HealthPill, PageError, PageHeader } from '../components/layout'
 
@@ -47,7 +47,6 @@ const QUERIES = '#38bdf8'
 const CACHED = '#34d399'
 const BLOCKED = '#f59e0b'
 const ERRORS = '#f87171'
-const TIP_BG = '#0f172a'
 
 /** The parts of the dashboard payload that older daemons may not send. */
 type DashboardView = Partial<DashboardData> & {
@@ -64,20 +63,6 @@ function view(data: DashboardData | null): DashboardView {
   return (data ?? {}) as DashboardView
 }
 
-/** `useToast()` shape is the UI kit's business; this just needs a call. */
-type ToastLike = {
-  success?: (message: string) => void
-  error?: (message: string) => void
-  toast?: (message: string) => void
-  push?: (message: string) => void
-}
-
-function notify(toast: unknown, kind: 'ok' | 'fail', message: string): void {
-  const t = toast as ToastLike
-  const fn = kind === 'ok' ? (t.success ?? t.toast ?? t.push) : (t.error ?? t.toast ?? t.push)
-  fn?.call(t, message)
-}
-
 type TipProps = {
   active?: boolean
   label?: number | string
@@ -222,7 +207,7 @@ export default function Dashboard() {
             <Button
               variant="outline"
               size="sm"
-              icon={<RefreshCw size={14} />}
+              icon={RefreshCw}
               onClick={refetch}
               loading={loading}
             >
@@ -251,26 +236,26 @@ export default function Dashboard() {
           label="Queries / s"
           value={formatCompact(d.queries_per_sec ?? 0)}
           hint={`${formatNumber(d.queries_total ?? 0)} total`}
-          icon={<Activity size={16} />}
+          icon={Activity}
         />
         <StatCard
           label="Cache hit rate"
           value={formatRate(hitRate)}
           hint="higher is better"
-          icon={<Gauge size={16} />}
+          icon={Gauge}
           tone={hitRate >= 0.7 ? 'ok' : hitRate >= 0.4 ? 'warn' : 'danger'}
         />
         <StatCard
           label="Blocked"
           value={formatPercent(d.blocked_pct ?? 0)}
           hint={`${formatNumber(d.blocked_total ?? 0)} queries`}
-          icon={<Ban size={16} />}
+          icon={Ban}
         />
         <StatCard
           label={latencyLabel}
           value={latencyValue != null ? formatMs(latencyValue) : '—'}
           hint={p99 != null ? 'from the server' : 'slowest upstream p95'}
-          icon={<Timer size={16} />}
+          icon={Timer}
         />
       </div>
 

web/src/pages/Domains.tsx

@@ -10,15 +10,16 @@ import { useEffect, useMemo, useState } from 'react'
 import type { ReactNode } from 'react'
 import { Plus, Search, Trash2 } from 'lucide-react'
 
-import { addDomain, deleteDomain, getDomains } from '../../lib/api'
+import { addDomain, deleteDomain, getDomains } from '../lib/api'
-import type { AddDomainInput } from '../../lib/api'
+import type { AddDomainInput } from '../lib/api'
-import type { DomainItem } from '../../lib/types'
+import type { DomainItem } from '../lib/types'
-import { useApi, useAsyncAction } from '../../hooks/useApi'
+import { useApi, useAsyncAction } from '../hooks/useApi'
-import { useDebounced } from '../../hooks/useDebounced'
+import { useDebounced } from '../hooks/useDebounced'
-import { formatNumber, formatRelative, toLines } from '../../lib/format'
+import { formatNumber, formatRelative, toLines } from '../lib/format'
 import {
   Badge,
   Button,
+  Card,
   ConfirmDialog,
   Dialog,
   EmptyState,
@@ -30,21 +31,21 @@ import {
   Textarea,
   useToast,
 } from '../components/ui'
+import type { ToastOptions } from '../components/ui'
 import { PageError, PageHeader } from '../components/layout'
 
 type Row = Record<string, ReactNode>
 
-type ToastLike = {
+/**
-  success?: (message: string) => void
+ * One line for every mutation on this page: the message plus the tone it
-  error?: (message: string) => void
+ * earned. `fail` is the danger tone, `ok` the success one.
-  toast?: (message: string) => void
+ */
-  push?: (message: string) => void
+function notify(
-}
+  toast: (options: ToastOptions) => void,
-
+  kind: 'ok' | 'fail',
-function notify(toast: unknown, kind: 'ok' | 'fail', message: string): void {
+  message: string,
-  const t = toast as ToastLike
+): void {
-  const fn = kind === 'ok' ? (t.success ?? t.toast ?? t.push) : (t.error ?? t.toast ?? t.push)
+  toast({ title: message, tone: kind === 'ok' ? 'ok' : 'danger' })
-  fn?.call(t, message)
 }
 
 const PER_PAGE = 50
@@ -88,7 +89,7 @@ type AddForm = {
 const EMPTY_FORM: AddForm = { name: '', action: 'block', list: '', bulk: '', advanced: false }
 
 export default function Domains() {
-  const toast = useToast()
+  const { toast } = useToast()
   const [search, setSearch] = useState('')
   const [action, setAction] = useState('')
   const [list, setList] = useState('')
@@ -220,7 +221,7 @@ export default function Domains() {
     updated: <span className="whitespace-nowrap text-xs">{formatRelative(item.updated_at)}</span>,
     delete: (
       <IconButton
-        icon={<Trash2 size={14} />}
+        icon={Trash2}
         label={`delete ${item.name}`}
         variant="ghost"
         onClick={() => setPendingDelete(item)}
@@ -234,7 +235,7 @@ export default function Domains() {
         title="Domains"
         description="Inline allow and block rules, checked before the downloaded lists."
         actions={
-          <Button icon={<Plus size={14} />} onClick={() => setAddOpen(true)}>
+          <Button icon={Plus} onClick={() => setAddOpen(true)}>
             Add rule
           </Button>
         }
@@ -245,7 +246,7 @@ export default function Domains() {
           <Input
             label="Search"
             placeholder="ads.example.com"
-            icon={<Search size={14} />}
+            icon={Search}
             value={search}
             onChange={(e) => setSearch(e.target.value)}
             autoComplete="off"

web/src/pages/Login.tsx

@@ -0,0 +1,182 @@
+/**
+ * Login — the only page that renders outside AppShell.
+ *
+ * The daemon may be running with `auth.mode: none`, `apikey` or `basic`, so the
+ * form is built from what `/api/whoami` reports rather than hard-coded. A 401
+ * from that probe is expected (that is why the visitor is here) and is not
+ * shown as an error.
+ */
+
+import { useEffect, useState } from 'react'
+import type { FormEvent } from 'react'
+import { Globe, KeyRound, LogIn, ShieldCheck, Terminal } from 'lucide-react'
+import { useSearchParams } from 'react-router-dom'
+
+import { ApiError, login, whoami } from '../lib/api'
+import type { WhoamiResponse } from '../lib/api'
+import { Button, Card, CardBody, Input } from '../components/ui'
+import { Spinner } from '../components/ui'
+
+/** "apikey" -> one field, "basic" -> two, anything else -> one field. */
+function wantsPassword(mode: string): boolean {
+  return mode === 'basic'
+}
+
+export default function Login() {
+  const [params] = useSearchParams()
+  const next = params.get('next') ?? '/'
+
+  const [mode, setMode] = useState<string | null>(null)
+  const [apiKey, setApiKey] = useState('')
+  const [username, setUsername] = useState('')
+  const [password, setPassword] = useState('')
+  const [pending, setPending] = useState(false)
+  const [error, setError] = useState<string | null>(null)
+
+  useEffect(() => {
+    let live = true
+    whoami()
+      .then((who: WhoamiResponse) => {
+        if (live) setMode(who.auth_mode ?? 'apikey')
+      })
+      .catch(() => {
+        // Unauthenticated probes and older daemons both land here: show the
+        // form rather than an error, the submit will say what is wrong.
+        if (live) setMode('apikey')
+      })
+    return () => {
+      live = false
+    }
+  }, [])
+
+  const onSubmit = async (event: FormEvent<HTMLFormElement>): Promise<void> => {
+    event.preventDefault()
+    if (pending) return
+    setError(null)
+    setPending(true)
+    try {
+      if (wantsPassword(mode ?? 'apikey')) {
+        await login({ username, password })
+      } else {
+        await login({ api_key: apiKey })
+      }
+      // A full navigation: the shell reads whoami and the token on boot.
+      window.location.assign(next.startsWith('/') ? next : '/')
+    } catch (err) {
+      setError(
+        err instanceof ApiError
+          ? err.message
+          : err instanceof Error
+            ? err.message
+            : 'login failed',
+      )
+    } finally {
+      setPending(false)
+    }
+  }
+
+  const passwordMode = wantsPassword(mode ?? 'apikey')
+
+  return (
+    <div className="flex min-h-svh items-center justify-center bg-canvas px-4 py-10">
+      <div className="w-full max-w-sm">
+        <div className="mb-5 flex items-center justify-center gap-2">
+          <span className="grid size-8 place-items-center rounded-lg bg-accent-soft text-accent">
+            <Globe size={17} strokeWidth={2.25} aria-hidden="true" />
+          </span>
+          <span className="text-base font-semibold tracking-tight text-fg">fancy-dns</span>
+        </div>
+
+        <Card
+          title="Sign in"
+          subtitle={
+            mode === null
+              ? 'Checking the authentication mode…'
+              : passwordMode
+                ? 'This node uses username and password.'
+                : 'This node uses an API key.'
+          }
+        >
+          <CardBody>
+            <form className="space-y-3" onSubmit={(event) => void onSubmit(event)}>
+              {passwordMode ? (
+                <>
+                  <Input
+                    label="Username"
+                    icon={ShieldCheck}
+                    value={username}
+                    onChange={(event) => setUsername(event.target.value)}
+                    autoComplete="username"
+                    autoFocus
+                    required
+                  />
+                  <Input
+                    label="Password"
+                    type="password"
+                    icon={KeyRound}
+                    value={password}
+                    onChange={(event) => setPassword(event.target.value)}
+                    autoComplete="current-password"
+                    required
+                  />
+                </>
+              ) : (
+                <Input
+                  label="API key"
+                  type="password"
+                  icon={KeyRound}
+                  placeholder="paste the node's API key"
+                  value={apiKey}
+                  onChange={(event) => setApiKey(event.target.value)}
+                  autoComplete="current-password"
+                  spellCheck={false}
+                  autoFocus
+                  required
+                />
+              )}
+
+              {error !== null && (
+                <p
+                  role="alert"
+                  className="rounded-lg border border-danger/40 bg-danger-soft px-3 py-2 text-xs text-danger"
+                >
+                  {error}
+                </p>
+              )}
+
+              <Button
+                type="submit"
+                variant="primary"
+                size="lg"
+                icon={pending ? undefined : LogIn}
+                loading={pending}
+                className="w-full justify-center"
+              >
+                {pending ? 'Signing in…' : 'Sign in'}
+              </Button>
+            </form>
+          </CardBody>
+        </Card>
+
+        <p className="mt-4 flex items-start gap-2 text-xs text-fg-subtle">
+          <Terminal size={13} className="mt-0.5 shrink-0" aria-hidden="true" />
+          <span>
+            Scripts can skip this page: send the same credential as an{' '}
+            <span className="font-mono text-fg-muted">X-API-Key</span> header, for example{' '}
+            <span className="font-mono text-fg-muted">
+              curl -H &quot;X-API-Key: $KEY&quot; http://node:8080/api/dashboard
+            </span>
+            .
+          </span>
+        </p>
+
+        {pending && (
+          <p className="mt-3 flex items-center justify-center gap-2 text-xs text-fg-muted">
+            <Spinner size="sm" />
+            waiting for the node
+          </p>
+        )}
+      </div>
+    </div>
+  )
+}

web/src/pages/Logs.tsx

@@ -0,0 +1,431 @@
+/**
+ * Logs — the query log, and the live tail of it.
+ *
+ * One table, two sources: the paged `/api/logs` result and the SSE stream from
+ * `/api/logs/stream`. While LIVE is on, streamed entries are prepended and run
+ * through the same filters client-side, so a live row can never contradict what
+ * the filter bar says. Rendering is capped: a busy node emits far faster than a
+ * browser can lay out rows, and an uncapped list is the tab's worst enemy.
+ */
+
+import { useEffect, useMemo, useRef, useState } from 'react'
+import type { ReactNode } from 'react'
+import { Eraser, Radio, Search, Signal, WifiOff } from 'lucide-react'
+
+import { clearLogs, getLogs, streamLogs } from '../lib/api'
+import type { LogFilters, LogItem, LiveStats } from '../lib/types'
+import { useApi, useAsyncAction } from '../hooks/useApi'
+import { useAutoRefresh } from '../hooks/useAutoRefresh'
+import { useDebounced } from '../hooks/useDebounced'
+import {
+  formatClock,
+  formatCompact,
+  formatMs,
+  formatNumber,
+  formatPercent,
+  formatRate,
+  formatRcode,
+} from '../lib/format'
+import { cn } from '../lib/cn'
+import {
+  Badge,
+  Button,
+  Card,
+  CardBody,
+  ConfirmDialog,
+  Input,
+  Pagination,
+  Select,
+  Switch,
+  Table,
+  useToast,
+} from '../components/ui'
+import { PageError, PageHeader } from '../components/layout'
+
+type Row = Record<string, ReactNode>
+
+const PER_PAGE = 50
+/** Hard cap on rendered rows, live tail included. */
+const MAX_ROWS = 300
+const REFRESH_LIVE_MS = 3000
+
+const ACTION_OPTIONS = [
+  { value: '', label: 'All actions' },
+  { value: 'allow', label: 'Allow' },
+  { value: 'cached', label: 'Cached' },
+  { value: 'block', label: 'Block' },
+  { value: 'error', label: 'Error' },
+  { value: 'drop', label: 'Drop' },
+]
+
+/** The filter set, reduced to what both the server and the live tail need. */
+interface Filter {
+  q: string
+  action: string
+  client: string
+  blocked: boolean
+}
+
+function actionTone(action: string): 'default' | 'ok' | 'warn' | 'danger' | 'info' | 'muted' {
+  switch (action) {
+    case 'allow':
+      return 'default'
+    case 'cached':
+      return 'info'
+    case 'block':
+      return 'warn'
+    case 'error':
+      return 'danger'
+    case 'drop':
+      return 'muted'
+    default:
+      return 'default'
+  }
+}
+
+/** The server filters; the live tail re-implements the same rules. */
+function matches(entry: LogItem, filter: Filter): boolean {
+  if (filter.action !== '' && entry.action !== filter.action) return false
+  if (filter.blocked && entry.action !== 'block') return false
+
+  if (filter.client !== '') {
+    const needle = filter.client.toLowerCase()
+    const hay = `${entry.client} ${entry.client_group}`.toLowerCase()
+    if (!hay.includes(needle)) return false
+  }
+
+  if (filter.q !== '') {
+    const needle = filter.q.toLowerCase()
+    const hay = [
+      entry.name,
+      entry.list,
+      entry.upstream,
+      entry.client,
+      entry.client_group,
+      entry.action,
+    ]
+      .join(' ')
+      .toLowerCase()
+    if (!hay.includes(needle)) return false
+  }
+
+  return true
+}
+
+/** Older daemons omit stats fields; a missing counter reads as zero, not NaN. */
+function liveStats(stats: LiveStats): LiveStats {
+  return {
+    qps: Number(stats.qps) || 0,
+    blocked_pct: Number(stats.blocked_pct) || 0,
+    cache_hit_rate: Number(stats.cache_hit_rate) || 0,
+  }
+}
+
+export default function Logs() {
+  const { toast } = useToast()
+  const notify = (title: string, description?: string, tone: 'ok' | 'fail' = 'fail'): void =>
+    toast({ title, description, tone: tone === 'ok' ? 'ok' : 'danger' })
+
+  const [search, setSearch] = useState('')
+  const [action, setAction] = useState('')
+  const [client, setClient] = useState('')
+  const [blockedOnly, setBlockedOnly] = useState(false)
+  const [page, setPage] = useState(1)
+  const q = useDebounced(search, 300)
+
+  const [live, setLive] = useState(false)
+  const [connected, setConnected] = useState(false)
+  const [streamError, setStreamError] = useState<string | null>(null)
+  const [liveRows, setLiveRows] = useState<LogItem[]>([])
+  const [stats, setStats] = useState<LiveStats | null>(null)
+  const [confirmClear, setConfirmClear] = useState(false)
+
+  const clear = useAsyncAction(async () => clearLogs())
+
+  // Every filter change invalidates both the current page and the live tail.
+  const applyFilter = (): void => {
+    setPage(1)
+    setLiveRows([])
+  }
+
+  const filters = useMemo<LogFilters>(
+    () => ({
+      q: q.trim() || undefined,
+      action: action === '' ? undefined : action,
+      client: client.trim() || undefined,
+      blocked: blockedOnly ? true : undefined,
+      limit: PER_PAGE,
+      offset: (page - 1) * PER_PAGE,
+    }),
+    [q, action, client, blockedOnly, page],
+  )
+
+  const { data, error, loading, refetch } = useApi(() => getLogs(filters), [filters])
+  useAutoRefresh(refetch, REFRESH_LIVE_MS, live)
+
+  // The stream outlives every filter change, so it reads the current filter
+  // through a ref instead of resubscribing on each keystroke.
+  const filterRef = useRef<Filter>({ q: '', action: '', client: '', blocked: false })
+  useEffect(() => {
+    filterRef.current = { q: q.trim(), action, client: client.trim(), blocked: blockedOnly }
+  }, [q, action, client, blockedOnly])
+
+  const toggleLive = (next: boolean): void => {
+    setLiveRows([])
+    setStats(null)
+    setConnected(false)
+    setStreamError(null)
+    setLive(next)
+  }
+
+  useEffect(() => {
+    if (!live) return
+    const stop = streamLogs({
+      onEntry: (entry) => {
+        if (!matches(entry, filterRef.current)) return
+        setLiveRows((prev) => [entry, ...prev].slice(0, MAX_ROWS))
+      },
+      onStats: (next) => setStats(liveStats(next)),
+      onReady: () => setConnected(true),
+      onOpen: () => {
+        setConnected(true)
+        setStreamError(null)
+      },
+      onError: (message) => {
+        setConnected(false)
+        setStreamError(message)
+      },
+    })
+    return stop
+  }, [live])
+
+  const items: LogItem[] = useMemo(() => data?.items ?? [], [data])
+  const total = data?.total ?? items.length
+  const rows = useMemo<LogItem[]>(
+    () => (liveRows.length > 0 ? [...liveRows, ...items] : items).slice(0, MAX_ROWS),
+    [liveRows, items],
+  )
+
+  const onClear = async (): Promise<void> => {
+    const res = await clear.run()
+    if (res === null) {
+      notify('Could not clear the log', clear.error ?? 'the server rejected the request')
+      return
+    }
+    setConfirmClear(false)
+    setLiveRows([])
+    notify('Log cleared', 'the server dropped every buffered entry', 'ok')
+    refetch()
+  }
+
+  if (error) return <PageError error={error} onRetry={refetch} />
+
+  const columns = [
+    { key: 'ts', header: 'Time', width: '9%' },
+    { key: 'client', header: 'Client', width: '13%' },
+    { key: 'group', header: 'Group', width: '10%' },
+    { key: 'name', header: 'Name', width: '24%' },
+    { key: 'qtype', header: 'Type', width: '6%' },
+    { key: 'rcode', header: 'Rcode', width: '12%' },
+    { key: 'action', header: 'Action', width: '8%' },
+    { key: 'duration', header: 'Time', width: '8%', align: 'right' as const },
+    { key: 'upstream', header: 'Upstream', width: '10%' },
+    { key: 'list', header: 'List', width: '12%' },
+  ]
+
+  const tableRows: Row[] = rows.map((entry, index) => ({
+    __key: `${entry.id}-${entry.ts}-${index}`,
+    ts: <span className="tabular font-mono text-xs text-fg-muted">{formatClock(entry.ts)}</span>,
+    client: (
+      <span className="block max-w-[180px] truncate font-mono text-xs">
+        {entry.client || '—'}
+      </span>
+    ),
+    group: entry.client_group ? (
+      <Badge tone="muted">{entry.client_group}</Badge>
+    ) : (
+      '—'
+    ),
+    name: (
+      <span className="block max-w-[320px] truncate font-mono text-xs" title={entry.name}>
+        {entry.name}
+      </span>
+    ),
+    qtype: <span className="font-mono text-xs">{entry.qtype || '—'}</span>,
+    rcode: <span className="font-mono text-xs text-fg-muted">{formatRcode(entry.rcode)}</span>,
+    action: <Badge tone={actionTone(entry.action)}>{entry.action || 'unknown'}</Badge>,
+    duration: <span className="font-mono text-xs">{formatMs(entry.duration_ms)}</span>,
+    upstream: (
+      <span className="block max-w-[160px] truncate font-mono text-xs text-fg-muted">
+        {entry.upstream || '—'}
+      </span>
+    ),
+    list: (
+      <span className="block max-w-[200px] truncate text-xs text-fg-muted" title={entry.list}>
+        {entry.list || '—'}
+      </span>
+    ),
+  }))
+
+  const qps = stats?.qps ?? 0
+  const blockedPct = stats?.blocked_pct ?? 0
+  const hitRate = stats?.cache_hit_rate ?? 0
+
+  return (
+    <div className="space-y-4">
+      <PageHeader
+        title="Logs"
+        description="Every query the daemon answered, with the rule that decided it."
+        actions={
+          <div className="flex flex-wrap items-center gap-2">
+            <Badge
+              tone={live ? (connected ? 'ok' : 'danger') : 'muted'}
+              className="gap-1.5"
+            >
+              <span
+                aria-hidden="true"
+                className={cn(
+                  'size-1.5 rounded-full',
+                  live ? (connected ? 'animate-pulse-soft bg-ok' : 'animate-pulse-soft bg-danger') : 'bg-fg-subtle',
+                )}
+              />
+              {live ? (connected ? 'LIVE' : 'DISCONNECTED') : 'PAUSED'}
+            </Badge>
+            <Switch
+              checked={live}
+              onChange={toggleLive}
+              label="Live tail"
+              description="Stream new entries over SSE"
+            />
+            <Button
+              variant="outline"
+              icon={Eraser}
+              onClick={() => setConfirmClear(true)}
+              loading={clear.pending}
+            >
+              Clear
+            </Button>
+          </div>
+        }
+      />
+
+      {live && (
+        <div className="flex flex-wrap items-center gap-x-5 gap-y-1 rounded-lg border border-line bg-surface px-4 py-2.5 text-xs shadow-panel">
+          <span className="inline-flex items-center gap-1.5 text-fg-muted">
+            <Signal size={13} aria-hidden="true" />
+            <span className="tabular font-mono text-fg">{formatCompact(qps)}</span> qps
+          </span>
+          <span className="text-fg-muted">
+            <span className="tabular font-mono text-fg">{formatPercent(blockedPct)}</span>{' '}
+            blocked
+          </span>
+          <span className="text-fg-muted">
+            <span className="tabular font-mono text-fg">{formatRate(hitRate)}</span> cache hits
+          </span>
+          <span className="text-fg-subtle">sampled by the daemon every few seconds</span>
+          {streamError !== null && (
+            <span className="text-danger" title={streamError}>
+              <WifiOff size={13} className="mr-1 inline align-[-2px]" aria-hidden="true" />
+              {streamError}
+            </span>
+          )}
+        </div>
+      )}
+
+      <Card title="Filters" subtitle="Free text covers name, list, upstream and client.">
+        <CardBody>
+          <div className="grid grid-cols-1 gap-3 sm:grid-cols-2 xl:grid-cols-4">
+            <Input
+              label="Search"
+              icon={Search}
+              placeholder="example.com, ads, 10.0.0.0/8"
+              value={search}
+              onChange={(event) => {
+                setSearch(event.target.value)
+                applyFilter()
+              }}
+              autoComplete="off"
+              spellCheck={false}
+            />
+            <Select
+              label="Action"
+              options={ACTION_OPTIONS}
+              value={action}
+              onChange={(event) => {
+                setAction(event.target.value)
+                applyFilter()
+              }}
+            />
+            <Input
+              label="Client"
+              placeholder="address or group name"
+              value={client}
+              onChange={(event) => {
+                setClient(event.target.value)
+                applyFilter()
+              }}
+              autoComplete="off"
+              spellCheck={false}
+            />
+            <div className="flex items-end pb-1">
+              <Switch
+                checked={blockedOnly}
+                onChange={(next) => {
+                  setBlockedOnly(next)
+                  applyFilter()
+                }}
+                label="Blocked only"
+                description="Hide allowed and cached answers"
+              />
+            </div>
+          </div>
+        </CardBody>
+      </Card>
+
+      <Card
+        title="Query log"
+        subtitle={`${rows.length} of ${formatNumber(total)} matching entries`}
+        padded={false}
+        actions={
+          rows.length >= MAX_ROWS ? (
+            <Badge tone="warn">capped at {MAX_ROWS} rows</Badge>
+          ) : (
+            <Badge tone="muted">
+              <Radio size={11} aria-hidden="true" />
+              {live ? 'streaming' : 'paused'}
+            </Badge>
+          )
+        }
+      >
+        <div className="overflow-x-auto">
+          <Table
+            columns={columns}
+            rows={tableRows}
+            rowKey={(row) => String(row.__key)}
+            loading={loading && items.length === 0 && liveRows.length === 0}
+            maxHeight={640}
+            empty="No query matched these filters."
+          />
+        </div>
+      </Card>
+
+      <Pagination
+        page={page}
+        perPage={PER_PAGE}
+        total={total}
+        onPage={setPage}
+      />
+
+      <ConfirmDialog
+        open={confirmClear}
+        onClose={() => setConfirmClear(false)}
+        onConfirm={() => void onClear()}
+        title="Clear the query log?"
+        message="Every buffered entry is dropped on the server. This cannot be undone, and the running log keeps filling from the next query."
+        confirmLabel="Clear log"
+        destructive
+        pending={clear.pending}
+      />
+    </div>
+  )
+}

web/src/pages/NotFound.tsx

@@ -0,0 +1,53 @@
+/**
+ * NotFound — the catch-all route.
+ */
+
+import { Compass, LayoutDashboard } from 'lucide-react'
+import { Link } from 'react-router-dom'
+
+import { Button, Card, CardBody, EmptyState } from '../components/ui'
+
+export default function NotFound() {
+  return (
+    <div className="space-y-4">
+      <EmptyState
+        icon={Compass}
+        title="404 — that page does not exist"
+        description="The console has no route for this address. It may have been renamed, or the link may be stale."
+        action={
+          <Link to="/">
+            <Button variant="primary" icon={LayoutDashboard}>
+              Back to the dashboard
+            </Button>
+          </Link>
+        }
+      />
+      <Card title="Available pages" subtitle="The console ships these routes">
+        <CardBody>
+          <ul className="grid grid-cols-1 gap-1.5 text-sm sm:grid-cols-2 lg:grid-cols-4">
+            {[
+              ['/', 'Dashboard'],
+              ['/blocklists', 'Blocklists'],
+              ['/domains', 'Domains'],
+              ['/clients', 'Clients'],
+              ['/upstreams', 'Upstreams'],
+              ['/logs', 'Logs'],
+              ['/cache', 'Cache'],
+              ['/settings', 'Settings'],
+            ].map(([to, label]) => (
+              <li key={to}>
+                <Link
+                  to={to}
+                  className="flex items-baseline gap-2 rounded-lg px-2 py-1.5 text-fg-muted transition-colors hover:bg-surface-3 hover:text-fg"
+                >
+                  <span className="font-mono text-xs text-fg-subtle">{to}</span>
+                  <span className="truncate">{label}</span>
+                </Link>
+              </li>
+            ))}
+          </ul>
+        </CardBody>
+      </Card>
+    </div>
+  )
+}

web/src/pages/Settings.tsx

@@ -0,0 +1,471 @@
+/**
+ * Settings — the YAML that the daemon was started from.
+ *
+ * The editor is deliberately plain: a line-numbered textarea rather than a
+ * YAML widget, because the operator's file already has comments and ordering
+ * that a form would silently destroy. Validation is server-side, so what is
+ * reported here is exactly what the daemon would do with the file.
+ */
+
+import { useMemo, useRef, useState } from 'react'
+import type { UIEvent } from 'react'
+import {
+  Ban,
+  CheckCircle2,
+  Database,
+  FileCode2,
+  Info,
+  RefreshCw,
+  Save,
+  ShieldAlert,
+  TriangleAlert,
+} from 'lucide-react'
+
+import {
+  flushCache,
+  getSettings,
+  putSettings,
+  refreshBlocklists,
+  reloadSettings,
+  validateSettings,
+} from '../lib/api'
+import type { ConfigIssue } from '../lib/types'
+import { useApi, useAsyncAction } from '../hooks/useApi'
+import { formatNumber } from '../lib/format'
+import {
+  Badge,
+  Button,
+  Card,
+  CardBody,
+  ConfirmDialog,
+  CopyButton,
+  Spinner,
+  Textarea,
+  useToast,
+} from '../components/ui'
+import { LoadingBlock, PageError, PageHeader } from '../components/layout'
+
+interface Report {
+  errors: ConfigIssue[]
+  warnings: ConfigIssue[]
+}
+
+/** A validation list is only usable if every entry has a message. */
+function issues(list: ConfigIssue[] | undefined): ConfigIssue[] {
+  if (!Array.isArray(list)) return []
+  return list.filter((item) => item != null && typeof item.message === 'string' && item.message !== '')
+}
+
+function IssueList({ issues: list, tone }: { issues: ConfigIssue[]; tone: 'danger' | 'warn' }) {
+  return (
+    <ul className="space-y-1.5">
+      {list.map((issue, index) => (
+        <li key={`${issue.field}-${index}`} className="flex min-w-0 items-start gap-2 text-xs">
+          {tone === 'danger' ? (
+            <TriangleAlert size={13} className="mt-0.5 shrink-0 text-danger" aria-hidden="true" />
+          ) : (
+            <Info size={13} className="mt-0.5 shrink-0 text-warn" aria-hidden="true" />
+          )}
+          <span className="min-w-0">
+            {issue.field !== '' && (
+              <span className="mr-1.5 font-mono text-fg-subtle">{issue.field}</span>
+            )}
+            <span className={tone === 'danger' ? 'text-danger' : 'text-fg-muted'}>
+              {issue.message}
+            </span>
+          </span>
+        </li>
+      ))}
+    </ul>
+  )
+}
+
+export default function Settings() {
+  const { toast } = useToast()
+  const notify = (title: string, description?: string, tone: 'ok' | 'fail' = 'fail'): void =>
+    toast({ title, description, tone: tone === 'ok' ? 'ok' : 'danger' })
+
+  const { data, error, loading, refetch } = useApi(() => getSettings(), [])
+  // The draft is keyed by the version it was started from: once the server
+  // reports a different version the edit is dropped, so a refetch can never
+  // stomp on an editor that is mid-change.
+  const [draft, setDraft] = useState<{ version: string; text: string } | null>(null)
+  const gutter = useRef<HTMLDivElement | null>(null)
+  const [report, setReport] = useState<Report | null>(null)
+  const [confirmReload, setConfirmReload] = useState(false)
+
+  const validate = useAsyncAction(async (yaml: string) => validateSettings(yaml))
+  const save = useAsyncAction(async (yaml: string) => putSettings(yaml))
+  const reload = useAsyncAction(async () => reloadSettings())
+  const flush = useAsyncAction(async () => flushCache())
+  const refreshLists = useAsyncAction(async () => refreshBlocklists())
+
+  const pristine = data?.config_yaml ?? ''
+  const version = data?.version ?? ''
+  const value = draft !== null && draft.version === version ? draft.text : pristine
+  const dirty = data !== null && value !== pristine
+  const writable = data?.writable !== false
+  const lineCount = useMemo(() => value.split('\n').length, [value])
+  const gutterLines = useMemo(() => Array.from({ length: lineCount }, (_, i) => i + 1), [lineCount])
+
+  // The gutter is a plain div; keep its scroll offset glued to the textarea's.
+  const onScroll = (event: UIEvent<HTMLTextAreaElement>): void => {
+    const element = gutter.current
+    if (element !== null) element.scrollTop = event.currentTarget.scrollTop
+  }
+
+  const onValidate = async (): Promise<void> => {
+    const res = await validate.run(value)
+    if (res === null) {
+      setReport(null)
+      notify('Validation failed', validate.error ?? 'the server could not check this file')
+      return
+    }
+    const found: Report = { errors: issues(res.errors), warnings: issues(res.warnings) }
+    setReport(found)
+    if (found.errors.length === 0) {
+      notify(
+        'Configuration is valid',
+        found.warnings.length > 0
+          ? `${found.warnings.length} ${found.warnings.length === 1 ? 'warning' : 'warnings'}`
+          : 'the daemon would accept this file',
+        'ok',
+      )
+      return
+    }
+    notify(
+      'Configuration is not valid',
+      `${found.errors.length} ${found.errors.length === 1 ? 'error' : 'errors'}, ${found.warnings.length} warnings`,
+    )
+  }
+
+  const onSave = async (): Promise<void> => {
+    const res = await save.run(value)
+    if (res === null) {
+      notify('Could not save', save.error ?? 'the daemon rejected the write')
+      return
+    }
+    const list = Array.isArray(res.warnings) ? res.warnings : []
+    notify(
+      'Configuration saved',
+      list.length > 0
+        ? `${list.length} ${list.length === 1 ? 'warning' : 'warnings'}: ${list.join('; ')}`
+        : `version ${res.version || 'unknown'}`,
+      'ok',
+    )
+    setDraft(null)
+    setReport(null)
+    refetch()
+  }
+
+  const onReload = async (): Promise<void> => {
+    const res = await reload.run()
+    setConfirmReload(false)
+    if (res === null) {
+      notify('Reload failed', reload.error ?? 'the daemon kept the running configuration')
+      return
+    }
+    setDraft(null)
+    setReport(null)
+    notify('Configuration reloaded', `running version ${res.version || 'unknown'}`, 'ok')
+    refetch()
+  }
+
+  const onFlush = async (): Promise<void> => {
+    const res = await flush.run()
+    if (res === null) {
+      notify('Could not flush the cache', flush.error ?? 'the server rejected the request')
+      return
+    }
+    notify(
+      'Cache flushed',
+      `${formatNumber(res.entries ?? 0)} ${res.entries === 1 ? 'entry' : 'entries'} dropped`,
+      'ok',
+    )
+  }
+
+  const onRefreshLists = async (): Promise<void> => {
+    const res = await refreshLists.run()
+    if (res === null) {
+      notify('Refresh failed', refreshLists.error ?? 'the server rejected the request')
+      return
+    }
+    notify('Refresh started', `${res.names?.length ?? 0} sources queued`, 'ok')
+  }
+
+  if (error) return <PageError error={error} onRetry={refetch} />
+
+  if (loading && data === null) {
+    return (
+      <div className="space-y-4">
+        <PageHeader title="Settings" description="The configuration the daemon is running." />
+        <Card>
+          <CardBody>
+            <LoadingBlock rows={8} />
+          </CardBody>
+        </Card>
+      </div>
+    )
+  }
+
+  const effective = data?.effective_yaml ?? ''
+  const defaults = data?.defaults_yaml ?? ''
+
+  return (
+    <div className="space-y-4">
+      <PageHeader
+        title="Settings"
+        description="The configuration the daemon is running, and the file behind it."
+        actions={
+          <div className="flex flex-wrap items-center gap-2">
+            <Badge tone={writable ? 'ok' : 'warn'}>{writable ? 'writable' : 'read-only'}</Badge>
+            <Button
+              variant="outline"
+              icon={RefreshCw}
+              onClick={() => setConfirmReload(true)}
+              loading={reload.pending}
+              disabled={data?.reload_available === false}
+            >
+              Reload from disk
+            </Button>
+          </div>
+        }
+      />
+
+      <div className="flex flex-wrap items-center gap-x-4 gap-y-1 text-xs text-fg-muted">
+        <span className="inline-flex min-w-0 items-center gap-1.5">
+          <FileCode2 size={12} aria-hidden="true" />
+          <span className="truncate font-mono" title={data?.path}>
+            {data?.path || 'unknown path'}
+          </span>
+        </span>
+        <span>version {data?.version || 'unknown'}</span>
+        {data?.reload_available === false && (
+          <span>the daemon cannot reload this file at runtime</span>
+        )}
+      </div>
+
+      {data?.last_error ? (
+        <div
+          role="alert"
+          className="flex items-start gap-3 rounded-lg border border-danger/40 bg-danger-soft px-4 py-3"
+        >
+          <ShieldAlert size={18} className="mt-0.5 shrink-0 text-danger" aria-hidden="true" />
+          <div className="min-w-0">
+            <p className="text-sm font-medium text-fg">The daemon reported a config error</p>
+            <p className="mt-0.5 font-mono text-xs break-words text-danger">{data.last_error}</p>
+          </div>
+        </div>
+      ) : null}
+
+      <div className="grid grid-cols-1 gap-3 xl:grid-cols-2">
+        <Card
+          title="Configuration file"
+          subtitle={`${lineCount} lines, as stored on disk`}
+          padded={false}
+          actions={<Badge tone={dirty ? 'warn' : 'muted'}>{dirty ? 'modified' : 'in sync'}</Badge>}
+        >
+          <div className="flex min-w-0 items-stretch">
+            <div
+              ref={gutter}
+              aria-hidden="true"
+              className="max-h-[520px] w-11 shrink-0 overflow-hidden border-r border-line bg-surface-2 py-2 pr-2 text-right font-mono text-xs leading-[1.625] text-fg-subtle select-none"
+            >
+              {gutterLines.map((line) => (
+                <div key={line}>{line}</div>
+              ))}
+            </div>
+            <Textarea
+              monospace
+              value={value}
+              onChange={(event) => setDraft({ version, text: event.currentTarget.value })}
+              onScroll={onScroll}
+              rows={26}
+              spellCheck={false}
+              wrap="off"
+              aria-label="Configuration YAML"
+              className="h-[520px] min-h-[520px] rounded-none border-0 bg-transparent"
+            />
+          </div>
+        </Card>
+
+        <div className="flex min-w-0 flex-col gap-3">
+          <Card
+            title="Validate and save"
+            subtitle="The daemon parses the exact text on the left."
+            actions={
+              <div className="flex items-center gap-2">
+                <Button
+                  icon={CheckCircle2}
+                  onClick={() => void onValidate()}
+                  loading={validate.pending}
+                >
+                  Validate
+                </Button>
+                <Button
+                  variant="primary"
+                  icon={Save}
+                  onClick={() => void onSave()}
+                  loading={save.pending}
+                  disabled={!writable || !dirty}
+                >
+                  Save
+                </Button>
+              </div>
+            }
+          >
+            <CardBody>
+              {!writable && (
+                <p className="flex items-start gap-2 rounded-lg border border-warn/40 bg-warn-soft px-3 py-2 text-xs text-fg">
+                  <Ban size={14} className="mt-0.5 shrink-0 text-warn" aria-hidden="true" />
+                  <span>
+                    This daemon reports the config file as read-only, so Save is disabled. Edit{' '}
+                    <span className="font-mono">{data?.path || 'the file'}</span> on the host and
+                    use Reload from disk.
+                  </span>
+                </p>
+              )}
+              {writable && !dirty && (
+                <p className="text-xs text-fg-subtle">
+                  Nothing to save: the editor matches the file on disk.
+                </p>
+              )}
+              {validate.pending && (
+                <p className="flex items-center gap-2 text-xs text-fg-muted">
+                  <Spinner size="sm" />
+                  validating {lineCount} lines…
+                </p>
+              )}
+              {validate.error !== null && <p className="text-xs text-danger">{validate.error}</p>}
+            </CardBody>
+          </Card>
+
+          <Card
+            title="Validation report"
+            subtitle={
+              report === null
+                ? 'Nothing validated yet'
+                : `${report.errors.length} ${report.errors.length === 1 ? 'error' : 'errors'}, ${report.warnings.length} ${report.warnings.length === 1 ? 'warning' : 'warnings'}`
+            }
+            actions={
+              report === null ? null : (
+                <Badge tone={report.errors.length === 0 ? 'ok' : 'danger'}>
+                  {report.errors.length === 0 ? 'valid' : 'invalid'}
+                </Badge>
+              )
+            }
+          >
+            <CardBody>
+              {report === null ? (
+                <p className="text-xs text-fg-subtle">
+                  Press Validate to parse the current text against the daemon&apos;s schema.
+                </p>
+              ) : report.errors.length === 0 && report.warnings.length === 0 ? (
+                <p className="text-xs text-ok">No errors and no warnings.</p>
+              ) : (
+                <div className="space-y-3">
+                  {report.errors.length > 0 && (
+                    <div>
+                      <p className="mb-1.5 text-xs font-semibold text-danger">Errors</p>
+                      <IssueList issues={report.errors} tone="danger" />
+                    </div>
+                  )}
+                  {report.warnings.length > 0 && (
+                    <div>
+                      <p className="mb-1.5 text-xs font-semibold text-warn">Warnings</p>
+                      <IssueList issues={report.warnings} tone="warn" />
+                    </div>
+                  )}
+                </div>
+              )}
+            </CardBody>
+          </Card>
+        </div>
+      </div>
+
+      <div className="grid grid-cols-1 gap-3 xl:grid-cols-2">
+        <Card
+          title="Effective configuration"
+          subtitle="Your file with every default filled in"
+          actions={<CopyButton value={effective} label="Copy effective config" />}
+        >
+          <CardBody>
+            <details>
+              <summary className="flex cursor-pointer list-none items-center gap-2 text-sm text-fg-muted transition-colors hover:text-fg">
+                <FileCode2 size={14} aria-hidden="true" />
+                Show the effective configuration
+                <span className="text-xs text-fg-subtle">
+                  {effective === '' ? '' : `${effective.split('\n').length} lines`}
+                </span>
+              </summary>
+              <pre className="mt-3 max-h-80 overflow-auto rounded-lg border border-line bg-surface-2 p-3 font-mono text-xs whitespace-pre-wrap text-fg-muted">
+                {effective || 'The server sent no effective configuration.'}
+              </pre>
+            </details>
+          </CardBody>
+        </Card>
+
+        <Card
+          title="Defaults"
+          subtitle="What the daemon falls back to for anything you leave out"
+          actions={<CopyButton value={defaults} label="Copy defaults" />}
+        >
+          <CardBody>
+            <details>
+              <summary className="flex cursor-pointer list-none items-center gap-2 text-sm text-fg-muted transition-colors hover:text-fg">
+                <Info size={14} aria-hidden="true" />
+                Show defaults
+                <span className="text-xs text-fg-subtle">read-only</span>
+              </summary>
+              <pre className="mt-3 max-h-80 overflow-auto rounded-lg border border-line bg-surface-2 p-3 font-mono text-xs whitespace-pre-wrap text-fg-muted">
+                {defaults || 'The server sent no defaults.'}
+              </pre>
+            </details>
+          </CardBody>
+        </Card>
+      </div>
+
+      <Card title="Quick actions" subtitle="The two things operators reach for from here">
+        <CardBody>
+          <div className="flex flex-wrap items-center gap-2">
+            <Button
+              icon={Database}
+              variant="outline"
+              onClick={() => void onFlush()}
+              loading={flush.pending}
+            >
+              Flush cache
+            </Button>
+            <Button
+              icon={RefreshCw}
+              variant="outline"
+              onClick={() => void onRefreshLists()}
+              loading={refreshLists.pending}
+            >
+              Refresh blocklists
+            </Button>
+            <span className="text-xs text-fg-subtle">
+              Both act on the running daemon; neither writes the config file.
+            </span>
+          </div>
+        </CardBody>
+      </Card>
+
+      <ConfirmDialog
+        open={confirmReload}
+        onClose={() => setConfirmReload(false)}
+        onConfirm={() => void onReload()}
+        title="Reload the configuration from disk?"
+        message={
+          <>
+            The daemon re-reads <span className="font-mono">{data?.path || 'its config'}</span>{' '}
+            and applies it to the running process. In-flight queries finish on the old
+            configuration; the next query uses the new one.
+          </>
+        }
+        confirmLabel="Reload"
+        pending={reload.pending}
+      />
+    </div>
+  )
+}

web/src/pages/Upstreams.tsx

@@ -20,10 +20,10 @@ import {
   Trash2,
 } from 'lucide-react'
 
-import { getUpstreams, putUpstreams } from '../../lib/api'
+import { getUpstreams, putUpstreams } from '../lib/api'
-import type { UpstreamBootstrap, UpstreamServer } from '../../lib/types'
+import type { UpstreamBootstrap, UpstreamHealth, UpstreamServer } from '../lib/types'
-import { useApi, useAsyncAction } from '../../hooks/useApi'
+import { useApi, useAsyncAction } from '../hooks/useApi'
-import { formatLatency, formatMs, formatNumber, formatRate } from '../../lib/format'
+import { formatLatency, formatMs, formatNumber, formatRate } from '../lib/format'
 import {
   Badge,
   Button,
@@ -40,21 +40,21 @@ import {
   Textarea,
   useToast,
 } from '../components/ui'
+import type { ToastOptions } from '../components/ui'
 import { PageError, PageHeader } from '../components/layout'
 
 type Row = Record<string, ReactNode>
 
-type ToastLike = {
+/**
-  success?: (message: string) => void
+ * One line for every mutation on this page: the message plus the tone it
-  error?: (message: string) => void
+ * earned. `fail` is the danger tone, `ok` the success one.
-  toast?: (message: string) => void
+ */
-  push?: (message: string) => void
+function notify(
-}
+  toast: (options: ToastOptions) => void,
-
+  kind: 'ok' | 'fail',
-function notify(toast: unknown, kind: 'ok' | 'fail', message: string): void {
+  message: string,
-  const t = toast as ToastLike
+): void {
-  const fn = kind === 'ok' ? (t.success ?? t.toast ?? t.push) : (t.error ?? t.toast ?? t.push)
+  toast({ title: message, tone: kind === 'ok' ? 'ok' : 'danger' })
-  fn?.call(t, message)
 }
 
 const POLICIES = [
@@ -167,7 +167,7 @@ function splitList(value: string): string[] {
 }
 
 export default function Upstreams() {
-  const toast = useToast()
+  const { toast } = useToast()
   const { data, error, loading, refetch } = useApi(() => getUpstreams(), [])
 
   const [draft, setDraft] = useState<Draft | null>(null)
@@ -210,10 +210,15 @@ export default function Upstreams() {
         insecure_skip_verify: s.insecure_skip_verify,
         weight: s.weight,
         comment: s.comment,
-        health: s.health,
+        health: s.health as UpstreamHealth,
         latency_ms: s.latency_ms,
         fail_rate: s.fail_rate,
         breaker_open: s.breaker_open,
+        // Runtime counters are not part of a write: the server owns them and
+        // ignores these fields on PUT /upstreams.
+        queries: 0,
+        avg_ms: 0,
+        p95_ms: 0,
       })),
       bootstraps: working.bootstraps.map((b) => ({
         name: b.name.trim(),
@@ -314,10 +319,10 @@ export default function Upstreams() {
     comment: <span className="block max-w-[160px] truncate text-xs text-slate-500">{s.comment}</span>,
     order: (
       <span className="flex justify-end gap-0.5">
-        <IconButton icon={<ArrowUp size={13} />} label={`move ${s.addr} up`} variant="ghost" disabled={i === 0} onClick={() => move(i, -1)} />
+        <IconButton icon={ArrowUp} label={`move ${s.addr} up`} variant="ghost" disabled={i === 0} onClick={() => move(i, -1)} />
-        <IconButton icon={<ArrowDown size={13} />} label={`move ${s.addr} down`} variant="ghost" disabled={i === working.servers.length - 1} onClick={() => move(i, 1)} />
+        <IconButton icon={ArrowDown} label={`move ${s.addr} down`} variant="ghost" disabled={i === working.servers.length - 1} onClick={() => move(i, 1)} />
-        <IconButton icon={<Pencil size={13} />} label={`edit ${s.addr}`} variant="ghost" onClick={() => { setEditing(s); setEditorOpen(true) }} />
+        <IconButton icon={Pencil} label={`edit ${s.addr}`} variant="ghost" onClick={() => { setEditing(s); setEditorOpen(true) }} />
-        <IconButton icon={<Trash2 size={13} />} label={`remove ${s.addr}`} variant="ghost" onClick={() => setPendingRemove(i)} />
+        <IconButton icon={Trash2} label={`remove ${s.addr}`} variant="ghost" onClick={() => setPendingRemove(i)} />
       </span>
     ),
   }))
@@ -331,13 +336,13 @@ export default function Upstreams() {
           <div className="flex items-center gap-2">
             <Button
               variant="outline"
-              icon={<RotateCw size={14} />}
+              icon={RotateCw}
               onClick={() => setDraft(null)}
               disabled={!dirty}
             >
               Discard
             </Button>
-            <Button icon={<Save size={14} />} onClick={() => void commit()} loading={save.pending} disabled={!dirty}>
+            <Button icon={Save} onClick={() => void commit()} loading={save.pending} disabled={!dirty}>
               Save
             </Button>
           </div>
@@ -386,7 +391,7 @@ export default function Upstreams() {
         actions={
           <Button
             size="sm"
-            icon={<Plus size={14} />}
+            icon={Plus}
             onClick={() => {
               setEditing(EMPTY_SERVER)
               setEditorOpen(true)
@@ -414,7 +419,7 @@ export default function Upstreams() {
         actions={
           <Button
             size="sm"
-            icon={<Plus size={14} />}
+            icon={Plus}
             onClick={() =>
               setDraft({
                 ...working,
@@ -428,7 +433,7 @@ export default function Upstreams() {
       >
         {working.bootstraps.length === 0 ? (
           <EmptyState
-            icon={<Gauge size={20} />}
+            icon={Gauge}
             title="No bootstrap resolvers"
             description="Without them, an upstream address must already be an IP."
           />
@@ -457,7 +462,7 @@ export default function Upstreams() {
                     placeholder="24h"
                   />
                   <IconButton
-                    icon={<Trash2 size={14} />}
+                    icon={Trash2}
                     label={`remove bootstrap ${b.name}`}
                     variant="ghost"
                     onClick={() =>

web/vite.config.ts

@@ -20,7 +20,13 @@ export default defineConfig({
     },
   },
   build: {
-    outDir: 'dist',
+    // The Go binary embeds the console via //go:embed all:dist in
+    // internal/web/web.go, so the build has to land there rather than in
+    // web/dist: a bundle nobody embeds is a console nobody can load.
+    outDir: '../internal/web/dist',
+    emptyOutDir: true,
     sourcemap: false,
+    // Hashed asset names let the server cache them immutably.
+    assetsDir: 'assets',
   },
 })