Files
go-cart-actor/pkg/actor/simple_grain_pool_test.go
mats cecf4abe76
Build and Publish / BuildAndDeployArm64 (push) Failing after 6s
Build and Publish / BuildAndDeployAmd64 (push) Failing after 4s
update
2026-07-03 17:45:51 +02:00

435 lines
11 KiB
Go

package actor
import (
"context"
"fmt"
"io"
"net/http"
"sync"
"sync/atomic"
"testing"
"time"
"google.golang.org/protobuf/proto"
)
type testState struct {
Value int `json:"value"`
}
type testGrain struct {
id uint64
accessed time.Time
}
func (g *testGrain) GetId() uint64 { return g.id }
func (g *testGrain) GetLastAccess() time.Time { return g.accessed }
func (g *testGrain) GetLastChange() time.Time { return g.accessed }
func (g *testGrain) GetCurrentState() (*testState, error) {
return &testState{Value: int(g.id)}, nil
}
type mockHost struct {
name string
healthy bool
closed atomic.Bool
closeCalls atomic.Int32
actorIDs []uint64
negotiateFn func([]string) ([]string, error)
mu sync.Mutex
ownership []uint64
expiry []uint64
}
func newMockHost(name string) *mockHost {
return &mockHost{name: name, healthy: true}
}
func (m *mockHost) AnnounceExpiry(ids []uint64, _ []int64) {
m.mu.Lock()
m.expiry = append(m.expiry, ids...)
m.mu.Unlock()
}
func (m *mockHost) Negotiate(otherHosts []string) ([]string, error) {
if m.negotiateFn != nil {
return m.negotiateFn(otherHosts)
}
return otherHosts, nil
}
func (m *mockHost) Name() string { return m.name }
func (m *mockHost) Proxy(_ uint64, _ http.ResponseWriter, _ *http.Request, _ io.Reader) (bool, error) {
return false, fmt.Errorf("not implemented")
}
func (m *mockHost) Apply(_ context.Context, id uint64, _ ...proto.Message) (*MutationResult[testState], error) {
return &MutationResult[testState]{Result: testState{Value: int(id)}}, nil
}
func (m *mockHost) Get(_ context.Context, id uint64) (*testState, error) {
return &testState{Value: int(id)}, nil
}
func (m *mockHost) GetActorIds() []uint64 { return m.actorIDs }
func (m *mockHost) Close() error {
m.closeCalls.Add(1)
m.closed.Store(true)
return nil
}
func (m *mockHost) Ping() bool { return m.healthy }
func (m *mockHost) IsHealthy() bool { return m.healthy }
func (m *mockHost) AnnounceOwnership(_ string, ids []uint64, _ []int64) {
m.mu.Lock()
m.ownership = append(m.ownership, ids...)
m.mu.Unlock()
}
func newTestPool(t *testing.T, spawnHost func(string) (Host[testState], error)) *SimpleGrainPool[testState] {
t.Helper()
reg := NewMutationRegistry()
pool, err := NewSimpleGrainPool(GrainPoolConfig[testState]{
Hostname: "local",
Spawn: func(_ context.Context, id uint64) (Grain[testState], error) {
return &testGrain{id: id, accessed: time.Now()}, nil
},
SpawnHost: spawnHost,
TTL: time.Hour,
PoolSize: 100,
MutationRegistry: reg,
})
if err != nil {
t.Fatal(err)
}
t.Cleanup(pool.Close)
return pool
}
func TestSimpleGrainPoolGetAndApply(t *testing.T) {
pool := newTestPool(t, func(string) (Host[testState], error) {
return nil, fmt.Errorf("no remotes")
})
got, err := pool.Get(context.Background(), 42)
if err != nil {
t.Fatal(err)
}
if got.Value != 42 {
t.Fatalf("Get value = %d, want 42", got.Value)
}
ids := pool.GetLocalIds()
if len(ids) != 1 || ids[0] != 42 {
t.Fatalf("GetLocalIds = %v, want [42]", ids)
}
}
func TestSimpleGrainPoolAddRemoteDedup(t *testing.T) {
var spawnCalls atomic.Int32
host := newMockHost("peer-a")
pool := newTestPool(t, func(name string) (Host[testState], error) {
spawnCalls.Add(1)
if name != host.name {
t.Fatalf("spawn host %q, want %q", name, host.name)
}
return host, nil
})
const workers = 8
var wg sync.WaitGroup
wg.Add(workers)
for range workers {
go func() {
defer wg.Done()
if _, err := pool.AddRemote(host.name); err != nil {
t.Error(err)
}
}()
}
wg.Wait()
if pool.RemoteCount() != 1 {
t.Fatalf("RemoteCount = %d, want 1", pool.RemoteCount())
}
if spawnCalls.Load() != 1 {
t.Fatalf("spawn calls = %d, want 1", spawnCalls.Load())
}
if host.closeCalls.Load() != 0 {
t.Fatalf("duplicate remote closed = %d, want 0", host.closeCalls.Load())
}
}
func TestSimpleGrainPoolAddRemoteRaceClosesDuplicate(t *testing.T) {
start := make(chan struct{})
var spawnCalls atomic.Int32
var spawned []*mockHost
pool := newTestPool(t, func(name string) (Host[testState], error) {
spawnCalls.Add(1)
h := newMockHost(name)
spawned = append(spawned, h)
if spawnCalls.Load() == 1 {
<-start
}
return h, nil
})
go func() {
if _, err := pool.AddRemote("peer-b"); err != nil {
t.Error(err)
}
}()
time.Sleep(20 * time.Millisecond)
close(start)
if _, err := pool.AddRemote("peer-b"); err != nil {
t.Fatal(err)
}
if pool.RemoteCount() != 1 {
t.Fatalf("RemoteCount = %d, want 1", pool.RemoteCount())
}
if spawnCalls.Load() != 2 {
t.Fatalf("spawn calls = %d, want 2", spawnCalls.Load())
}
closed := 0
deadline := time.After(2 * time.Second)
for closed < 1 {
closed = 0
for _, h := range spawned {
if h.closeCalls.Load() > 0 {
closed++
}
}
if closed >= 1 {
break
}
select {
case <-deadline:
t.Fatal("timed out waiting for duplicate remote host Close")
default:
time.Sleep(10 * time.Millisecond)
}
}
}
func TestSimpleGrainPoolRemoveHost(t *testing.T) {
host := newMockHost("peer-c")
pool := newTestPool(t, func(string) (Host[testState], error) { return host, nil })
if _, err := pool.AddRemote(host.name); err != nil {
t.Fatal(err)
}
pool.remoteMu.Lock()
pool.remoteOwners[99] = host
pool.remoteMu.Unlock()
pool.RemoveHost(host.name)
if pool.RemoteCount() != 0 {
t.Fatalf("RemoteCount = %d, want 0", pool.RemoteCount())
}
if owner, ok := pool.OwnerHost(99); ok || owner != nil {
t.Fatalf("OwnerHost still mapped: %#v ok=%v", owner, ok)
}
if host.closeCalls.Load() != 1 {
t.Fatalf("Close calls = %d, want 1", host.closeCalls.Load())
}
}
func TestSimpleGrainPoolHandleOwnershipChange(t *testing.T) {
host := newMockHost("peer-d")
pool := newTestPool(t, func(string) (Host[testState], error) { return host, nil })
// First-spawn-wins arbitration: a remote broadcast wins only when its
// lastChange stamp is strictly earlier than the local grain's stamp
// (or the local grain is absent). Capture the local spawn time once
// and pass a stamp 1ns earlier to the remote so the test continues to
// assert "external owner takes over given an earlier remote spawn"
// under the new arbitration rule.
spawnTime := time.Now()
pool.localMu.Lock()
pool.grains[7] = &testGrain{id: 7, accessed: spawnTime}
pool.localMu.Unlock()
remoteStamp := spawnTime.UnixNano() - 1
if err := pool.HandleOwnershipChange(host.name, []uint64{7}, []int64{remoteStamp}); err != nil {
t.Fatal(err)
}
pool.localMu.RLock()
_, local := pool.grains[7]
pool.localMu.RUnlock()
if local {
t.Fatal("local grain should be removed after ownership change")
}
owner, ok := pool.OwnerHost(7)
if !ok || owner.Name() != host.name {
t.Fatalf("OwnerHost = (%v, %v), want (%q, true)", owner, ok, host.name)
}
}
// TestSimpleGrainPoolHandleOwnershipChangeArbitration verifies the
// first-spawn-wins arbitration rules for HandleOwnershipChange:
//
// - Remote with earlier lastChange than local → accept remote.
// - Remote with later lastChange than local → keep local.
// - Equal lastChange → lower hostname wins.
// - Remote stamp == -1 (legacy path) → accept remote (pre-arbitration).
// - Local grain absent → accept remote.
func TestSimpleGrainPoolHandleOwnershipChangeArbitration(t *testing.T) {
host := newMockHost("peer-arbitrate")
// Each table case constructs a fresh pool inside t.Run — we don't
// share state across cases because per-case hostname differs
// (hostname is the tie-break oracle on equal lastChange stamps).
now := time.Now().UnixNano()
cases := []struct {
name string
name_local string // pool hostname for ordering check
id uint64
localStamp int64 // UnixNano of the local grain; -1 = no local grain
remoteStamp int64 // UnixNano stamped on the broadcast
expectLocal bool // true → local grain still resident
expectRemote bool // true → remoteOwners[id] = host
}{
{
name: "remote older stamp wins",
name_local: "zzz", // pool above "peer-arbitrate" lex
id: 10,
localStamp: now + 100,
remoteStamp: now,
expectLocal: false,
expectRemote: true,
},
{
name: "local earlier stamp keeps",
name_local: "aaa", // pool below "peer-arbitrate" lex but stamp decides
id: 11,
localStamp: now,
remoteStamp: now + 100,
expectLocal: true,
expectRemote: false,
},
{
name: "equal stamp + lower local hostname keeps",
name_local: "a-peer", // below host
id: 12,
localStamp: now,
remoteStamp: now,
expectLocal: true,
expectRemote: false,
},
{
name: "equal stamp + higher local hostname defers",
name_local: "zzz-peer", // above host
id: 13,
localStamp: now,
remoteStamp: now,
expectLocal: false,
expectRemote: true,
},
{
name: "legacy (no remote stamp) accepts remote",
name_local: "zzz", // we'd otherwise keep local
id: 14,
localStamp: now + 100,
remoteStamp: -1,
expectLocal: false,
expectRemote: true,
},
{
name: "no local grain accepts remote",
name_local: "aaa",
id: 15,
localStamp: -1,
remoteStamp: now,
expectLocal: false,
expectRemote: true,
},
}
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
// Fresh pool per case so hostname differences don't leak.
p := newTestPool(t, func(string) (Host[testState], error) { return host, nil })
p.hostname = tc.name_local
p.localMu.Lock()
if tc.localStamp != -1 {
p.grains[tc.id] = &testGrain{id: tc.id, accessed: time.Unix(0, tc.localStamp)}
}
p.localMu.Unlock()
if err := p.HandleOwnershipChange(host.name, []uint64{tc.id}, []int64{tc.remoteStamp}); err != nil {
t.Fatal(err)
}
p.localMu.RLock()
_, hasLocal := p.grains[tc.id]
p.localMu.RUnlock()
if hasLocal != tc.expectLocal {
t.Fatalf("local grain presence = %v, want %v", hasLocal, tc.expectLocal)
}
owner, ok := p.OwnerHost(tc.id)
if ok != tc.expectRemote {
t.Fatalf("OwnerHost ok = %v, want %v", ok, tc.expectRemote)
}
if tc.expectRemote && (owner == nil || owner.Name() != host.name) {
t.Fatalf("Owner = %v, want host %q", owner, host.name)
}
})
}
}
func TestSimpleGrainPoolPurgeEvictsStaleGrains(t *testing.T) {
pool := newTestPool(t, func(string) (Host[testState], error) {
return nil, fmt.Errorf("no remotes")
})
pool.ttl = time.Minute
stale := &testGrain{id: 1, accessed: time.Now().Add(-2 * time.Minute)}
fresh := &testGrain{id: 2, accessed: time.Now()}
pool.localMu.Lock()
pool.grains[1] = stale
pool.grains[2] = fresh
pool.localMu.Unlock()
pool.purge()
pool.localMu.RLock()
defer pool.localMu.RUnlock()
if _, ok := pool.grains[1]; ok {
t.Fatal("stale grain should be purged")
}
if _, ok := pool.grains[2]; !ok {
t.Fatal("fresh grain should remain")
}
}
func TestSimpleGrainPoolBroadcastOwnership(t *testing.T) {
host := newMockHost("peer-e")
pool := newTestPool(t, func(string) (Host[testState], error) { return host, nil })
if _, err := pool.AddRemote(host.name); err != nil {
t.Fatal(err)
}
pool.broadcastOwnership([]uint64{11, 12})
host.mu.Lock()
defer host.mu.Unlock()
if len(host.ownership) != 2 {
t.Fatalf("ownership announces = %v, want 2 ids", host.ownership)
}
}