package topology

import (
	"bufio"
	"context"
	"sort"
	"time"

	corev1 "k8s.io/api/core/v1"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
)

// PodLog is what looking up a running pod behind a Deployment found.
type PodLog struct {
	Pod   string   // empty unless Found
	Lines []string // most recent log lines, oldest first
	Found bool     // a pod exists and was queried (Lines may still be empty: a quiet pod)
	Err   string   // set when the lookup itself failed; Found is false

	// The pod's own lifecycle facts, which explain errors the log lines alone
	// cannot: a 503 burst while a replica is being replaced, a crash loop the
	// kubelet is already retrying, or a pod that runs but never becomes ready.
	Ready            bool      // container passing its readiness probe
	Phase            string    // Pending | Running | Succeeded | Failed | Unknown
	Restarts         int32     // restarts of the selected pod's containers
	PodRestarts      int32     // restarts summed over every pod of the Deployment
	Reason           string    // CrashLoopBackOff, OOMKilled, ImagePullBackOff, ... ("" while healthy)
	ExitCode         *int32    // exit code of the last container termination
	Message          string    // message of the last container termination
	LastTerminatedAt time.Time // when the last container terminated
}

// rollingWindow is how recently a container must have died for the resulting
// connection errors to be attributed to its replacement still coming up.
const rollingWindow = 5 * time.Minute

// CrashLooping reports whether the pod is stuck restarting instead of serving.
// Restarting the Deployment then only repeats what the kubelet is already
// doing, so the answer is a new release, not a restart.
func (p PodLog) CrashLooping() bool {
	if p.Reason == "CrashLoopBackOff" {
		return true
	}
	return p.Restarts > 0 && !p.LastTerminatedAt.IsZero()
}

// Restarting reports whether a pod of this Deployment is right now being
// replaced, which is what makes a short 503 burst expected rather than new.
func (p PodLog) Restarting() bool {
	return !p.LastTerminatedAt.IsZero() && time.Since(p.LastTerminatedAt) < rollingWindow
}

// podLogTimeout bounds one lookup so a slow or unreachable API server never
// stalls alert delivery.
const podLogTimeout = 5 * time.Second

// PodLogs finds a pod behind namespace/deployment and returns its last `tail`
// log lines. A Deployment with no live pods right now - scaled to 0, or mid
// rollout with none ready yet - comes back with Found=false and no Err: that
// absence is itself the answer, not a failure.
func (r *Registry) PodLogs(ctx context.Context, namespace, deployment string, tail int) PodLog {
	if r == nil || r.client == nil {
		return PodLog{}
	}
	ctx, cancel := context.WithTimeout(ctx, podLogTimeout)
	defer cancel()

	dep, err := r.client.AppsV1().Deployments(namespace).Get(ctx, deployment, metav1.GetOptions{})
	if err != nil {
		return PodLog{Err: err.Error()}
	}
	sel, err := metav1.LabelSelectorAsSelector(dep.Spec.Selector)
	if err != nil {
		return PodLog{Err: err.Error()}
	}
	pods, err := r.client.CoreV1().Pods(namespace).List(ctx, metav1.ListOptions{LabelSelector: sel.String()})
	if err != nil {
		return PodLog{Err: err.Error()}
	}
	live := make([]corev1.Pod, 0, len(pods.Items))
	for _, p := range pods.Items {
		if p.DeletionTimestamp == nil {
			live = append(live, p)
		}
	}
	if len(live) == 0 {
		return PodLog{} // scaled to 0, or nothing ready yet
	}
	// Prefer a Running pod; among ties prefer the newest, most likely to hold
	// the log lines behind whatever just triggered the alert.
	sort.Slice(live, func(i, j int) bool {
		ri, rj := live[i].Status.Phase == corev1.PodRunning, live[j].Status.Phase == corev1.PodRunning
		if ri != rj {
			return ri
		}
		return live[i].CreationTimestamp.After(live[j].CreationTimestamp.Time)
	})
	pod := live[0]
	state := podState(pod, live)
	container := ""
	if len(pod.Spec.Containers) > 0 {
		container = pod.Spec.Containers[0].Name
	}
	tailLines := int64(tail)
	stream, err := r.client.CoreV1().Pods(namespace).GetLogs(pod.Name, &corev1.PodLogOptions{Container: container, TailLines: &tailLines}).Stream(ctx)
	if err != nil {
		return withState(state, PodLog{Pod: pod.Name, Found: true, Err: err.Error()})
	}
	defer stream.Close()
	var lines []string
	sc := bufio.NewScanner(stream)
	sc.Buffer(make([]byte, 0, 64*1024), 1024*1024)
	for sc.Scan() {
		lines = append(lines, sc.Text())
	}
	if len(lines) > tail {
		lines = lines[len(lines)-tail:]
	}
	return withState(state, PodLog{Pod: pod.Name, Lines: lines, Found: true})
}

// withState keeps the query result but carries the lifecycle facts along.
func withState(state, result PodLog) PodLog {
	result.Ready, result.Phase, result.Restarts, result.PodRestarts = state.Ready, state.Phase, state.Restarts, state.PodRestarts
	result.Reason, result.ExitCode, result.Message, result.LastTerminatedAt = state.Reason, state.ExitCode, state.Message, state.LastTerminatedAt
	return result
}

// podState reads the lifecycle facts off the selected pod and totals the
// restarts of its siblings, so one unhealthy replica is not mistaken for a
// healthy deployment.
func podState(selected corev1.Pod, all []corev1.Pod) PodLog {
	state := PodLog{Found: true, Ready: len(selected.Status.ContainerStatuses) > 0, Phase: string(selected.Status.Phase)}
	// The most recent termination across the pod's containers is the
	// interesting one: a pod being replaced is explained by its last death.
	keepTermination := func(term *corev1.ContainerStateTerminated) {
		if term == nil || !term.FinishedAt.Time.After(state.LastTerminatedAt) {
			return
		}
		state.LastTerminatedAt = term.FinishedAt.Time
		state.Message = term.Message
		code := term.ExitCode
		state.ExitCode = &code
		switch {
		case term.Reason != "":
			state.Reason = term.Reason
		case term.Message != "":
			state.Reason = "Error"
		}
	}
	for i := range selected.Status.ContainerStatuses {
		cs := &selected.Status.ContainerStatuses[i]
		state.Restarts += cs.RestartCount
		if !cs.Ready {
			state.Ready = false
		}
		keepTermination(cs.LastTerminationState.Terminated)
	}
	if selected.Status.Reason != "" {
		state.Reason = selected.Status.Reason
	}
	for _, p := range all {
		for _, cs := range p.Status.ContainerStatuses {
			state.PodRestarts += cs.RestartCount
		}
	}
	return state
}