teraWeb/backend/internal/startup/status.go

package startup

import (
	"context"
	"sync"
	"time"
)

type Stage struct {
	ID    string `json:"id"`
	Label string `json:"label"`
	State string `json:"state"`
}
type Status struct {
	ClockSyncToken string  `json:"clock_sync_token,omitempty"`
	Ready          bool    `json:"ready"`
	State          string  `json:"state"`
	ElapsedMS      int64   `json:"elapsed_ms"`
	Completed      int     `json:"completed"`
	Total          int     `json:"total"`
	Stages         []Stage `json:"stages"`
}
type Tracker struct {
	mu       sync.Mutex
	started  time.Time
	finished time.Time
	stages   []Stage
}

func New(started time.Time) *Tracker { return &Tracker{started: started} }
func (t *Tracker) Add(id, label string) {
	t.mu.Lock()
	defer t.mu.Unlock()
	t.stages = append(t.stages, Stage{ID: id, Label: label, State: "pending"})
}
func (t *Tracker) Set(id, state string) {
	t.mu.Lock()
	defer t.mu.Unlock()
	for i := range t.stages {
		if t.stages[i].ID == id {
			t.stages[i].State = state
		}
	}
	done := len(t.stages) > 0
	for _, s := range t.stages {
		if s.State != "ready" {
			done = false
		}
	}
	if done && t.finished.IsZero() {
		t.finished = time.Now()
	}
}

// Run bounds the readiness decision even if a device call fails to return.
// A late result can never overwrite the recorded timeout.
func (t *Tracker) Run(id string, timeout time.Duration, work func() error) error {
	t.Set(id, "running")
	result := make(chan error, 1)
	go func() { result <- work() }()
	timer := time.NewTimer(timeout)
	defer timer.Stop()
	var err error
	select {
	case err = <-result:
	case <-timer.C:
		err = context.DeadlineExceeded
	}
	if err != nil {
		t.Set(id, "failed")
	} else {
		t.Set(id, "ready")
	}
	return err
}

// Retry never overlaps attempts, including when a device call stops responding.
// Delayed means work is still pending, not that it succeeded or was abandoned.
func (t *Tracker) Retry(ctx context.Context, id string, delayedAfter, interval time.Duration, work func() error, onError func(error)) error {
	t.Set(id, "running")
	delay := time.NewTimer(delayedAfter)
	defer delay.Stop()
	delayed := false
	markDelayed := func() { delayed = true; t.Set(id, "delayed") }
	for {
		if err := ctx.Err(); err != nil {
			return err
		}
		result := make(chan error, 1)
		go func() { result <- work() }()
		var err error
	wait:
		for {
			select {
			case <-ctx.Done():
				return ctx.Err()
			case <-delay.C:
				markDelayed()
			case err = <-result:
				break wait
			}
		}
		if ctx.Err() != nil {
			return ctx.Err()
		}
		if err == nil {
			t.Set(id, "ready")
			return nil
		}
		if onError != nil {
			onError(err)
		}
		if !delayed {
			t.Set(id, "retrying")
		}
		pause := time.NewTimer(interval)
		select {
		case <-ctx.Done():
			pause.Stop()
			return ctx.Err()
		case <-delay.C:
			markDelayed()
			select {
			case <-ctx.Done():
				pause.Stop()
				return ctx.Err()
			case <-pause.C:
			}
		case <-pause.C:
		}
	}
}
func (t *Tracker) Snapshot() Status {
	t.mu.Lock()
	defer t.mu.Unlock()
	end := time.Now()
	if !t.finished.IsZero() {
		end = t.finished
	}
	s := Status{State: "starting", ElapsedMS: end.Sub(t.started).Milliseconds(), Total: len(t.stages), Stages: append([]Stage(nil), t.stages...)}
	for _, stage := range t.stages {
		if stage.State == "ready" {
			s.Completed++
		}
		if stage.State == "failed" {
			s.State = "failed"
		}
		if stage.State == "delayed" && s.State != "failed" {
			s.State = "delayed"
		}
	}
	s.Ready = s.Total > 0 && s.Completed == s.Total
	if s.Ready {
		s.State = "ready"
	}
	return s
}