teraWeb/backend/cmd/tera-api/startup_timing.go

package main

import (
	"context"
	"errors"
	"fmt"
	"log"
	"tera-clone/services/tera-api/internal/serial"
	"tera-clone/services/tera-api/internal/sessionhealth"
	"tera-clone/services/tera-api/internal/startup"
	"time"
)

func retryStartupSerial(ctx context.Context, work func() error) error {
	for {
		if err := ctx.Err(); err != nil {
			return err
		}
		err := work()
		if !errors.Is(err, serial.ErrStreamLeaseBusy) {
			return err
		}
		select {
		case <-ctx.Done():
			return ctx.Err()
		case <-time.After(125 * time.Millisecond):
		}
	}
}

func logStartupTiming(stage string, elapsed time.Duration, err error) {
	if err != nil {
		log.Printf("tera-api startup stage=%s elapsed_ms=%d error=%v", stage, elapsed.Milliseconds(), err)
		return
	}
	log.Printf("tera-api startup stage=%s elapsed_ms=%d", stage, elapsed.Milliseconds())
}

func requiredStartupTask(ctx context.Context, boot *startup.Tracker, name string, task func() error) func() {
	return func() {
		started := time.Now()
		err := boot.Retry(ctx, name, 25*time.Second, 3*time.Second, task, func(err error) { log.Printf("tera-api startup retry stage=%s error=%v", name, err) })
		logStartupTiming("runtime."+name, time.Since(started), err)
	}
}

func waitStartupServices(ctx context.Context, ports serial.Controller, sessions sessionhealth.Controller, requireSession bool) error {
	for {
		probeCtx, cancel := context.WithTimeout(ctx, 2*time.Second)
		states, err := ports.States(probeCtx)
		if err == nil && len(states) != serial.ChannelCount {
			err = fmt.Errorf("serial channel count %d", len(states))
		}
		if err == nil {
			for _, s := range states {
				if !s.Available || !s.ProfileApplied {
					err = fmt.Errorf("serial channel unavailable")
					break
				}
			}
		}
		if err == nil && requireSession {
			status, e := sessions.Status(probeCtx)
			err = e
			if err == nil && status.Status != "ok" {
				err = fmt.Errorf("session service not ready")
			}
		}
		cancel()
		if err == nil {
			return nil
		}
		select {
		case <-ctx.Done():
			return fmt.Errorf("service verification: %w", err)
		case <-time.After(time.Second):
		}
	}
}

func timedStartupTask(name string, task func()) func() {
	return func() {
		started := time.Now()
		defer func() { logStartupTiming("runtime."+name, time.Since(started), nil) }()
		task()
	}
}

func verifyStartupPortObservation(ctx context.Context, ports serial.Controller, required bool, after ...time.Time) error {
	observer, ok := ports.(serial.CRProbeStateController)
	if !ok {
		if required {
			return fmt.Errorf("initial port observation unavailable")
		}
		return nil
	}
	states, err := observer.CRProbeStates(ctx)
	if err != nil {
		return err
	}
	if len(states) != serial.ChannelCount {
		return fmt.Errorf("initial port observation incomplete")
	}
	seen := make(map[string]bool)
	for _, state := range states {
		if state.Channel < "CH01" || state.Channel > "CH48" || seen[state.Channel] {
			return fmt.Errorf("invalid initial port observation")
		}
		seen[state.Channel] = true
		if !state.Enabled && state.State == serial.CRProbeStateDisabled {
			continue
		}
		if len(after) > 0 && (state.LastProbeAt == nil || state.LastProbeAt.Before(after[0])) {
			return fmt.Errorf("port %s awaiting observation after settings applied", state.Channel)
		}
		switch state.State {
		case serial.CRProbeStateResponded, serial.CRProbeStateNoResponse:
			if state.LastProbeAt == nil {
				return fmt.Errorf("port %s has no observation", state.Channel)
			}
		case serial.CRProbeStateBusy, serial.CRProbeStateSuppressed:
			if state.LastConfirmedAt == nil && state.ConsecutiveMisses == 0 {
				return fmt.Errorf("port %s awaiting first observation", state.Channel)
			}
		default:
			return fmt.Errorf("port %s initial observation pending (%s)", state.Channel, state.State)
		}
	}
	return nil
}