feat: 增加 Runner 生命周期结构化日志

internal/assignmentqueue/jetstream.go

@@ -78,6 +78,22 @@ type Processor struct {
 	Admission    Admission
 	RetryDelay   time.Duration
 	ClaimTimeout time.Duration
+	OnEvent      func(Event)
+}
+
+// Event describes a non-sensitive assignment handoff transition. It never
+// contains task payloads, credentials, capabilities, or workload identities.
+type Event struct {
+	Name         string
+	AssignmentID string
+	Backend      taskassignment.Backend
+	RetryDelay   time.Duration
+}
+
+func (p Processor) event(name string, assignment taskassignment.Assignment, retryDelay time.Duration) {
+	if p.OnEvent != nil {
+		p.OnEvent(Event{Name: name, AssignmentID: assignment.ID, Backend: assignment.Backend, RetryDelay: retryDelay})
+	}
 }
 
 func (p Processor) Process(ctx context.Context, message Message) error {
@@ -88,6 +104,7 @@ func (p Processor) Process(ctx context.Context, message Message) error {
 	if err != nil {
 		return errors.Join(err, message.TermWithReason("invalid assignment"))
 	}
+	p.event("received", assignment, 0)
 	if _, err := p.Claims.Offer(assignment); err != nil {
 		return errors.Join(err, message.TermWithReason("conflicting assignment"))
 	}
@@ -96,8 +113,10 @@ func (p Processor) Process(ctx context.Context, message Message) error {
 		if delay <= 0 {
 			delay = 2 * time.Second
 		}
+		p.event("capacity_wait", assignment, delay)
 		return message.NakWithDelay(delay)
 	}
+	p.event("capacity_acquired", assignment, 0)
 	accepted, err := p.Accepter.Accept(ctx, assignment)
 	if err != nil {
 		p.Admission.Release(assignment.ID)
@@ -105,9 +124,11 @@ func (p Processor) Process(ctx context.Context, message Message) error {
 		if delay <= 0 {
 			delay = 15 * time.Second
 		}
+		p.event("backend_retry", assignment, delay)
 		return errors.Join(err, message.NakWithDelay(delay))
 	}
 	if accepted {
+		p.event("backend_ready", assignment, 0)
 		timeout := p.ClaimTimeout
 		if timeout <= 0 {
 			timeout = 4 * time.Minute
@@ -120,17 +141,21 @@ func (p Processor) Process(ctx context.Context, message Message) error {
 			if delay <= 0 {
 				delay = 2 * time.Second
 			}
+			p.event("claim_timeout", assignment, delay)
 			return errors.Join(err, message.NakWithDelay(delay))
 		}
+		p.event("runner_claimed", assignment, 0)
 		if err := message.DoubleAck(ctx); err != nil {
 			return fmt.Errorf("ack assignment %s: %w", assignment.ID, err)
 		}
+		p.event("acked", assignment, 0)
 		return nil
 	}
 	delay := p.RetryDelay
 	if delay <= 0 {
 		delay = 2 * time.Second
 	}
+	p.event("backend_pending", assignment, delay)
 	return message.NakWithDelay(delay)
 }