feat: parallel fan-out delivery + circuit breaker for retry targets

- Fan out all targets for an event in parallel goroutines (fire-and-forget)
- Add per-target circuit breaker for retry targets (closed/open/half-open)
- Circuit breaker trips after 5 consecutive failures, 30s cooldown
- Open circuit skips delivery and reschedules after cooldown
- Half-open allows one probe delivery to test recovery
- HTTP/database/log targets unaffected (no circuit breaker)
- Recovery path also fans out in parallel
- Update README with parallel delivery and circuit breaker docs

internal/delivery/engine.go

@@ -104,6 +104,11 @@ type EngineParams struct {
 // DeliveryTask so retries also avoid unnecessary DB reads. The database
 // stores delivery status for crash recovery only; on startup the engine
 // scans for interrupted deliveries and re-queues them.
+//
+// All targets for a single event are delivered in parallel — each
+// DeliveryTask is dispatched in its own goroutine for maximum fan-out
+// speed. Retry targets are protected by a per-target circuit breaker
+// that stops hammering a down target after consecutive failures.
 type Engine struct {
 	database  *database.Database
 	dbManager *database.WebhookDBManager
@@ -113,6 +118,11 @@ type Engine struct {
 	wg        sync.WaitGroup
 	notifyCh  chan []DeliveryTask
 	retryCh   chan DeliveryTask
+
+	// circuitBreakers stores a *CircuitBreaker per target ID. Only used
+	// for retry targets — HTTP, database, and log targets do not need
+	// circuit breakers because they either fire once or are local ops.
+	circuitBreakers sync.Map
 }
 
 // New creates and registers the delivery engine with the fx lifecycle.
@@ -347,10 +357,12 @@ func (e *Engine) recoverWebhookDeliveries(ctx context.Context, webhookID string)
 }
 
 // processDeliveryTasks handles a batch of delivery tasks from the webhook
-// handler. In the happy path (body ≤ 16KB), the engine delivers without
-// reading from any database — it trusts the handler's inline data and
-// only touches the DB to record results. For large bodies (body > 16KB),
-// the body is fetched from the per-webhook database on demand.
+// handler. Each task is dispatched in its own goroutine for parallel
+// fan-out — all targets for a single event start delivering simultaneously.
+// In the happy path (body ≤ 16KB), the engine delivers without reading
+// from any database — it trusts the handler's inline data and only touches
+// the DB to record results. For large bodies (body > 16KB), the body is
+// fetched once and shared across all goroutines in the batch.
 func (e *Engine) processDeliveryTasks(ctx context.Context, tasks []DeliveryTask) {
 	if len(tasks) == 0 {
 		return
@@ -367,10 +379,25 @@ func (e *Engine) processDeliveryTasks(ctx context.Context, tasks []DeliveryTask)
 		return
 	}
 
-	// For the large-body case, we may need to fetch the event body once
-	// for all tasks sharing the same event. Cache it here.
+	// For the large-body case, pre-fetch the event body once before
+	// fanning out so all goroutines share the same data.
 	var fetchedBody *string
+	if tasks[0].Body == nil {
+		var dbEvent database.Event
+		if err := webhookDB.Select("body").
+			First(&dbEvent, "id = ?", tasks[0].EventID).Error; err != nil {
+			e.log.Error("failed to fetch event body from database",
+				"event_id", tasks[0].EventID,
+				"error", err,
+			)
+			return
+		}
+		fetchedBody = &dbEvent.Body
+	}
 
+	// Fan out: spin up a goroutine per task for parallel delivery.
+	// Each goroutine is independent (fire-and-forget) and records its
+	// own result. No need to wait for all goroutines to finish.
 	for i := range tasks {
 		select {
 		case <-ctx.Done():
@@ -378,60 +405,61 @@ func (e *Engine) processDeliveryTasks(ctx context.Context, tasks []DeliveryTask)
 		default:
 		}
 
-		task := &tasks[i]
+		task := tasks[i] // copy for goroutine closure safety
 
-		// Build Event from task data
-		event := database.Event{
-			Method:      task.Method,
-			Headers:     task.Headers,
-			ContentType: task.ContentType,
-		}
-		event.ID = task.EventID
-		event.WebhookID = task.WebhookID
-
-		if task.Body != nil {
-			// Happy path: body inline, no DB read needed
-			event.Body = *task.Body
-		} else {
-			// Large body path: fetch from per-webhook DB (once per batch)
-			if fetchedBody == nil {
-				var dbEvent database.Event
-				if err := webhookDB.Select("body").
-					First(&dbEvent, "id = ?", task.EventID).Error; err != nil {
-					e.log.Error("failed to fetch event body from database",
-						"event_id", task.EventID,
-						"error", err,
-					)
-					continue
-				}
-				fetchedBody = &dbEvent.Body
-			}
-			event.Body = *fetchedBody
-		}
-
-		// Build Target from task data (no main DB query needed)
-		target := database.Target{
-			Name:       task.TargetName,
-			Type:       task.TargetType,
-			Config:     task.TargetConfig,
-			MaxRetries: task.MaxRetries,
-		}
-		target.ID = task.TargetID
-
-		// Build Delivery struct for the processing chain
-		d := &database.Delivery{
-			EventID:  task.EventID,
-			TargetID: task.TargetID,
-			Status:   database.DeliveryStatusPending,
-			Event:    event,
-			Target:   target,
-		}
-		d.ID = task.DeliveryID
-
-		e.processDelivery(ctx, webhookDB, d, task)
+		go func() {
+			e.deliverTask(ctx, webhookDB, &task, fetchedBody)
+		}()
 	}
 }
 
+// deliverTask prepares and executes a single delivery task. Called from
+// a dedicated goroutine for parallel fan-out.
+func (e *Engine) deliverTask(ctx context.Context, webhookDB *gorm.DB, task *DeliveryTask, fetchedBody *string) {
+	// Build Event from task data
+	event := database.Event{
+		Method:      task.Method,
+		Headers:     task.Headers,
+		ContentType: task.ContentType,
+	}
+	event.ID = task.EventID
+	event.WebhookID = task.WebhookID
+
+	switch {
+	case task.Body != nil:
+		event.Body = *task.Body
+	case fetchedBody != nil:
+		event.Body = *fetchedBody
+	default:
+		e.log.Error("no body available for delivery task",
+			"delivery_id", task.DeliveryID,
+			"event_id", task.EventID,
+		)
+		return
+	}
+
+	// Build Target from task data (no main DB query needed)
+	target := database.Target{
+		Name:       task.TargetName,
+		Type:       task.TargetType,
+		Config:     task.TargetConfig,
+		MaxRetries: task.MaxRetries,
+	}
+	target.ID = task.TargetID
+
+	// Build Delivery struct for the processing chain
+	d := &database.Delivery{
+		EventID:  task.EventID,
+		TargetID: task.TargetID,
+		Status:   database.DeliveryStatusPending,
+		Event:    event,
+		Target:   target,
+	}
+	d.ID = task.DeliveryID
+
+	e.processDelivery(ctx, webhookDB, d, task)
+}
+
 // processRetryTask handles a single delivery task fired by a retry timer.
 // The task carries all data needed for delivery (same as the initial
 // notification). The only DB read is a status check to verify the delivery
@@ -562,41 +590,47 @@ func (e *Engine) processWebhookPendingDeliveries(ctx context.Context, webhookID
 		targetMap[t.ID] = t
 	}
 
+	// Fan out recovered deliveries in parallel — same as the normal
+	// delivery path, each task gets its own goroutine.
 	for i := range deliveries {
 		select {
 		case <-ctx.Done():
 			return
 		default:
-			target, ok := targetMap[deliveries[i].TargetID]
-			if !ok {
-				e.log.Error("target not found for delivery",
-					"delivery_id", deliveries[i].ID,
-					"target_id", deliveries[i].TargetID,
-				)
-				continue
-			}
-			deliveries[i].Target = target
-
-			// Build task from DB data for the recovery path
-			bodyStr := deliveries[i].Event.Body
-			task := &DeliveryTask{
-				DeliveryID:   deliveries[i].ID,
-				EventID:      deliveries[i].EventID,
-				WebhookID:    webhookID,
-				TargetID:     target.ID,
-				TargetName:   target.Name,
-				TargetType:   target.Type,
-				TargetConfig: target.Config,
-				MaxRetries:   target.MaxRetries,
-				Method:       deliveries[i].Event.Method,
-				Headers:      deliveries[i].Event.Headers,
-				ContentType:  deliveries[i].Event.ContentType,
-				Body:         &bodyStr,
-				AttemptNum:   1,
-			}
-
-			e.processDelivery(ctx, webhookDB, &deliveries[i], task)
 		}
+
+		target, ok := targetMap[deliveries[i].TargetID]
+		if !ok {
+			e.log.Error("target not found for delivery",
+				"delivery_id", deliveries[i].ID,
+				"target_id", deliveries[i].TargetID,
+			)
+			continue
+		}
+		deliveries[i].Target = target
+
+		// Build task from DB data for the recovery path
+		bodyStr := deliveries[i].Event.Body
+		task := DeliveryTask{
+			DeliveryID:   deliveries[i].ID,
+			EventID:      deliveries[i].EventID,
+			WebhookID:    webhookID,
+			TargetID:     target.ID,
+			TargetName:   target.Name,
+			TargetType:   target.Type,
+			TargetConfig: target.Config,
+			MaxRetries:   target.MaxRetries,
+			Method:       deliveries[i].Event.Method,
+			Headers:      deliveries[i].Event.Headers,
+			ContentType:  deliveries[i].Event.ContentType,
+			Body:         &bodyStr,
+			AttemptNum:   1,
+		}
+
+		d := deliveries[i] // copy for goroutine closure safety
+		go func() {
+			e.processDelivery(ctx, webhookDB, &d, &task)
+		}()
 	}
 }
 
@@ -683,6 +717,26 @@ func (e *Engine) deliverRetry(_ context.Context, webhookDB *gorm.DB, d *database
 		return
 	}
 
+	// Check the circuit breaker for this target before attempting delivery.
+	cb := e.getCircuitBreaker(task.TargetID)
+	if !cb.Allow() {
+		// Circuit is open — skip delivery, mark as retrying, and
+		// schedule a retry for after the cooldown expires.
+		remaining := cb.CooldownRemaining()
+		e.log.Info("circuit breaker open, skipping delivery",
+			"target_id", task.TargetID,
+			"target_name", task.TargetName,
+			"delivery_id", d.ID,
+			"cooldown_remaining", remaining,
+		)
+		e.updateDeliveryStatus(webhookDB, d, database.DeliveryStatusRetrying)
+
+		retryTask := *task
+		// Don't increment AttemptNum — this wasn't a real attempt
+		e.scheduleRetry(retryTask, remaining)
+		return
+	}
+
 	attemptNum := task.AttemptNum
 
 	// Attempt delivery immediately — backoff is handled by the timer
@@ -698,10 +752,14 @@ func (e *Engine) deliverRetry(_ context.Context, webhookDB *gorm.DB, d *database
 	e.recordResult(webhookDB, d, attemptNum, success, statusCode, respBody, errMsg, duration)
 
 	if success {
+		cb.RecordSuccess()
 		e.updateDeliveryStatus(webhookDB, d, database.DeliveryStatusDelivered)
 		return
 	}
 
+	// Delivery failed — record failure in circuit breaker
+	cb.RecordFailure()
+
 	maxRetries := d.Target.MaxRetries
 	if maxRetries <= 0 {
 		maxRetries = 5 // default
@@ -727,6 +785,20 @@ func (e *Engine) deliverRetry(_ context.Context, webhookDB *gorm.DB, d *database
 	}
 }
 
+// getCircuitBreaker returns the circuit breaker for the given target ID,
+// creating one if it doesn't exist yet. Circuit breakers are in-memory
+// only and reset on restart (startup recovery rescans the DB anyway).
+func (e *Engine) getCircuitBreaker(targetID string) *CircuitBreaker {
+	if val, ok := e.circuitBreakers.Load(targetID); ok {
+		cb, _ := val.(*CircuitBreaker) //nolint:errcheck // type is guaranteed by LoadOrStore below
+		return cb
+	}
+	fresh := NewCircuitBreaker()
+	actual, _ := e.circuitBreakers.LoadOrStore(targetID, fresh)
+	cb, _ := actual.(*CircuitBreaker) //nolint:errcheck // we only store *CircuitBreaker values
+	return cb
+}
+
 // deliverDatabase handles the database target type. Since events are already
 // stored in the per-webhook database (that's the whole point of per-webhook
 // databases), the database target simply marks the delivery as successful.