[WIP] hitless #3447

ndyakov · 2025-07-25T12:35:33Z

No description provided.

ndyakov · 2025-08-05T14:43:59Z

hitless/config.go

+	// Enabled controls whether hitless upgrades are enabled.
+	// Requires RESP3 protocol for push notifications.
+	// Default: false
+	Enabled bool


should be enabled by default?

ndyakov · 2025-08-05T14:44:44Z

hitless/config.go

+	// PostHandoffRelaxedDuration is how long to keep relaxed timeouts on the new connection
+	// after a handoff completes. This provides additional resilience during cluster transitions.
+	// Default: 10 seconds
+	PostHandoffRelaxedDuration time.Duration


default should be RelaxedTimeoutDuration*2 ?

hitless/config.go

ndyakov · 2025-08-05T14:53:15Z

hitless/hitless_manager.go

+
+	// Register handlers for all hitless upgrade notifications with the client's processor
+	if err := processor.RegisterHandler("MOVING", handler, true); err != nil {
+		return err
+	}
+	if err := processor.RegisterHandler("MIGRATING", handler, true); err != nil {
+		return err
+	}
+	if err := processor.RegisterHandler("MIGRATED", handler, true); err != nil {
+		return err
+	}
+	if err := processor.RegisterHandler("FAILING_OVER", handler, true); err != nil {
+		return err
+	}
+	if err := processor.RegisterHandler("FAILED_OVER", handler, true); err != nil {
+		return err
+	}
+


maybe rewrite in a loop? Try couple of different approaches and decide which works the best.

hitless/hitless_manager.go

ndyakov · 2025-08-05T15:29:41Z

hitless/hitless_manager.go

+// GetCurrentState returns current state based on active operations.
+func (hm *HitlessManager) GetCurrentState() State {
+	hm.mu.RLock()
+	defer hm.mu.RUnlock()
+
+	if len(hm.activeMovingOps) > 0 {
+		return StateMoving
+	}
+	return StateIdle
+}


atomic flag instead of lock?

hitless/hitless_manager.go

ndyakov force-pushed the ndyakov/CAE-1072-hitless-upgrades-2 branch 2 times, most recently from 49e5814 to 43aef14 Compare July 25, 2025 12:51

ndyakov changed the base branch from ndyakov/CAE-1088-resp3-notification-handlers to master July 25, 2025 12:52

ndyakov force-pushed the ndyakov/CAE-1072-hitless-upgrades-2 branch 21 times, most recently from 8608f93 to a39a23a Compare July 30, 2025 07:05

ndyakov force-pushed the ndyakov/CAE-1072-hitless-upgrades-2 branch 2 times, most recently from e88e673 to 4542e8f Compare August 4, 2025 13:01

ndyakov changed the base branch from master to ndyakov/CAE-1088-resp3-notification-handlers August 4, 2025 13:02

ndyakov force-pushed the ndyakov/CAE-1072-hitless-upgrades-2 branch 3 times, most recently from 9590c26 to 100c3d2 Compare August 4, 2025 13:57

ndyakov force-pushed the ndyakov/CAE-1072-hitless-upgrades-2 branch from 100c3d2 to de641c5 Compare August 4, 2025 13:58

This was referenced Aug 5, 2025

fix the maxActiveConns sematic #2934

Closed

feat: Allow for pubsub connections to optionally be taken from pool #2869

Closed