fix: subscribing to events should not be blocked by agent actor state (#910)

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: mattjohnsonpint

CHANGELOG.md

@@ -2,7 +2,11 @@
 
 # Change Log
 
-## 2025-06-24 - Runtime v0.18.0
+## 2025-06-24 - Runtime v0.18.1
+
+- fix: subscribing to events should not be blocked by agent actor state [#910](https://github.com/hypermodeinc/modus/pull/910)
+
+## 2025-06-23 - Runtime v0.18.0
 
 NOTE: These notes are inclusive of all preview releases in the v0.18.0 series.
 
@@ -38,7 +42,7 @@ Other items:
 - fix: time zone retrieval and logging [#906](https://github.com/hypermodeinc/modus/pull/906)
 - fix: update dockerfile to include legacy time zones [#909](https://github.com/hypermodeinc/modus/pull/909)
 
-## 2025-06-24 - Go SDK v0.18.0
+## 2025-06-23 - Go SDK v0.18.0
 
 NOTE: These notes are inclusive of all preview releases in the v0.18.0 series.
 
@@ -62,7 +66,7 @@ Other items:
 - feat: return user and chat errors in API response [#863](https://github.com/hypermodeinc/modus/pull/863)
 - feat: delete collections features [#872](https://github.com/hypermodeinc/modus/pull/872)
 
-## 2025-06-24 - AssemblyScript SDK v0.18.0
+## 2025-06-23 - AssemblyScript SDK v0.18.0
 
 NOTE: These notes are inclusive of all preview releases in the v0.18.0 series.
 

runtime/actors/agents.go

@@ -168,10 +168,18 @@ func GetAgentInfo(ctx context.Context, agentId string) (*AgentInfo, error) {
 		}, nil
 	}
 
-	// If the actor is not found, or if the request timed out, we can check the database for the agent state.
-	// This is useful for agents that are terminated or suspended, or just busy processing another request.
-	if errors.Is(err, goakt.ErrActorNotFound) || errors.Is(err, goakt.ErrRequestTimeout) {
-		return getAgentInfoFromDatabase(ctx, agentId)
+	// If the actor is not responding, we can check the database for the agent state.
+	// This is useful for agents that are still starting, are terminated or suspended, or just busy processing another request.
+	allowedErrs := []error{
+		goakt.ErrActorNotFound,
+		goakt.ErrRequestTimeout,
+		goakt.ErrRemoteSendFailure,
+		goakt.ErrDead,
+	}
+	for _, r := range allowedErrs {
+		if errors.Is(err, r) {
+			return getAgentInfoFromDatabase(ctx, agentId)
+		}
 	}
 
 	return nil, fmt.Errorf("error getting agent info: %w", err)

runtime/actors/subscriber.go

@@ -33,7 +33,9 @@ func SubscribeForAgentEvents(ctx context.Context, agentId string, update func(da
 	span, ctx := utils.NewSentrySpanForCurrentFunc(ctx)
 	defer span.Finish()
 
-	if a, err := GetAgentInfo(ctx, agentId); err != nil {
+	// Go directly to the database for the agent status, because we don't want subscribing to events to fail
+	// if there is any issue with the agent actor.
+	if a, err := getAgentInfoFromDatabase(ctx, agentId); err != nil {
 		return err
 	} else if a.Status == AgentStatusStopping || a.Status == AgentStatusTerminated {
 		return fmt.Errorf("agent %s is %s, cannot subscribe to events", agentId, a.Status)