fix: websocket connections unreliable

Constant disconnection and reconnection every couple minutes or less.

go/internal/pusher/conn.go

@@ -38,9 +38,10 @@ func (c *Conn) ReadPump(ctx context.Context, hub *Hub) {
 		_, data, err := c.ws.Read(ctx)
 		if err != nil {
 			if ctx.Err() != nil {
+				slog.Info("websocket context cancelled", "connId", c.id, "humanId", c.humanID, "error", ctx.Err())
 				return
 			}
-			slog.Debug("websocket read error", "connId", c.id, "error", err)
+			slog.Info("websocket read error", "connId", c.id, "humanId", c.humanID, "error", err)
 			return
 		}
 

go/k8s/dev/pusher.yaml

@@ -80,6 +80,24 @@ spec:
 
 ---
 
+# WebSocket connections need a much longer timeout than the default 30s.
+# Without this, GKE terminates idle WS connections after 30 seconds.
+apiVersion: networking.gke.io/v1
+kind: GCPBackendPolicy
+metadata:
+  name: pusher-backend-policy
+spec:
+  default:
+    timeoutSec: 60
+    connectionDraining:
+      drainingTimeoutSec: 30
+  targetRef:
+    group: ""
+    kind: Service
+    name: pusher
+
+---
+
 apiVersion: networking.gke.io/v1
 kind: HealthCheckPolicy
 metadata:

go/k8s/prod/pusher.yaml

@@ -77,6 +77,24 @@ spec:
 
 ---
 
+# WebSocket connections need a much longer timeout than the default 30s.
+# Without this, GKE terminates idle WS connections after 30 seconds.
+apiVersion: networking.gke.io/v1
+kind: GCPBackendPolicy
+metadata:
+  name: pusher-backend-policy
+spec:
+  default:
+    timeoutSec: 60
+    connectionDraining:
+      drainingTimeoutSec: 30
+  targetRef:
+    group: ""
+    kind: Service
+    name: pusher
+
+---
+
 apiVersion: networking.gke.io/v1
 kind: HealthCheckPolicy
 metadata: