From 66aec4478f0a0945072a6823b5657dc6793548c6 Mon Sep 17 00:00:00 2001
From: talksik <patel.arjun50@gmail.com>
Date: Thu, 9 Apr 2026 16:18:38 -0700
Subject: [PATCH] fix: websocket connections unreliable

Constant disconnection and reconnection every couple minutes or less.
---
 go/internal/pusher/conn.go |  3 ++-
 go/k8s/dev/pusher.yaml     | 18 ++++++++++++++++++
 go/k8s/prod/pusher.yaml    | 18 ++++++++++++++++++
 3 files changed, 38 insertions(+), 1 deletion(-)

diff --git a/go/internal/pusher/conn.go b/go/internal/pusher/conn.go
index c28ab33..89d46c7 100644
--- a/go/internal/pusher/conn.go
+++ b/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
 		}
 
diff --git a/go/k8s/dev/pusher.yaml b/go/k8s/dev/pusher.yaml
index 3053761..8548375 100644
--- a/go/k8s/dev/pusher.yaml
+++ b/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:
diff --git a/go/k8s/prod/pusher.yaml b/go/k8s/prod/pusher.yaml
index 2eed66b..e4649a5 100644
--- a/go/k8s/prod/pusher.yaml
+++ b/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: