From 6d5f0278480889b6e7018e9ae53ddc12e9b22c2a Mon Sep 17 00:00:00 2001
From: Ray <19251440+swjturay@users.noreply.github.com>
Date: Sat, 29 Aug 2026 17:25:33 +0800
Subject: [PATCH] fix(openai-ws): recycle stale idle pool connections

---
 backend/internal/service/openai_ws_pool.go    | 18 +++++++++++--
 .../internal/service/openai_ws_pool_test.go   | 26 +++++++++++++++++++
 2 files changed, 42 insertions(+), 2 deletions(-)

diff --git a/backend/internal/service/openai_ws_pool.go b/backend/internal/service/openai_ws_pool.go
index 2c62b6b4c0..5a04392399 100644
--- a/backend/internal/service/openai_ws_pool.go
+++ b/backend/internal/service/openai_ws_pool.go
@@ -18,8 +18,11 @@ import (
 )
 
 const (
-	openAIWSConnMaxAge             = 60 * time.Minute
-	openAIWSConnHealthCheckIdle    = 90 * time.Second
+	openAIWSConnMaxAge          = 60 * time.Minute
+	openAIWSConnHealthCheckIdle = 90 * time.Second
+	// coder/websocket cannot consume pong frames without a reader. Recycle
+	// unsupported idle sockets before the upstream keepalive window expires.
+	openAIWSConnIdleRecycleAfter   = 90 * time.Second
 	openAIWSConnHealthCheckTO      = 2 * time.Second
 	openAIWSConnPrewarmExtraDelay  = 2 * time.Second
 	openAIWSAcquireCleanupInterval = 3 * time.Second
@@ -1353,6 +1356,17 @@ func (p *openAIWSConnPool) cleanupAccountLocked(ap *openAIWSAccountPool, now tim
 		if p.isConnPinnedLocked(ap, id) {
 			continue
 		}
+		if !conn.isLeased() && conn.waiters.Load() == 0 &&
+			!conn.supportsIdlePingWithoutReader() &&
+			conn.idleDuration(now) >= openAIWSConnIdleRecycleAfter {
+			delete(ap.conns, id)
+			if len(ap.pinnedConns) > 0 {
+				delete(ap.pinnedConns, id)
+			}
+			evicted = append(evicted, conn)
+			p.metrics.scaleDownTotal.Add(1)
+			continue
+		}
 		if maxAge > 0 && !conn.isLeased() && conn.age(now) > maxAge {
 			delete(ap.conns, id)
 			if len(ap.pinnedConns) > 0 {
diff --git a/backend/internal/service/openai_ws_pool_test.go b/backend/internal/service/openai_ws_pool_test.go
index ada18af8d3..1a4480dccb 100644
--- a/backend/internal/service/openai_ws_pool_test.go
+++ b/backend/internal/service/openai_ws_pool_test.go
@@ -1307,6 +1307,32 @@ func TestOpenAIWSConnPool_BackgroundCleanupSweep_WithoutAcquire(t *testing.T) {
 	require.False(t, exists, "后台清理应在无新 acquire 时也回收过期连接")
 }
 
+func TestOpenAIWSConnPool_RecyclesUnsupportedIdlePingConnection(t *testing.T) {
+	cfg := &config.Config{}
+	cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 2
+	cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 2
+	pool := &openAIWSConnPool{cfg: cfg}
+
+	accountID := int64(303)
+	ap := &openAIWSAccountPool{conns: make(map[string]*openAIWSConn)}
+	conn := newOpenAIWSConn("stale_unsupported_idle_ping", accountID, &openAIWSIdlePingUnsupportedConn{}, nil)
+	conn.lastUsedNano.Store(time.Now().Add(-openAIWSConnIdleRecycleAfter - time.Second).UnixNano())
+	ap.conns[conn.id] = conn
+	pool.accounts.Store(accountID, ap)
+
+	pool.runBackgroundCleanupSweep(time.Now())
+
+	ap.mu.Lock()
+	_, exists := ap.conns[conn.id]
+	ap.mu.Unlock()
+	require.False(t, exists, "不支持无 reader idle ping 的陈旧连接应被主动回收")
+	select {
+	case <-conn.closedCh:
+	default:
+		t.Fatal("被回收的陈旧连接应已关闭")
+	}
+}
+
 func TestOpenAIWSConnPool_BackgroundWorkerGuardBranches(t *testing.T) {
 	var nilPool *openAIWSConnPool
 	require.NotPanics(t, func() {