diff --git a/deploy/kubernetes/production/database.yaml b/deploy/kubernetes/production/database.yaml
index 99e831e..8dd6c38 100644
--- a/deploy/kubernetes/production/database.yaml
+++ b/deploy/kubernetes/production/database.yaml
@@ -128,6 +128,13 @@ spec:
       shared_buffers: 512MB
       wal_compression: "on"
       wal_keep_size: 1GB
+      # CNPG defaults these cross-region liveness timers to 5 seconds. A short
+      # control-plane or pod-network stall can otherwise tear down a healthy
+      # WAL stream and make preferred synchronous replication fall back to
+      # async. Keep failure detection bounded without amplifying a transient
+      # multi-second stall into repeated reconnects.
+      wal_sender_timeout: 30s
+      wal_receiver_timeout: 30s
   monitoring:
     enablePodMonitor: false
   projectedVolumeTemplate:
diff --git a/docs/operations/k3s-ha-runbook.md b/docs/operations/k3s-ha-runbook.md
index aed8df7..e65fcbb 100644
--- a/docs/operations/k3s-ha-runbook.md
+++ b/docs/operations/k3s-ha-runbook.md
@@ -41,7 +41,7 @@ NodePort 公网入口。节点防火墙继续由原有 UFW 管理，接入脚本
 [生产同构验收模式与高媒体压力测试](production-acceptance.md)。
 
 - 健康双库：同步复制，RPO=0。
-- 单库降级：`dataDurability: preferred` 保持写入，`archive_timeout=5min`，灾难 RPO 目标不超过 5 分钟。生产 WAL 段为 16 MiB；禁止把该值降到 60 秒，否则仅空闲切段就要求跨地域链路持续承载至少 2.24 Mbit/s，会使同步副本在当前带宽下永久追不平。
+- 单库降级：`dataDurability: preferred` 保持写入，`archive_timeout=5min`，灾难 RPO 目标不超过 5 分钟。生产 WAL 段为 16 MiB；禁止把该值降到 60 秒，否则仅空闲切段就要求跨地域链路持续承载至少 2.24 Mbit/s，会使同步副本在当前带宽下永久追不平。生产同时显式设置 `wal_sender_timeout=30s` 和 `wal_receiver_timeout=30s`，避免 5 秒级控制面或 Pod 网络抖动反复拆除 WAL 流；该配置只降低瞬时抖动放大，不替代同步复制、六向链路和复制日志门禁。
 - 数据库故障 RTO：不超过 5 分钟。
 - 应用发布：香港验收后再滚动宁波，计划内零停机。
 - 公网入口：双 NGINX 加人工 AliDNS 切换。接入带健康检查的 GTM 前，不宣称公网入口自动高可用。
diff --git a/tests/release/manual-release-test.sh b/tests/release/manual-release-test.sh
index b5a7e13..d92448c 100755
--- a/tests/release/manual-release-test.sh
+++ b/tests/release/manual-release-test.sh
@@ -305,6 +305,8 @@ if grep -Fq 'archive_timeout: 60s' "$root/deploy/kubernetes/production/database.
   echo 'production archive timeout would exceed the certified cross-region bandwidth floor' >&2
   exit 1
 fi
+grep -Fq 'wal_sender_timeout: 30s' "$root/deploy/kubernetes/production/database.yaml"
+grep -Fq 'wal_receiver_timeout: 30s' "$root/deploy/kubernetes/production/database.yaml"
 
 "$root/tests/release/cluster-release-helper-test.sh" >/dev/null
 "$root/tests/release/production-acceptance-script-test.sh" >/dev/null