From fcbac56989acd35e79aafacb3ed10c134bf5ad4f Mon Sep 17 00:00:00 2001
From: xiaoxia <dev@xiaoxiajianji.com>
Date: Tue, 18 Aug 2026 13:37:55 +0800
Subject: [PATCH] fix: add 25min timeout to docker buildx build to prevent
 infinite hang

---
 scripts/ci/docker_build_push.sh | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/scripts/ci/docker_build_push.sh b/scripts/ci/docker_build_push.sh
index f3ba4bb80..bff4325ca 100755
--- a/scripts/ci/docker_build_push.sh
+++ b/scripts/ci/docker_build_push.sh
@@ -3,6 +3,9 @@
 # 用法: docker_build_push.sh [--no-cache] <Dockerfile> <image_tag> <cache_ref> [build_arg...]
 set -eu
 
+# 单次 build 超时时间（秒），防止 docker buildx build 无限挂起
+BUILD_TIMEOUT=1500
+
 NO_CACHE_FLAG=""
 if [ "$1" = "--no-cache" ]; then
   NO_CACHE_FLAG="--no-cache"
@@ -43,7 +46,7 @@ build_with_cache_retry() {
     local build_output
     local exit_code
     set +e
-    build_output=$(docker buildx build \
+    build_output=$(timeout ${BUILD_TIMEOUT} docker buildx build \
       $NO_CACHE_FLAG \
       $BUILD_ARGS \
       --cache-from "type=local,src=${LOCAL_CACHE_DIR}" \
@@ -60,6 +63,12 @@ build_with_cache_retry() {
       echo "$build_output"
       return 0
     fi
+    # 超时退出（exit code 124）
+    if [ $exit_code -eq 124 ]; then
+      echo "❌ Docker build TIMEOUT after ${BUILD_TIMEOUT}s - build hung and was killed"
+      echo "$build_output" | tail -20
+      return $exit_code
+    fi
     # 检测到缓存损坏类错误，清掉本地缓存重试
     if echo "$build_output" | grep -qE "parent snapshot.*not found|snapshot.*does not exist|cache.*corrupt|failed to compute cache key"; then
       echo "$build_output"
@@ -68,7 +77,7 @@ build_with_cache_retry() {
       rm -rf "${LOCAL_CACHE_DIR}"
       mkdir -p "${LOCAL_CACHE_DIR}"
       # 清理buildx builder的内部snapshot状态
-      docker buildx prune -f -a >/dev/null 2>&1 || true
+      docker buildx prune -f -a > /dev/null 2>&1 || true
       attempt=$((attempt + 1))
     else
       # 非缓存类错误，直接输出并返回
@@ -78,7 +87,7 @@ build_with_cache_retry() {
   done
   # 重试完还是失败，不用本地缓存最后试一次（只从registry读）
   echo "⚠️  All cached attempts failed, building without local cache..."
-  docker buildx build \
+  timeout ${BUILD_TIMEOUT} docker buildx build \
     $NO_CACHE_FLAG \
     $BUILD_ARGS \
     --cache-from "type=registry,ref=${CACHE_REF}" \
@@ -93,6 +102,7 @@ build_with_cache_retry() {
 echo "=== Step 1: Build & push image (local cache + registry cache, with auto-repair) ==="
 echo "Local cache: ${LOCAL_CACHE_DIR}"
 echo "Registry cache: ${CACHE_REF}"
+echo "Build timeout: ${BUILD_TIMEOUT}s"
 echo ""
 
 build_with_cache_retry
-- 
2.54.0