fix(ci): docker_build_only (PR构建) 同步增加缓存损坏串行重建容错

scripts/ci/docker_build_only.sh

@@ -36,13 +36,44 @@ echo "Image tag:  ${IMAGE_TAG}"
 echo "Builder:    ${BUILDER_NAME}"
 echo ""
 
-docker buildx build \
-  $NO_CACHE_FLAG \
-  $BUILD_ARGS \
-  --cache-from "type=registry,ref=${CACHE_REF}" \
-  -f "${DOCKERFILE}" \
-  -t "${IMAGE_TAG}" \
-  .
+run_build() {
+  docker buildx build \
+    $NO_CACHE_FLAG \
+    $BUILD_ARGS \
+    --cache-from "type=registry,ref=${CACHE_REF}" \
+    -f "${DOCKERFILE}" \
+    -t "${IMAGE_TAG}" \
+    .
+}
+
+# PR 构建同样容错：检测到 builder 缓存损坏时，用 flock 串行重建共享 builder 后重试一次
+if ! build_output=$(run_build 2>&1); then
+  if echo "$build_output" | grep -qE "parent snapshot.*not found|snapshot.*does not exist|cache.*corrupt|failed to compute cache key|no such file or directory.*cache"; then
+    echo "$build_output"
+    echo "⚠️  builder 缓存异常，串行重建持久 builder 后重试..."
+    LOCK_FILE="/tmp/ci-builder-persist-rebuild.lock"
+    exec 9>"$LOCK_FILE"
+    flock -w 120 9 || echo "⚠️ 等待重建锁超时，直接重试 build"
+    if ! docker buildx inspect "$BUILDER_NAME" --bootstrap >/dev/null 2>&1; then
+      echo "🔨 锁内重建持久 builder..."
+      docker buildx rm "$BUILDER_NAME" >/dev/null 2>&1 || true
+      docker buildx create --name "$BUILDER_NAME" --driver docker-container \
+        --driver-opt network=host \
+        --buildkitd-flags "--allow-insecure-entitlement network.host" \
+        --platform linux/amd64
+      docker buildx use "$BUILDER_NAME"
+      docker buildx inspect "$BUILDER_NAME" --bootstrap
+    else
+      echo "✅ builder 已被其他并发 job 重建/恢复，直接复用"
+    fi
+    run_build
+  else
+    echo "$build_output"
+    exit 1
+  fi
+else
+  echo "$build_output"
+fi
 
 echo ""
 echo "PR build OK (build only, no output): ${IMAGE_TAG}"