test: harden live release validation flakes

extensions/google/google.live.test.ts

@@ -13,6 +13,17 @@ const GOOGLE_API_KEY =
 const LIVE = isLiveTestEnabled() && GOOGLE_API_KEY.length > 0;
 const describeLive = LIVE ? describe : describe.skip;
 
+function isTransientGeminiSearchError(error: unknown): boolean {
+  if (!(error instanceof Error)) {
+    return false;
+  }
+  if (error.name === "AbortError") {
+    return true;
+  }
+  const message = error.message.toLowerCase();
+  return message.includes("timeout") || message.includes("aborted");
+}
+
 const registerGooglePlugin = () =>
   registerProviderPlugin({
     plugin,
@@ -87,10 +98,26 @@ describeLive("google plugin live", () => {
     const provider = createGeminiWebSearchProvider();
     const tool = provider.createTool?.({
       config: {},
-      searchConfig: { gemini: { apiKey: GOOGLE_API_KEY }, cacheTtlMinutes: 0 },
+      searchConfig: { gemini: { apiKey: GOOGLE_API_KEY }, cacheTtlMinutes: 0, timeoutSeconds: 90 },
     } as never);
 
-    const result = await tool?.execute({ query: "OpenClaw GitHub", count: 1 });
+    let result: { provider?: string; content?: unknown; citations?: unknown } | undefined;
+    let lastError: unknown;
+    for (let attempt = 0; attempt < 2; attempt += 1) {
+      try {
+        result = await tool?.execute({ query: "OpenClaw GitHub", count: 1 });
+        lastError = undefined;
+        break;
+      } catch (error) {
+        lastError = error;
+        if (!isTransientGeminiSearchError(error) || attempt === 1) {
+          throw error;
+        }
+      }
+    }
+    if (lastError) {
+      throw lastError;
+    }
 
     expect(result?.provider).toBe("gemini");
     expect(typeof result?.content).toBe("string");

scripts/lib/docker-build.sh

@@ -19,7 +19,7 @@ docker_build_on_missing_enabled() {
   [ "${OPENCLAW_TESTBOX:-0}" = "1" ]
 }
 
-docker_build_exec() {
+docker_build_command() {
   local build_cmd=(docker build)
   if [ "${OPENCLAW_DOCKER_BUILD_USE_BUILDX:-0}" = "1" ]; then
     build_cmd=(docker buildx build --load)
@@ -31,23 +31,66 @@ docker_build_exec() {
     fi
   fi
 
-  env DOCKER_BUILDKIT=1 "${build_cmd[@]}" "$@"
+  printf '%s\0' env DOCKER_BUILDKIT=1 "${build_cmd[@]}" "$@"
+}
+
+docker_build_transient_failure() {
+  local log_file="$1"
+  grep -Eqi \
+    'frontend grpc server closed unexpectedly|failed to dial gRPC|no active session|buildkit.*connection.*closed|rpc error: code = Unavailable' \
+    "$log_file"
+}
+
+docker_build_retry_count() {
+  local configured="${OPENCLAW_DOCKER_BUILD_RETRIES:-2}"
+  if [[ "$configured" =~ ^[0-9]+$ ]]; then
+    echo "$configured"
+    return 0
+  fi
+  echo 2
+}
+
+docker_build_with_retries() {
+  local label="$1"
+  shift
+  local retries
+  retries="$(docker_build_retry_count)"
+  local attempt=1
+  local max_attempts=$((retries + 1))
+  local log_file
+  local command=()
+  while IFS= read -r -d '' part; do
+    command+=("$part")
+  done < <(docker_build_command "$@")
+
+  while true; do
+    log_file="$(docker_e2e_run_log "$label")"
+    if "${command[@]}" >"$log_file" 2>&1; then
+      rm -f "$log_file"
+      return 0
+    fi
+
+    if [ "$attempt" -ge "$max_attempts" ] || ! docker_build_transient_failure "$log_file"; then
+      docker_e2e_print_log "$log_file"
+      rm -f "$log_file"
+      return 1
+    fi
+
+    echo "Docker build failed with a transient BuildKit transport error; retrying ($attempt/$retries)..." >&2
+    docker_e2e_print_log "$log_file"
+    rm -f "$log_file"
+    attempt=$((attempt + 1))
+    sleep "$attempt"
+  done
+}
+
+docker_build_exec() {
+  docker_build_with_retries docker-build "$@"
 }
 
 docker_build_run() {
   local label="$1"
   shift
 
-  local build_cmd=(docker build)
-  if [ "${OPENCLAW_DOCKER_BUILD_USE_BUILDX:-0}" = "1" ]; then
-    build_cmd=(docker buildx build --load)
-    if [ -n "${OPENCLAW_DOCKER_BUILD_CACHE_FROM:-}" ]; then
-      build_cmd+=(--cache-from "${OPENCLAW_DOCKER_BUILD_CACHE_FROM}")
-    fi
-    if [ -n "${OPENCLAW_DOCKER_BUILD_CACHE_TO:-}" ]; then
-      build_cmd+=(--cache-to "${OPENCLAW_DOCKER_BUILD_CACHE_TO}")
-    fi
-  fi
-
-  run_logged "$label" env DOCKER_BUILDKIT=1 "${build_cmd[@]}" "$@"
+  docker_build_with_retries "$label" "$@"
 }

test/scripts/docker-build-helper.test.ts

@@ -37,6 +37,9 @@ describe("docker build helper", () => {
     expect(helper).toContain("docker_build_exec()");
     expect(helper).toContain("docker_build_run()");
     expect(helper).toContain("docker buildx build --load");
+    expect(helper).toContain("docker_build_transient_failure()");
+    expect(helper).toContain("OPENCLAW_DOCKER_BUILD_RETRIES");
+    expect(helper).toContain("frontend grpc server closed unexpectedly");
   });
 
   it("keeps shell-script Docker builds behind the helper", () => {