fix: wait for qa gateway restart boundary

extensions/qa-lab/src/gateway-child.test.ts

@@ -328,6 +328,32 @@ describe("buildQaRuntimeEnv", () => {
     expect(__testing.isRetryableGatewayCallError("permission denied")).toBe(false);
   });
 
+  it("waits for a fresh in-process restart boundary after the current log offset", async () => {
+    let logs = "old restart mode: in-process restart\n";
+    const offset = logs.length;
+    const wait = __testing.waitForQaGatewayRestartBoundary({
+      logs: () => logs,
+      offset,
+      pollMs: 1,
+      timeoutMs: 100,
+    });
+
+    logs += "signal SIGUSR1 received\nrestart mode: in-process restart\n";
+
+    await expect(wait).resolves.toBeUndefined();
+  });
+
+  it("times out when a SIGUSR1 restart never reaches the boundary", async () => {
+    await expect(
+      __testing.waitForQaGatewayRestartBoundary({
+        logs: () => "signal SIGUSR1 received\n",
+        offset: 0,
+        pollMs: 1,
+        timeoutMs: 1,
+      }),
+    ).rejects.toThrow("qa gateway child did not reach restart boundary");
+  });
+
   it("stages a live Anthropic setup-token profile for isolated QA workers", async () => {
     const stateDir = await mkdtemp(path.join(os.tmpdir(), "qa-setup-token-state-"));
     cleanups.push(async () => {

extensions/qa-lab/src/gateway-child.ts

@@ -268,6 +268,24 @@ async function fetchLocalGatewayHealth(params: {
   }
 }
 
+async function waitForQaGatewayRestartBoundary(params: {
+  logs: () => string;
+  offset: number;
+  pollMs?: number;
+  timeoutMs?: number;
+}) {
+  const timeoutMs = params.timeoutMs ?? 30_000;
+  const pollMs = params.pollMs ?? 100;
+  const startedAt = Date.now();
+  while (Date.now() - startedAt < timeoutMs) {
+    if (params.logs().slice(params.offset).includes("restart mode:")) {
+      return;
+    }
+    await sleep(pollMs);
+  }
+  throw new Error(`qa gateway child did not reach restart boundary within ${timeoutMs}ms`);
+}
+
 export const __testing = {
   assertQaArtifactDirWithinRepo,
   buildQaRuntimeEnv,
@@ -283,6 +301,7 @@ export const __testing = {
   stageQaLiveAnthropicSetupToken,
   stageQaMockAuthProfiles,
   resolveQaLiveCliAuthEnv,
+  waitForQaGatewayRestartBoundary,
   resolveQaOwnerPluginIdsForProviderIds,
   resolveQaBundledPluginSourceDir,
   resolveQaRuntimeHostVersion,
@@ -816,7 +835,20 @@ export async function startQaGatewayChild(params: {
         if (!activeChild.pid) {
           throw new Error("qa gateway child has no pid");
         }
+        const restartLogOffset = logs().length;
         process.kill(activeChild.pid, signal);
+        if (signal === "SIGUSR1") {
+          await waitForQaGatewayRestartBoundary({
+            logs,
+            offset: restartLogOffset,
+          });
+          await waitForGatewayReady({
+            baseUrl,
+            logs,
+            child: activeChild,
+            timeoutMs: 120_000,
+          });
+        }
       },
       async restartAfterStateMutation(
         mutateState: (context: QaGatewayChildStateMutationContext) => Promise<void>,