fix(gateway): preserve lifecycle cleanup

samzong · samzong · commit f4f945f814a4 · 2026-05-22T16:14:29.000+08:00
Signed-off-by: samzong &lt;samzong.lu@gmail.com&gt;
diff --git a/src/gateway/server-chat.agent-events.test.ts b/src/gateway/server-chat.agent-events.test.ts
@@ -2050,6 +2050,20 @@ describe("agent event handler", () => {
     expect(fallbackPayload.runId).toBe("run-fallback-client");
     expect(fallbackPayload.data?.phase).toBe("fallback");
 
+    vi.advanceTimersByTime(100);
+
+    expect(chatRunState.registry.peek("run-fallback-retry")).toEqual({
+      sessionKey: "session-fallback",
+      clientRunId: "run-fallback-client",
+    });
+    expect(
+      chatBroadcastCalls(broadcast).some(
+        ([, payload]) => (payload as { state?: string }).state === "error",
+      ),
+    ).toBe(false);
+    expect(clearAgentRunContext).not.toHaveBeenCalled();
+    expect(agentRunSeq.get("run-fallback-retry")).toBe(3);
+
     emitLifecycleEnd(handler, "run-fallback-retry", 4);
 
     expect(
@@ -2172,6 +2186,52 @@ describe("agent event handler", () => {
     ).toBe(true);
   });
 
+  it("keeps deferred lifecycle-error cleanup across phase-less lifecycle events", () => {
+    vi.useFakeTimers();
+    const { broadcast, clearAgentRunContext, agentRunSeq, handler } = createHarness({
+      resolveSessionKeyForRun: () => "session-terminal-error",
+      lifecycleErrorRetryGraceMs: 100,
+    });
+    registerAgentRunContext("run-terminal-late-lifecycle", {
+      sessionKey: "session-terminal-error",
+    });
+
+    handler({
+      runId: "run-terminal-late-lifecycle",
+      seq: 1,
+      stream: "lifecycle",
+      ts: Date.now(),
+      data: { phase: "start" },
+    });
+    handler({
+      runId: "run-terminal-late-lifecycle",
+      seq: 2,
+      stream: "lifecycle",
+      ts: Date.now(),
+      data: { phase: "error", error: "request timed out" },
+    });
+    handler({
+      runId: "run-terminal-late-lifecycle",
+      seq: 3,
+      stream: "lifecycle",
+      ts: Date.now(),
+      data: { msg: "status update" },
+    });
+
+    vi.advanceTimersByTime(100);
+
+    const finalPayload = chatBroadcastCalls(broadcast).at(-1)?.[1] as {
+      state?: string;
+      runId?: string;
+      errorMessage?: string;
+    };
+    expect(finalPayload.state).toBe("error");
+    expect(finalPayload.runId).toBe("run-terminal-late-lifecycle");
+    expect(finalPayload.errorMessage).toContain("request timed out");
+    expect(clearAgentRunContext).toHaveBeenCalledWith("run-terminal-late-lifecycle");
+    expect(agentRunSeq.has("run-terminal-late-lifecycle")).toBe(false);
+  });
+
   it("cancels deferred lifecycle-error cleanup when the run restarts", () => {
     vi.useFakeTimers();
     const { broadcast, clearAgentRunContext, agentRunSeq, handler } = createHarness({
diff --git a/src/gateway/server-chat.ts b/src/gateway/server-chat.ts
@@ -814,7 +814,7 @@ export function createAgentEventHandler({
   return (evt: AgentEventPayload) => {
     const lifecyclePhase =
       evt.stream === "lifecycle" && typeof evt.data?.phase === "string" ? evt.data.phase : null;
-    if (lifecyclePhase === "start") {
+    if (lifecyclePhase !== null && lifecyclePhase !== "error") {
       clearPendingTerminalLifecycleError(evt.runId);
     }
 

Original file line number	Diff line number	Diff line change
`@@ -814,7 +814,7 @@ export function createAgentEventHandler({`
`814`	`814`	`return (evt: AgentEventPayload) => {`
`815`	`815`	`const lifecyclePhase =`
`816`	`816`	`evt.stream === "lifecycle" && typeof evt.data?.phase === "string" ? evt.data.phase : null;`
`817`		`- if (lifecyclePhase === "start") {`
	`817`	`+ if (lifecyclePhase !== null && lifecyclePhase !== "error") {`
`818`	`818`	`clearPendingTerminalLifecycleError(evt.runId);`
`819`	`819`	`}`
`820`	`820`