fix(FN-2048): prevent stale SSE reconnects after channel teardown
- Add a closed flag to sse-bus channel state so teardown permanently disables reconnect scheduling - Guard useTasks onReconnect callbacks against stale effect instances when toggling views - Add regression tests for sse-bus and useTasks to ensure closed/unmounted channels do not reconnect - Document the leak root cause and closed-flag pattern in .fusion/memory.md
This commit is contained in:
@@ -1233,3 +1233,61 @@ When a mission feature's implementation task completes, `MissionExecutionLoop.pr
|
||||
- **VALID_TRANSITIONS**: `"in-progress"` → `"done"` was added specifically for validation tasks (note in `types.ts`)
|
||||
|
||||
The error status from `parseValidationResult()` (empty response, invalid JSON, invalid status) is now handled in `processTaskOutcome()` alongside pass/fail/blocked.
|
||||
|
||||
## FN-2048: SSE Connection Leak — `closed` Flag Pattern in sse-bus
|
||||
|
||||
Rapid view transitions (e.g., Board↔Missions) caused zombie EventSource connections to accumulate, exhausting the browser's HTTP/1.1 connection pool (6 per origin) and blocking subsequent `fetchMissions` calls. Two fixes prevent this:
|
||||
|
||||
### 1. `closed` flag in sse-bus `Channel`
|
||||
|
||||
The `closeChannel` function sets `channel.closed = true` **before** closing the EventSource. This prevents `forceReconnect()` (triggered by error events) from scheduling a reconnect timer after the channel has already been torn down. Additionally, the reconnect timer callback itself checks `channel.closed` before calling `openChannel`.
|
||||
|
||||
```typescript
|
||||
interface Channel {
|
||||
// ...
|
||||
closed: boolean;
|
||||
}
|
||||
|
||||
function closeChannel(channel: Channel): void {
|
||||
channel.closed = true; // Set BEFORE es.close() to block synchronous reconnect
|
||||
if (channel.es) channel.es.close();
|
||||
// ...
|
||||
}
|
||||
|
||||
function forceReconnect(channel: Channel): void {
|
||||
// ...
|
||||
if (channel.closed) return; // Guard: do not schedule reconnect on teardown
|
||||
// ...
|
||||
channel.reconnectTimer = setTimeout(() => {
|
||||
if (channel.closed) return; // Guard: do not reopen after teardown
|
||||
// ...
|
||||
}, RECONNECT_DELAY_MS);
|
||||
}
|
||||
```
|
||||
|
||||
**Key insight**: The `closed` check at the top of `forceReconnect` prevents the timer from being set when `closeChannel` runs synchronously after an error event. Without this, the reconnect timer could be set after the channel was already marked for teardown, creating a zombie connection.
|
||||
|
||||
### 2. `active` flag in useTasks SSE effect
|
||||
|
||||
The SSE effect in `useTasks.ts` uses an `active` boolean to prevent stale `onReconnect` callbacks from firing after the effect has cleaned up (e.g., when `sseEnabled` flips to `false`):
|
||||
|
||||
```typescript
|
||||
useEffect(() => {
|
||||
let active = true;
|
||||
// ...
|
||||
return subscribeSse(url, {
|
||||
onReconnect: () => {
|
||||
if (!active) return; // Block stale callbacks after effect cleanup
|
||||
// ...
|
||||
},
|
||||
});
|
||||
return () => { active = false; }; // Runs after subscribeSse's cleanup
|
||||
}, [projectId, sseEnabled]);
|
||||
```
|
||||
|
||||
**Important**: The cleanup function that sets `active = false` must be returned **after** the `subscribeSse()` return value — not as a second `return` (which would be unreachable). The unsubscribe from `subscribeSse` is called first, then `active` is set to false.
|
||||
|
||||
### Files affected
|
||||
- `packages/dashboard/app/sse-bus.ts` — `closed` flag on Channel
|
||||
- `packages/dashboard/app/hooks/useTasks.ts` — `active` flag in SSE effect
|
||||
- Tests: `packages/dashboard/app/__tests__/sse-bus.test.ts`, `packages/dashboard/app/hooks/__tests__/useTasks.test.ts`
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
import { describe, it, expect, afterEach } from "vitest";
|
||||
import { describe, it, expect, afterEach, vi, beforeEach } from "vitest";
|
||||
import { MockEventSource } from "../../vitest.setup";
|
||||
import { subscribeSse, __resetSseBus, __sseBusChannelCount } from "../sse-bus";
|
||||
|
||||
@@ -129,4 +129,43 @@ describe("sse-bus", () => {
|
||||
expect(reconnects).toBe(1);
|
||||
unsub();
|
||||
});
|
||||
|
||||
it("does not set a reconnect timer after closeChannel is called", () => {
|
||||
vi.useFakeTimers();
|
||||
const url = "/api/events";
|
||||
const unsub = subscribeSse(url, { events: { "task:created": () => {} } });
|
||||
const es = MockEventSource.instances[0];
|
||||
|
||||
// Simulate an error which triggers forceReconnect (and schedules reconnect timer)
|
||||
es._emit("error");
|
||||
|
||||
// Immediately unsubscribe (triggers closeChannel which sets closed=true)
|
||||
unsub();
|
||||
|
||||
// Advance timers past RECONNECT_DELAY_MS (3 seconds)
|
||||
vi.advanceTimersByTime(4_000);
|
||||
|
||||
// No new EventSource should be created — the closed flag prevented reconnect
|
||||
expect(MockEventSource.instances).toHaveLength(1);
|
||||
vi.useRealTimers();
|
||||
});
|
||||
|
||||
it("does not leak channels on rapid subscribe/unsubscribe cycles", () => {
|
||||
const url = "/api/events";
|
||||
for (let i = 0; i < 5; i++) {
|
||||
const unsub = subscribeSse(url, { events: { "task:created": () => {} } });
|
||||
unsub();
|
||||
}
|
||||
expect(__sseBusChannelCount()).toBe(0);
|
||||
// After 5 subscribe/unsubscribe cycles the channel has been opened and closed 5
|
||||
// times, creating 5 EventSource instances (channel is deleted from the map on
|
||||
// every unsubscribe). The key leak concern is zombie reconnect timers — verify
|
||||
// no additional instances are created when reconnect delay fires after teardown.
|
||||
const countBeforeTimers = MockEventSource.instances.length;
|
||||
vi.useFakeTimers();
|
||||
vi.advanceTimersByTime(4_000);
|
||||
vi.useRealTimers();
|
||||
// No new instances should be created by the (blocked) reconnect timer
|
||||
expect(MockEventSource.instances.length).toBe(countBeforeTimers);
|
||||
});
|
||||
});
|
||||
|
||||
@@ -1621,5 +1621,70 @@ describe("useTasks", () => {
|
||||
|
||||
expect(MockEventSource.instances.length).toBe(0);
|
||||
});
|
||||
|
||||
it("does not grow EventSource instances on repeated sseEnabled toggles", async () => {
|
||||
vi.useFakeTimers({ shouldAdvanceTime: true });
|
||||
const { rerender } = renderHook(
|
||||
({ sseEnabled }: { sseEnabled?: boolean }) => useTasks({ projectId: "test-project", sseEnabled }),
|
||||
{ initialProps: { sseEnabled: true } }
|
||||
);
|
||||
|
||||
await waitFor(() => {
|
||||
expect(MockEventSource.instances.length).toBeGreaterThanOrEqual(1);
|
||||
});
|
||||
|
||||
// Toggle false → true multiple times
|
||||
for (let i = 0; i < 3; i++) {
|
||||
await act(async () => {
|
||||
rerender({ sseEnabled: false });
|
||||
});
|
||||
await act(async () => {
|
||||
rerender({ sseEnabled: true });
|
||||
});
|
||||
}
|
||||
|
||||
const countAfterToggles = MockEventSource.instances.length;
|
||||
|
||||
// Advance fake timers — no pending reconnect timers should fire after teardown
|
||||
vi.advanceTimersByTime(4_000);
|
||||
|
||||
// Count must not grow after timer advancement (the closed flag in sse-bus prevents
|
||||
// reconnect timers from creating zombie connections after channel teardown).
|
||||
expect(MockEventSource.instances.length).toBe(countAfterToggles);
|
||||
vi.useRealTimers();
|
||||
});
|
||||
|
||||
it("does not trigger onReconnect refetch after sseEnabled flips to false", async () => {
|
||||
vi.useFakeTimers({ shouldAdvanceTime: true });
|
||||
const { rerender } = renderHook(
|
||||
({ sseEnabled }: { sseEnabled?: boolean }) => useTasks({ projectId: "test-project", sseEnabled }),
|
||||
{ initialProps: { sseEnabled: true } }
|
||||
);
|
||||
|
||||
await waitFor(() => {
|
||||
expect(MockEventSource.instances.length).toBeGreaterThanOrEqual(1);
|
||||
});
|
||||
|
||||
const es = MockEventSource.instances[0];
|
||||
mockFetchTasks.mockClear();
|
||||
|
||||
// Simulate an error on the EventSource (triggers reconnect flow)
|
||||
act(() => {
|
||||
es._emit("error");
|
||||
});
|
||||
expect(mockFetchTasks).toHaveBeenCalledTimes(1); // onReconnect fires once during error
|
||||
|
||||
// Before the reconnect timer fires, flip sseEnabled to false
|
||||
await act(async () => {
|
||||
rerender({ sseEnabled: false });
|
||||
});
|
||||
|
||||
// Advance timers past RECONNECT_DELAY_MS (3 seconds)
|
||||
vi.advanceTimersByTime(4_000);
|
||||
|
||||
// No additional fetchTasks should have been called — active flag blocked it
|
||||
expect(mockFetchTasks).toHaveBeenCalledTimes(1);
|
||||
vi.useRealTimers();
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
@@ -166,6 +166,9 @@ export function useTasks(options?: UseTasksOptions) {
|
||||
const query = projectId ? `?projectId=${encodeURIComponent(projectId)}` : "";
|
||||
|
||||
const isStale = () => projectContextVersionRef.current !== contextVersionAtStart;
|
||||
// Guards against reconnect callbacks firing after the effect has cleaned up
|
||||
// (e.g., sseEnabled flipped to false during a pending reconnect timer in sse-bus).
|
||||
let active = true;
|
||||
|
||||
const handleCreated = (e: MessageEvent) => {
|
||||
if (isStale()) return;
|
||||
@@ -256,7 +259,7 @@ export function useTasks(options?: UseTasksOptions) {
|
||||
);
|
||||
};
|
||||
|
||||
return subscribeSse(`/api/events${query}`, {
|
||||
const unsubscribe = subscribeSse(`/api/events${query}`, {
|
||||
events: {
|
||||
"task:created": handleCreated,
|
||||
"task:moved": handleMoved,
|
||||
@@ -265,10 +268,15 @@ export function useTasks(options?: UseTasksOptions) {
|
||||
"task:merged": handleMerged,
|
||||
},
|
||||
onReconnect: () => {
|
||||
if (!active) return;
|
||||
if (isStale()) return;
|
||||
void refreshTasksRef.current();
|
||||
},
|
||||
});
|
||||
return () => {
|
||||
active = false;
|
||||
unsubscribe();
|
||||
};
|
||||
}, [projectId, sseEnabled]);
|
||||
|
||||
const createTask = useCallback(async (input: TaskCreateInput): Promise<Task> => {
|
||||
|
||||
@@ -28,6 +28,8 @@ interface Channel {
|
||||
heartbeatTimer: ReturnType<typeof setTimeout> | null;
|
||||
reconnectTimer: ReturnType<typeof setTimeout> | null;
|
||||
hasOpenedOnce: boolean;
|
||||
/** Set true at the start of closeChannel to prevent reconnect after teardown. */
|
||||
closed: boolean;
|
||||
}
|
||||
|
||||
const channels = new Map<string, Channel>();
|
||||
@@ -50,6 +52,7 @@ function forceReconnect(channel: Channel): void {
|
||||
}
|
||||
channel.nativeListeners.clear();
|
||||
|
||||
if (channel.closed) return;
|
||||
if (channel.subscribers.size === 0 || channel.reconnectTimer) return;
|
||||
|
||||
// A teardown means events may have been missed while the stream was
|
||||
@@ -59,12 +62,14 @@ function forceReconnect(channel: Channel): void {
|
||||
|
||||
channel.reconnectTimer = setTimeout(() => {
|
||||
channel.reconnectTimer = null;
|
||||
if (channel.closed) return;
|
||||
if (channel.subscribers.size > 0) openChannel(channel);
|
||||
}, RECONNECT_DELAY_MS);
|
||||
}
|
||||
|
||||
function openChannel(channel: Channel): void {
|
||||
if (channel.es) return;
|
||||
if (channel.closed) return;
|
||||
if (channel.reconnectTimer) {
|
||||
clearTimeout(channel.reconnectTimer);
|
||||
channel.reconnectTimer = null;
|
||||
@@ -123,6 +128,7 @@ function reattachNativeListeners(channel: Channel): void {
|
||||
}
|
||||
|
||||
function closeChannel(channel: Channel): void {
|
||||
channel.closed = true;
|
||||
if (channel.heartbeatTimer) clearTimeout(channel.heartbeatTimer);
|
||||
if (channel.reconnectTimer) clearTimeout(channel.reconnectTimer);
|
||||
if (channel.es) channel.es.close();
|
||||
@@ -158,6 +164,7 @@ export function subscribeSse(url: string, sub: SseSubscription = {}): () => void
|
||||
heartbeatTimer: null,
|
||||
reconnectTimer: null,
|
||||
hasOpenedOnce: false,
|
||||
closed: false,
|
||||
};
|
||||
channels.set(url, channel);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user