fix(FN-4122): use unique tmp filename for task.json atomic writes

Two TaskStore instances writing to the same task concurrently raced on a
shared task.json.tmp filename: one writer's rename consumed the tmp file,
the other ENOENTed. withTaskLock only serializes within a single store
instance, so cross-process writers (engine + dashboard server) were
unprotected.

Each write now uses task.json.<pid>.<uuid>.tmp and cleans up its own tmp
on rename failure. Adds a regression test that drives 40 concurrent
same-task updates across 4 TaskStore instances.

Fixes FN-4122, FN-4123, FN-4148.
This commit is contained in:
gsxdsm
2026-05-12 09:41:43 -07:00
parent 7a982a2a5c
commit 6124535759
3 changed files with 44 additions and 2 deletions

View File

@@ -0,0 +1,5 @@
---
"@runfusion/fusion": patch
---
Fix `ENOENT: ... rename 'task.json.tmp' -> 'task.json'` failures when two TaskStore instances (e.g. engine + dashboard server) write to the same task concurrently. The shared `task.json.tmp` filename caused one writer's `rename` to consume the tmp file and the other's to ENOENT. Each write now uses a unique tmp filename (`task.json.<pid>.<uuid>.tmp`) and cleans up its own tmp on rename failure.

View File

@@ -222,6 +222,29 @@ describe("TaskStore concurrent writes", () => {
}); });
}); });
it("FN-4122/FN-4123/FN-4148: concurrent same-task writes across store instances don't ENOENT on task.json.tmp", async () => {
// Reproducer for the in-review failure mode where two TaskStore instances
// (e.g. engine + dashboard server) wrote to the same task simultaneously.
// Both writers used a shared `task.json.tmp` filename: one rename consumed
// the tmp, the other ENOENTed because it was no longer there. Fix uses a
// unique tmp filename per write.
const task = await primary.createTask({ description: "Cross-instance same-task race" });
const writes = Array.from({ length: 40 }, (_, index) =>
stores[index % stores.length].updateTask(task.id, {
title: `Race title ${index}`,
}),
);
// None should reject with ENOENT on task.json.tmp.
const results = await Promise.allSettled(writes);
const rejections = results.filter((r): r is PromiseRejectedResult => r.status === "rejected");
expect(rejections.map((r) => (r.reason as Error).message)).toEqual([]);
const reloaded = await primary.getTask(task.id);
expect(reloaded.title).toMatch(/^Race title \d+$/);
});
it("moves different tasks concurrently without SQLITE_BUSY failures", async () => { it("moves different tasks concurrently without SQLITE_BUSY failures", async () => {
const tasks: Task[] = await Promise.all( const tasks: Task[] = await Promise.all(
Array.from({ length: 10 }, (_, index) => primary.createTask({ description: `Move task ${index}` })), Array.from({ length: 10 }, (_, index) => primary.createTask({ description: `Move task ${index}` })),

View File

@@ -1834,11 +1834,25 @@ export class TaskStore extends EventEmitter<TaskStoreEvents> {
private async writeTaskJsonFile(dir: string, task: Task): Promise<void> { private async writeTaskJsonFile(dir: string, task: Task): Promise<void> {
const taskJsonPath = join(dir, "task.json"); const taskJsonPath = join(dir, "task.json");
const tmpPath = join(dir, "task.json.tmp"); // Use a unique tmp filename per write so concurrent writers to the same task
// don't race on a shared `task.json.tmp` (one rename consumes it, the other
// ENOENTs). See FN-4122/FN-4123/FN-4148 for the reproducer.
const tmpPath = join(dir, `task.json.${process.pid}.${randomUUID()}.tmp`);
this.suppressWatcher(taskJsonPath); this.suppressWatcher(taskJsonPath);
await mkdir(dir, { recursive: true }); await mkdir(dir, { recursive: true });
await writeFile(tmpPath, JSON.stringify(task)); await writeFile(tmpPath, JSON.stringify(task));
await rename(tmpPath, taskJsonPath); try {
await rename(tmpPath, taskJsonPath);
} catch (err) {
// Best-effort cleanup of our tmp on rename failure so we don't leave
// orphaned `task.json.*.tmp` files behind.
try {
await unlink(tmpPath);
} catch {
// ignore — tmp may already be gone
}
throw err;
}
} }
/** /**