OfflineExecutor.dispose() never settles pendingTransactionPromises
Les mainteneurs répondent en général sous 1 jour
Personne n'a encore pris cette issue.
Évaluation
- Difficulté
- 2/5
- Temps estimé
- 1-3 heures
- Accessibilité débutants
- 78/100
- Type d'issue
- Bug
- Clarté
- Clairement spécifiée
- Activité
- Active
- Stack technique
- node.js, typescript
Piste de recherche
Commence par packages/offline-transactions/src/OfflineExecutor.ts, en particulier waitForTransactionCompletion, resolveTransaction/rejectTransaction et dispose(), comme indiqué dans les liens de l’issue. Suis la correspondance entre les deferreds en attente et les transactions, puis examine les tests de disposal et de transaction du paquet. Reproduis la séquence de commit offline, disposal et replay décrite dans l’issue ; le travail est terminé lorsque la transaction d’origine atteint un état résolu et que son isPersisted.promise ne reste plus bloquée.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Description
- I've validated the bug against the latest version of DB packages
Describe the bug
OfflineExecutor keeps a pendingTransactionPromises map of deferreds, created by waitForTransactionCompletion and settled by resolveTransaction / rejectTransaction. dispose() pauses the executor and tears down the collection refresh hooks, the online listener, leadership, and the online detector, but it never touches the map. Every deferred in it is abandoned, so a transaction committed on a disposed executor never leaves persisting, and its isPersisted.promise never settles.
To Reproduce
npm i @tanstack/[email protected] @tanstack/dband save the script below asrepro.mjs. It runs in plain Node: in-memory storage stands in for IndexedDB and is shared by both executors, and connectivity and leader election are stubbed.- Start an executor offline and commit a transaction. It persists to the outbox, and the mutation fn does not run.
- Call
dispose()on that executor and start a successor on the same storage, online. - The successor replays the row, runs the mutation fn once, and empties the outbox.
- See that the original transaction still reads
persistingand itsisPersisted.promiseis unsettled (node repro.mjs --dispose). The control run without the dispose resolves (node repro.mjs).
repro.mjs
import { createCollection, localOnlyCollectionOptions } from "@tanstack/db";
import { startOfflineExecutor } from "@tanstack/offline-transactions";
const withDispose = process.argv.includes("--dispose");
// Shared in-memory storage, standing in for IndexedDB across both executors.
const store = new Map();
const storage = {
get: async (k) => store.get(k) ?? null,
set: async (k, v) => void store.set(k, v),
delete: async (k) => void store.delete(k),
keys: async () => [...store.keys()],
clear: async () => void store.clear(),
};
// Always-leader election: no Web Locks in Node.
const leaderElection = {
requestLeadership: async () => true,
releaseLeadership: () => {},
isLeader: () => true,
onLeadershipChange: () => () => {},
};
// Manually toggled connectivity.
function makeDetector(initiallyOnline) {
let online = initiallyOnline;
const listeners = new Set();
return {
subscribe: (cb) => (listeners.add(cb), () => listeners.delete(cb)),
notifyOnline: () => listeners.forEach((cb) => cb()),
isOnline: () => online,
dispose: () => listeners.clear(),
goOnline() {
online = true;
listeners.forEach((cb) => cb());
},
};
}
const todos = createCollection(
localOnlyCollectionOptions({ id: "todos", getKey: (t) => t.id }),
);
const calls = [];
const mutationFns = {
syncTodos: async ({ transaction }) => {
calls.push(transaction.id);
},
};
const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
const settledWithin = (promise, ms) =>
Promise.race([
promise.then(
() => "resolved",
(e) => `rejected: ${e?.message}`,
),
sleep(ms).then(() => "STILL PENDING"),
]);
// 1. Start offline and commit a transaction; it persists to the outbox.
const detector1 = makeDetector(false);
const executor1 = startOfflineExecutor({
collections: { todos },
mutationFns,
storage,
leaderElection,
onlineDetector: detector1,
});
await executor1.waitForInit();
const tx = executor1.createOfflineTransaction({ mutationFnName: "syncTodos" });
// autoCommit defaults to true, so mutate() commits.
const transaction = tx.mutate(() => todos.insert({ id: "1", text: "hello" }));
const isPersisted = transaction.isPersisted.promise;
await sleep(100);
console.log("outbox rows after offline commit:", store.size);
console.log("mutationFn calls while offline:", calls.length);
let completer;
if (withDispose) {
// 2. Dispose (the recreate pattern) and start a successor on the same storage.
executor1.dispose();
const detector2 = makeDetector(true);
completer = startOfflineExecutor({
collections: { todos },
mutationFns,
storage,
leaderElection,
onlineDetector: detector2,
});
await completer.waitForInit();
} else {
// Control: no dispose, just come back online on the same executor.
detector1.goOnline();
completer = executor1;
}
// 3. Let the outbox drain.
await sleep(500);
console.log("mutationFn calls after replay:", calls.length);
console.log("outbox rows after replay:", store.size);
console.log("pending count on completing executor:", completer.getPendingCount());
// 4. The promises the first executor handed out.
console.log("transaction.state:", transaction.state);
console.log("isPersisted.promise:", await settledWithin(isPersisted, 2000));
process.exit(0);
Output:
== node repro.mjs (control: no dispose)
outbox rows after offline commit: 1
mutationFn calls while offline: 0
mutationFn calls after replay: 1
outbox rows after replay: 0
pending count on completing executor: 0
transaction.state: completed
isPersisted.promise: resolved
== node repro.mjs --dispose
outbox rows after offline commit: 1
mutationFn calls while offline: 0
mutationFn calls after replay: 1
outbox rows after replay: 0
pending count on completing executor: 0
transaction.state: persisting
isPersisted.promise: STILL PENDING
Expected behavior
dispose() settles every deferred in pendingTransactionPromises, so no caller is left awaiting a transaction from a disposed executor. Either of these would work:
- Reject each deferred with a disposal error the caller can treat as retriable.
- Expose the pending ids so a successor executor can re-bridge them, and resolve them when it completes the replayed transaction.
Screenshots
N/A. The repro's console output is above.
Desktop (please complete the following information):
- OS: Linux (kernel 6.18)
- Browser: N/A. Reproduced in Node v24.22.1. The executor code path is environment-independent.
- Version:
@tanstack/offline-transactions1.1.4,@tanstack/db0.15.2.dispose()is unchanged onmainat d2935d0.
Smartphone (please complete the following information):
- N/A
Additional context
I first hit this on 1.0.42 in an app whose outbox recreates its executor whenever a new collection registers. I work around it with a completion ledger that carries isPersisted settlement across the recreate, but an upstream fix would let me delete this workaround.
- Langage dominant
- TypeScript
- Étoiles
- 4k
- Forks
- 277
- Merge moyen
- 17 h 40 min
- PR mergées (30 j)
- 223
Préparer son environnement
- Aucun Dockerfile ni fichier Docker Compose
- Propose un modèle de pull request
- Lire le guide de contribution
Par où commencer
- Lisez l'issue en entier, puis le guide de contribution du projet.
- Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
- Forkez le dépôt et travaillez sur une branche.
- Ouvrez une pull request qui référence le numéro de l'issue.
Autres issues de TanStack/db
-
Difficulté 2/5 1-3 heures Accessibilité débutants 78/100
Les mainteneurs répondent en général sous 1 jour
-
Difficulté 2/5 1-3 heures Accessibilité débutants 82/100
Les mainteneurs répondent en général sous 1 jour
-
Garbage collection cleans up a collection while one of its transactions is still persistingPeut-être pris @Nabil372 l’a pris il y a 1 jour. Ouverte
Difficulté 3/5 1-2 jours Accessibilité débutants 35/100
Les mainteneurs répondent en général sous 1 jour
-
persistedCollectionOptions without a schema can't be passed to createCollection (follow-up to #1452)Ouverte
Difficulté 3/5 1-2 jours Accessibilité débutants 68/100
Les mainteneurs répondent en général sous 1 jour
-
Electric: scoped move-in snapshots can prematurely resolve awaitTxId and remove optimistic rowsOuverte
Difficulté 4/5 3-5 jours Accessibilité débutants 38/100
Les mainteneurs répondent en général sous 1 jour
Toutes les issues de TanStack/db
Issues similaires
-
type: bug
Difficulté 2/5 1-3 heures Accessibilité débutants 79/100
Les mainteneurs répondent en général sous 1 jour
-
0️⃣ phase/new
Difficulté 2/5 1-3 heures Accessibilité débutants 82/100
Les mainteneurs répondent en général sous 1 jour
-
Difficulté 2/5 1-3 heures Accessibilité débutants 68/100
-
Difficulté 1/5 Moins d'une heure Accessibilité débutants 70/100
microsoft/playwright#43640 ·
Les mainteneurs répondent en général sous 1 jour
-
Difficulté 2/5 1-3 heures Accessibilité débutants 70/100