[query-core] [performances] `setQueryData` dispatches a `success` event (and triggers broadcast/persist) even when the new data is structurally identical to the cached data

Offen
  • Stimmt etwas nicht?
#12,443 0 Kommentare 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen

Maintainer antworten meist innerhalb von 1 Tag

@praveen-kumary arbeitet bereits daran.

Seit 30.10.2026.

  • #12462 von @praveen-kumary — offen

Bewertung

Schwierigkeit
4/5
Geschätzter Aufwand
3-5 Tage
Anfängerfreundlichkeit
52/100
Issue-Typ
Feature
Klarheit
Klar beschrieben
Aktivitätsstatus
Aktiv
Tech-Stack
typescript
Bereich
backend-api-design, performance

Rechercherichtung

Start by reading setData and replaceData in the query core, then trace how setQueryData and setQueriesData call them. Check the tests covering manual cache writes, cache events, and explicit updatedAt values. Done means structurally identical manual writes can skip success events while explicit timestamp updates and fetch writes retain their intended behavior.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

Describe the bug

queryClient.setQueryData always writes the new data through Query#setData, which dispatches a success cache event, increments state.dataUpdateCount and refreshes timestamps — even when the new data is structurally identical to what is already cached.

Since v5, setData already runs a deep comparison on every write for structural sharing (replaceData), so the equality is already computed — but it is only used to preserve references, not to skip the write.

Minimal repro (no framework needed, @tanstack/query-core 5.105.0):

import { QueryClient } from '@tanstack/query-core';

const client = new QueryClient();

let events = 0;
client.getQueryCache().subscribe((cacheEvent) => {
  if (cacheEvent.type === 'event' && cacheEvent.event.type === 'success') events++;
});

const base = { id: '1', name: 'item', meta: { tag: 'a', n: 1 } };

client.setQueryData(['item', '1'], base);

// Same content, new reference — the typical upsert pattern used by cache-sync helpers:
for (let i = 0; i < 10; i++) {
  client.setQueryData(['item', '1'], (old) => ({ ...old, ...JSON.parse(JSON.stringify(base)) }));
}

console.log(events);                            // 11 — ten extra 'success' events for data that never changed
console.log(client.getQueryCache().find({ queryKey: ['item', '1'] })!.state.dataUpdateCount); // 11

Expected: a manual write whose result is structurally identical to the current data should not dispatch a cache event (or at least should be able to).

Why it matters

The no-op write is invisible to components — because the structural-sharing result is referentially identical to the old data, observers do not re-render — so the cost never shows up as a visible bug. But every dispatched success event is still fully processed by:

  • @tanstack/query-broadcast-client-experimental — on every success event it structured-clones the query's full state and posts it on the channel.
  • @tanstack/query-persist-client-core with manual persistQueryByKey (the pattern recommended for manual cache updates, see #6310) — every call writes the full state to storage, and an IndexedDB put structured-clones its value synchronously on the main thread.

So an application that upserts a parent entity plus its N children into the cache per page transition (a common normalized-cache pattern) pays N state clones + N cross-tab posts per transition even when nothing changed.

Measured in an application with that pattern (one page transition entering an edit form that upserts ~10 cached entities, warm cache):

no guard no-op skip
success events dispatched 609 193
full-state writes to IndexedDB 218 16

The 416 / 202 eliminated operations were all structurally no-op writes. (We currently work around this at the application level with a deep-equality pre-check in a setQueryData wrapper plus a persistQueryByKey dedup keyed on dataUpdateCount; an in-library fix would cover all call sites at once.)

Suggested fix

setData already computes the shared result, and when the whole tree is structurally identical, replaceData returns the previous reference itself — so the short-circuit is a zero-cost reference comparison:

setData(newData, options) {
    const data = replaceData(this.state.data, newData, this.options);
    if (options?.manual && data === this.state.data) return data; // no-op manual write: skip the dispatch
    this.#dispatch({ data, type: 'success', dataUpdatedAt: options?.updatedAt, manual: options?.manual });
    return data;
}

Scoping to manual: true (writes from setQueryData / setQueriesData) leaves the fetch path untouched.

Behavior change to flag

  • dataUpdatedAt / dataUpdateCount would not bump on a skipped no-op write. For manual writes this seems correct (nothing was updated), but it is an observable change.
  • A call that passes an explicit updatedAt option should probably still dispatch, since the caller explicitly asked for a timestamp change.

Versions

"@tanstack/eslint-plugin-query": "5.105.0",
"@tanstack/query-broadcast-client-experimental": "5.105.0",
"@tanstack/query-persist-client-core": "5.105.0",
"@tanstack/svelte-query": "6.3.6",
"@tanstack/svelte-query-devtools": "6.3.6",
Vorherrschende Sprache
TypeScript
Sterne
50.5k
Forks
4.2k
Ø Merge
7 Std. 24 Min.
Gemergte PRs (30 T.)
570

Entwicklungsumgebung

Erste Schritte

  1. Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
  3. Forken Sie das Repository und arbeiten Sie in einem Branch.
  4. Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus TanStack/query

Alle Issues in TanStack/query

Ähnliche Issues

Weitere Issues zu TypeScript

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.