Skip to content
Source
Sign inSign up
TanStack/db

fix(electric): a stream error that onError retries does not fail waiting loads #2127

MergedKyleAMathews(GitHub profile, opens in a new tab)merged 14 commits intomain fromrefs/pull/2127/headmerged by KyleAMathews(GitHub profile, opens in a new tab)

Showing 8 changed files with 73 additions and 29 deletions (loading 5).

x
0 of 8 viewed, 8 left
7 additions and 0 deletions.changeset/electric-retry-is-success.mdAdded
View file
@@ -0,0 +1,7 @@
+---
+'@tanstack/electric-db-collection': minor
+---
+​
+A stream error that `onError` retries no longer fails the initial sync or the subset loads that are waiting for it. The collection stays `loading` while `onError` decides, and waiting loads settle when the retried stream delivers its snapshot. Before, the collection entered `error` and waiting loads rejected even when the retry succeeded, which could leave a joined or included live query in error for good. A stream error without `onError`, or one where `onError` gives up or throws, still fails the initial sync and its waiting loads. The initial sync also fails when the Electric client gives up by itself after `onError` asked for a retry, for example on a response without Electric headers or after its retry limit, and when the stream's external `signal` aborts before the first ready. Before, the collection stayed `loading` in both cases. When the initial sync fails, waiting loads and readiness waits such as `preload()` now reject with the same cause, including when `onError` throws. A status listener that cleans the Collection up while it enters `error` no longer makes a waiting load succeed without rows. Once the first up-to-date is committed, its persistence step alone decides the initial sync: the Collection becomes ready when that snapshot is accepted, even after a retry or a give-up, and fails with the persistence error when it is not.
+​
+Migration note: while `onError` decides whether to retry, the Collection now stays `loading`, and `preload()` and other readiness waits stay pending. Before, they entered `error` and rejected at once, even when the retry later succeeded. Code that showed a reconnecting state on `status === 'error'` during a retry should read `onError` itself instead.
1 addition and 1 deletiondocs/contributing/oracle-coverage.mdModified6 lines can hide code
View file
@@ -433,7 +433,7 @@
Code far to the right: this line has a long run of spaces or tabs before more code. Scroll right to read it.433433
| QueryRef operators and user-value boundaries | `packages/db/tests/query/subquery-user-value-oracle.test.ts`, `packages/db/tests/transactions.test.ts`, [repair record](oracle-reviews/2026-09-30-queryref-publication-repair.md) | A finite array/Set model checks the compiled output bag for a joined DISTINCT subquery with an outer WHERE. Live-query drivers check selected containers, no-select, functional, and predicate-literal user objects with IR-like fields, top-level and aggregate-subquery proxy-shaped fields, arrays of selected references, outer virtual-field filters on joined DISTINCT and nested aggregate QueryRefs, and a renamed no-select source. The transaction suite documents the existing development-browser duplicate-load guard. The bounded cases run in `test:oracles` or the full DB suite. Joined DISTINCT, nested and flat aggregate source-update histories pass. Ordered joined `findOne()` checks initial, singleton, empty, restored, and changed-join-key cuts; an unordered `findOne()` QueryRef on either side of a join checks default-key candidate deletion and reinsertion. A materialized joined `findOne()` supplies a separate receiving formulation. Other `singleResult` forms, join forms, schedules, and cross-copy value handling outside the duplicate-load guard remain unproved. |
Code far to the right: this line has a long run of spaces or tabs before more code. Scroll right to read it.434434
| Opaque backend pagination | [window oracle](https://github.com/TanStack/db/blob/main/packages/query-db-collection/tests/cursor-pagination.oracle.test.ts), [cache histories](https://github.com/TanStack/db/blob/main/packages/query-db-collection/tests/cursor-pagination.cache-oracle.test.ts), [cache publication](https://github.com/TanStack/db/blob/main/packages/query-db-collection/tests/cursor-pagination.publication-oracle.test.ts), [browser acquisition boundaries](https://github.com/TanStack/db/blob/main/packages/query-db-collection/tests/cursor-pagination.boundary-oracle.test.ts), [QueryCollection integration](https://github.com/TanStack/db/blob/main/packages/query-db-collection/tests/cursor-pagination-oracle.integration.test.ts) | Full filter/sort/slice reference, opaque token transport, actual Query cache expiry/invalidation/GC, forced refresh during growth, protocol failure publication/recovery, bounded slice work, nested cancellation/replacement, reader abort, browser retry defaults, manual-write cache isolation, and production window publications. Stable backend sequences; not snapshot guarantees for changing endpoints. Peek-ahead remains enabled. |
| Experimental no-peek continuation | [test-only live-window probe](https://github.com/TanStack/db/blob/main/packages/query-db-collection/tests/cursor-pagination-oracle.no-peek.integration.test.ts) | The candidate uses an authoritative fact only for the same complete output prefix and otherwise requests the extra row. It remains test-only; the shipped opaque-pagination owner keeps peek-ahead. See [experiment limits](https://github.com/TanStack/db/blob/main/packages/query-db-collection/tests/cursor-pagination/NO-PEEK-RESULTS.md). |
Code far to the right: this line has a long run of spaces or tabs before more code. Scroll right to read it.436
-| Electric and TrailBase | [Electric histories](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/tests/electric-oracle.property.test.ts), [recovery histories](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/tests/electric-recovery-oracle.test.ts), [held resume snapshots](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/tests/electric-resume-snapshot-races.test.ts), [PostgreSQL semantics](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/e2e/sql-predicate-semantics-oracle.e2e.test.ts), [TrailBase contract](https://github.com/TanStack/db/blob/main/packages/trailbase-db-collection/tests/ORACLE.md) | Installed SDK delivery/framing, independent predicates, exact subscription arguments, restart/reset lineage, held certification and durability races, source-order publication before durability, and late errors. The queued-presence property runs identical fixed/random generators plus isolated seed-and-path replay across insert, update, delete, and truncate callbacks. The recovery fixtures use a mocked ShapeStream; they do not establish live Electric-service framing or native persistence-host behavior. The TrailBase lifecycle oracle's early-demand lane checks that an on-demand load issued before the change listener runs does not fetch until the listener runs or cannot start, through a real live query and the subset boundary: 16 histories cross listen, subscribe rejection, reader failure, abort, and cleanup with a service change before the subscription; the adapter from before the wait fails all 16. Its page-event lane checks that a page read before a concurrent change event neither reverts nor resurrects that row, and its stream-end lane pins that the adapter does not resubscribe: 17 histories, 6 of which fail on the adapter before the page fence. A generated page lane (`trailbase.page-event`) crosses two-page loads, cursor and offset paging, overlapping on-demand loads, and up to four changes among page releases. A held-commit lane checks that a page never fails a load for an older event that core holds while a user mutation persists; deciding insert or update from visible rows fails it. After a generated history drains, the Collection holds exactly the rows a page returned or an event touched. Both page lanes deliver events for changes made before the list call at once, during the page, or after it applies, and check that the listener survives a late insert for a row a page already holds. |
Code far to the right: this line has a long run of spaces or tabs before more code. Scroll right to read it.436
+| Electric and TrailBase | [Electric histories](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/tests/electric-oracle.property.test.ts), [recovery histories](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/tests/electric-recovery-oracle.test.ts), [held resume snapshots](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/tests/electric-resume-snapshot-races.test.ts), [PostgreSQL semantics](https://github.com/TanStack/db/blob/main/packages/electric-db-collection/e2e/sql-predicate-semantics-oracle.e2e.test.ts), [TrailBase contract](https://github.com/TanStack/db/blob/main/packages/trailbase-db-collection/tests/ORACLE.md) | Installed SDK delivery/framing, independent predicates, exact subscription arguments, restart/reset lineage, held certification and durability races, source-order publication before durability, and late errors. The queued-presence property runs identical fixed/random generators plus isolated seed-and-path replay across insert, update, delete, and truncate callbacks. The recovery fixtures use a mocked ShapeStream; they do not establish live Electric-service framing or native persistence-host behavior. The TrailBase lifecycle oracle's early-demand lane checks that an on-demand load issued before the change listener runs does not fetch until the listener runs or cannot start, through a real live query and the subset boundary: 16 histories cross listen, subscribe rejection, reader failure, abort, and cleanup with a service change before the subscription; the adapter from before the wait fails all 16. Its page-event lane checks that a page read before a concurrent change event neither reverts nor resurrects that row, and its stream-end lane pins that the adapter does not resubscribe: 17 histories, 6 of which fail on the adapter before the page fence. A generated page lane (`trailbase.page-event`) crosses two-page loads, cursor and offset paging, overlapping on-demand loads, and up to four changes among page releases. A held-commit lane checks that a page never fails a load for an older event that core holds while a user mutation persists; deciding insert or update from visible rows fails it. After a generated history drains, the Collection holds exactly the rows a page returned or an event touched. Both page lanes deliver events for changes made before the list call at once, during the page, or after it applies, and check that the listener survives a late insert for a row a page already holds. The retry-decision property (`electric.retry-decision`) generates histories over full-log and progressive waits, with and without `onError`, crossing synchronous and asynchronous retries and give-ups, a throwing `onError`, a later error after a retry, an Electric client give-up, an external abort before or during a decision, new demands, the first up-to-date, a persisting optimistic transaction that holds the retried snapshot and later settles, and cleanup. The retry-decision grammar also crosses eager and changes-only sources, an external signal already aborted at sync start, and a status listener that cleans up inside the failure; it checks that a readiness wait rejects with the same cause as every waiting demand, and that such a cleanup leaves a rejected demand rejected. It also holds the acceptance of the first up-to-date commit across later errors, retries, give-ups, and aborts, and checks that the Collection becomes ready with the snapshot's rows when the commit is accepted, and fails with the acceptance's cause when it is not. The descriptor-isolation oracle holds that acceptance across a managed-cache rotation and checks that the retired session's acceptance settles nothing, whether it resolves or rejects, and that the replacement session's snapshot decides readiness. A progressive snapshot's truncate applies at once; a demand whose snapshot arrived before an external abort settles when its rows apply, and cleanup after the abort rejects it. An independent model checks the status, every demand's settlement and cause, and row visibility after each step: a source that still tries stays `loading`, and only a give-up fails its waits. `main` before #2127 and mutants that fail before the decision, skip the failure on a throw or give-up, or skip `markError` on a client give-up or an external abort each fail it at an assertion. An installed-client lane in the SDK delivery oracle drives the same model (`electric-retry-decision-model.ts`) through the real Electric client and HTTP: a retried 503, a response without Electric headers, and an exhausted 50-retry budget, in eager, full-log, and progressive modes, with a synchronous or asynchronous retry (18 cells). Skipping `markError` on the client's give-up fails the 12 give-up cells; failing before the decision, or `main` before #2127, fails 15; the 3 survivors give up in the same continuation that decides, so an early failure is not observable there. A mutant that settles a held demand on acceptance instead of visibility fails the held replays. Excluded: stream steps while `onError` decides or after a final failure, which the client cannot produce. |
Code far to the right: this line has a long run of spaces or tabs before more code. Scroll right to read it.437437
| PowerSync | [tests](https://github.com/TanStack/db/tree/main/packages/powersync-db-collection/tests), `tests/correctness-oracle.test.ts` | Applied receipt positions crossed with held peers, native SQLite/SDK and cleanup evidence. A metadata-bearing `Collection.update` passes through an observed SDK watcher callback to the public and SQLite rows and an exact queued CRUD patch. Held callback ordering and synthetic zero-field cases remain controlled-only, and remote upload has no receiving fixture here. Run the focused owner with the package's `test:oracles` command. A timeout mutant proves a progress failure, not every value assertion. |
| SQLite persistence and native hosts | [persisted histories](https://github.com/TanStack/db/blob/main/packages/db-sqlite-persistence-core/tests/persisted-oracle.test.ts), [reset/resume histories](https://github.com/TanStack/db/blob/main/packages/db-sqlite-persistence-core/tests/sqlite-core-adapter-oracle.test.ts), [dual-adapter resume snapshots](https://github.com/TanStack/db/blob/main/packages/db-sqlite-persistence-core/tests/sqlite-resume-snapshot-oracle.test.ts), [Browser composed-owner histories](https://github.com/TanStack/db/blob/main/packages/browser-db-sqlite-persistence/tests/per-collection-coordinator-oracle.test.ts), [Browser coordinator RPC](https://github.com/TanStack/db/blob/main/packages/browser-db-sqlite-persistence/tests/browser-coordinator-oracle.test.ts), [shared-driver fairness](https://github.com/TanStack/db/blob/main/packages/browser-db-sqlite-persistence/tests/shared-driver-fairness-oracle.test.ts), [driver contracts](https://github.com/TanStack/db/blob/main/packages/db-sqlite-persistence-core/tests/contracts/sqlite-driver-contract.ts), [Node shared-handle scheduling](https://github.com/TanStack/db/blob/main/packages/node-db-sqlite-persistence/tests/node-driver.test.ts), [OP-SQLite shared-handle scheduling](https://github.com/TanStack/db/blob/main/packages/react-native-db-sqlite-persistence/tests/op-sqlite-driver-oracle.test.ts), [browser OPFS lifecycle](https://github.com/TanStack/db/blob/main/packages/browser-db-sqlite-persistence/tests/opfs-page-lifecycle-oracle.test.ts), [worker diagnostics](https://github.com/TanStack/db/blob/main/packages/browser-db-sqlite-persistence/tests/opfs-worker-diagnostics-oracle.test.ts), [Electron IPC and composed owner](https://github.com/TanStack/db/blob/main/packages/electron-db-sqlite-persistence/tests/electron-ipc.test.ts), [113-law manifest](https://github.com/TanStack/db/blob/main/packages/db-collection-e2e/src/fixtures/persisted-conformance-manifest.ts) | Core cache/remote rejection/peer/reopen histories, atomic reset/resume lineage, key-set evidence, dual-adapter races, queued coordinator reload/gap notifications during startup, unscheduled startup/reset overlap, and exact driver results. The shared core-adapter contract checks custom local comparator dispatch for ascending, descending, and comparator-equal public-key ordering at `loadSubset` completion. Host packages register this contract, but the local sqlite3 CLI run does not prove native host execution. Browser composes public source commits with per-collection elected-owner routing and covers the complete committed-transaction wire partition through deterministic Node transport seams. Remote-subset histories distinguish logical demand, physical acquisitions, exact acquisition leases, and released replay tombstones. Electron composes source commits with a per-collection renderer owner, IPC adapter, real SQLite, and reopen checks. The shared-driver fairness owner checks K=1 complete-logical-hydrate scheduling with identical fixed, random, and seed-plus-path campaigns. It records public rows, raw dequeue reach, and a persist-first FIFO hostile control. Its Chromium OPFS fixture refines the provider boundary but does not establish a browser matrix, elapsed-time latency, unbounded eventuality, or multi-process coordination. Same-handle Node and OP-SQLite tests cover transaction admission. Controlled OPFS page/worker histories cover ownership and diagnostic-cause retention. The reset/resume owners use sqlite3 CLI and in-memory node:sqlite seams; they do not prove multi-process WAL, mobile/Tauri, or other native-device execution. Distinct database handles rely on SQLite lock admission rather than one in-process queue. React Native hosts without async-context propagation must use the transaction driver supplied to the callback for nested work. Fake workers and synthetic page events do not prove native handle release or real bfcache admission. The Browser composed seams are not real multi-context/OPFS-worker execution; the Electron harness is not an actual Electron process unless its explicit runtime-bridge mode runs. An ownerless elected node suppresses core routing, while a follower may route demand to the elected owner; host coordinators retry only classified transport or admission failures while demand remains retained. The manifest excludes progressive and move suites; registration and shim runs are not device execution. |
Code far to the right: this line has a long run of spaces or tabs before more code. Scroll right to read it.439439
| SQLite expression-index planning | [Node expression-index oracle](https://github.com/TanStack/db/blob/main/packages/node-db-sqlite-persistence/tests/expression-index-oracle.test.ts) | RFC #1659 invariant 8 owns identical persisted-index and runtime-expression shapes. Independent expected keys are checked against direct captured SQL, adapter results, and named-index plans. Generated BigInts use SQLite's signed range; one fixed case checks legacy oversized-value reads. Other limits: bounded unqualified JSON paths/scalars, Node BetterSQLite, and no null, arbitrary raw SQL, or native-host planning. Run the package's `test:oracles` campaign. |
65 additions and 28 deletionspackages/electric-db-collection/src/electric.tsModified
View file