core/saga-manager

Call store.runSaga(sagaFn) after store.init(); it dispatches startSaga and returns a cancel function that dispatches stopSaga. Overlapping calls for the same saga share one task via reference counting, so the saga stops only after every cancel function is invoked.

On an unhandled saga error, autoRestart records the crash, logs it, waits, and restarts the saga. getBackOffDelay computes min(1000 * 2^restarts, 10 minutes), starting at 1s and doubling to the cap, with restart counts decaying one per minute of stable runtime.

No. addCrash, clearCrashes, reducer state paths, and the reserved @internal_sagaManager name are package-owned internals. App code should only use the public Store facade such as store.runSaga and store.dispose.

No. clearCrashes(sagaName) deletes only that saga name's crash entry; reports for other saga names remain. Each entry holds serialized reports plus an omittedCount that grows when older reports are trimmed past MAX_SAGA_CRASH_REPORTS.

Use store.dispose() only for whole-store teardown, since it stops all Store-owned saga tasks. For normal per-mount saga lifetimes, call the cancel function returned by store.runSaga instead of disposing the entire Store context.