Repair Resource Lifecycle Race Bugs

The issue is a classic resurrection race where the system lacks a terminal state (tombstone) to prevent late-arriving create confirmations from overwriting a completed deletion.

Diagnosis

• Identity: user-123
• Conflict: The delete operation finished locally, but the provider's create callback was still in flight.

Root Cause: The system treats the absence of a record as "not yet created" rather than "already deleted," allowing the late callback to insert a stale record.

Implementation: Tombstone Pattern

Replace the hard delete with a soft-delete tombstone that includes a deleted_at timestamp or a monotonically increasing version ID to discard late-arriving operations.

interface ResourceState {
  id: string;
  status: 'active' | 'deleted';
  version: number;
  deletedAt?: Date;
}

async function handleCreateCallback(providerResult: any) {
  const { id, version } = providerResult;
  
  // Atomic check-and-set or transaction
  const existing = await db.resources.find(id);
  
  if (existing && (existing.status === 'deleted' || existing.version >= version)) {
    console.log(`[Lifecycle] Ignoring late callback for ${id}. Resource already terminal.`);
    // Cleanup provider resource if necessary
    await provider.cleanup(id);
    return;
  }

  await db.resources.upsert({
    id,
    status: 'active',
    version,
    updatedAt: new Date()
  });
}

Deterministic Test Plan

it('should discard create callback if delete occurred during flight', async () => {
  const store = new MockStore();
  const lifecycle = new LifecycleManager(store);

  // 1. Start creation
  const createPromise = lifecycle.create('res-1');
  
  // 2. Immediate delete before callback returns
  await lifecycle.delete('res-1');
  
  // 3. Resolve late create callback
  await mockProvider.resolveCreate('res-1');

  // 4. Invariant: Store must still reflect 'deleted' state
  const finalState = await store.get('res-1');
  expect(finalState.status).toBe('deleted');
});

Next steps

1. Update your database schema to include a version or deleted_at column.
2. Modify the delete handler to write the tombstone instead of purging the row.
3. Add a background TTL task to purge expired tombstones after your max-retry window.