chore: import upstream snapshot with attribution
CI / Migrate Dev DB (push) Has been skipped
CI / Detect Version (push) Has been cancelled
CI / Migrate DB (push) Has been cancelled
CI / Build Dev ECR (./docker/app.Dockerfile, ECR_APP) (push) Has been cancelled
CI / Build Dev ECR (./docker/db.Dockerfile, ECR_MIGRATIONS) (push) Has been cancelled
CI / Build Dev ECR (./docker/pii.Dockerfile, ECR_PII) (push) Has been cancelled
CI / Build Dev ECR (./docker/realtime.Dockerfile, ECR_REALTIME) (push) Has been cancelled
CI / Deploy Trigger.dev (Dev) (push) Has been cancelled
CI / Build AMD64 (./docker/app.Dockerfile, ECR_APP, ghcr.io/simstudioai/simstudio) (push) Has been cancelled
CI / Build AMD64 (./docker/db.Dockerfile, ECR_MIGRATIONS, ghcr.io/simstudioai/migrations) (push) Has been cancelled
CI / Build AMD64 (./docker/pii.Dockerfile, ECR_PII, ghcr.io/simstudioai/pii) (push) Has been cancelled
CI / Build AMD64 (./docker/realtime.Dockerfile, ECR_REALTIME, ghcr.io/simstudioai/realtime) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (./docker/app.Dockerfile, ghcr.io/simstudioai/simstudio) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (./docker/db.Dockerfile, ghcr.io/simstudioai/migrations) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (./docker/pii.Dockerfile, ghcr.io/simstudioai/pii) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (./docker/realtime.Dockerfile, ghcr.io/simstudioai/realtime) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/migrations) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/pii) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/realtime) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/simstudio) (push) Has been cancelled
CI / Check Docs Changes (push) Has been cancelled
CI / Process Docs (push) Has been cancelled
CI / Create GitHub Release (push) Has been cancelled
CI / Test and Build (push) Has been cancelled
Publish CLI Package / publish-npm (push) Has been cancelled
Publish Python SDK / publish-pypi (push) Has been cancelled
Publish TypeScript SDK / publish-npm (push) Has been cancelled

This commit is contained in:
wehub-resource-sync
2026-07-13 13:20:55 +08:00
commit d25d482dc2
13754 changed files with 4996608 additions and 0 deletions
+389
View File
@@ -0,0 +1,389 @@
/**
* @vitest-environment node
*/
import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
type OutboxRow = {
id: string
eventType: string
payload: unknown
status: 'pending' | 'processing' | 'completed' | 'dead_letter'
attempts: number
maxAttempts: number
availableAt: Date
lockedAt: Date | null
lastError: string | null
createdAt: Date
processedAt: Date | null
}
// Hoisted mock state — all tests manipulate these directly.
const { state, mockDb } = vi.hoisted(() => {
const state = {
// Rows returned from the FOR UPDATE SKIP LOCKED select in claimBatch.
claimedRows: [] as OutboxRow[],
// Whether the terminal update (lease CAS) should report a match.
leaseHeld: true,
// IDs the reaper's UPDATE should return (simulates stuck `processing` rows).
reapedRowIds: [] as string[],
// Everything written (for assertions).
inserts: [] as Array<{ values: unknown }>,
updates: [] as Array<{ set: Record<string, unknown>; where?: unknown }>,
}
const makeUpdateChain = () => {
const row: { set: Record<string, unknown>; where?: unknown } = { set: {} }
const chain: Record<string, unknown> = {}
chain.set = vi.fn((s: Record<string, unknown>) => {
row.set = s
return chain
})
chain.where = vi.fn((w: unknown) => {
row.where = w
state.updates.push(row)
return chain
})
chain.returning = vi.fn(async () => {
// Terminal UPDATE (lease CAS): has `attempts` + `availableAt`
// on retry, or explicit completed/dead_letter. Reaper path sets
// status='pending' without attempts/availableAt.
const isReaperUpdate =
row.set.status === 'pending' && !('attempts' in row.set) && !('availableAt' in row.set)
if (isReaperUpdate) {
return state.reapedRowIds.map((id) => ({ id }))
}
if (
row.set.status === 'completed' ||
row.set.status === 'dead_letter' ||
(row.set.status === 'pending' && 'attempts' in row.set && 'availableAt' in row.set) ||
(!('status' in row.set) && 'attempts' in row.set && 'lockedAt' in row.set)
) {
return state.leaseHeld ? [{ id: 'evt-1' }] : []
}
return []
})
return chain
}
const makeSelectChain = () => {
const chain: Record<string, unknown> = {}
const self = () => chain
chain.from = vi.fn(self)
chain.where = vi.fn(self)
chain.orderBy = vi.fn(self)
chain.limit = vi.fn(self)
chain.for = vi.fn(async () => state.claimedRows.splice(0, 1))
return chain
}
const mockDb = {
insert: vi.fn(() => {
const chain: Record<string, unknown> = {}
chain.values = vi.fn(async (v: unknown) => {
state.inserts.push({ values: v })
})
return chain
}),
update: vi.fn(() => makeUpdateChain()),
select: vi.fn(() => makeSelectChain()),
transaction: vi.fn(async (fn: (tx: unknown) => Promise<unknown>) => fn(mockDb)),
}
return { state, mockDb }
})
vi.mock('@sim/db', () => ({ db: mockDb }))
vi.mock('@sim/db/schema', () => ({
outboxEvent: {
id: 'outbox_event.id',
eventType: 'outbox_event.event_type',
payload: 'outbox_event.payload',
status: 'outbox_event.status',
attempts: 'outbox_event.attempts',
maxAttempts: 'outbox_event.max_attempts',
availableAt: 'outbox_event.available_at',
lockedAt: 'outbox_event.locked_at',
lastError: 'outbox_event.last_error',
createdAt: 'outbox_event.created_at',
processedAt: 'outbox_event.processed_at',
$inferSelect: {} as OutboxRow,
},
}))
vi.mock('@sim/logger', () => ({
createLogger: () => ({ info: vi.fn(), warn: vi.fn(), error: vi.fn(), debug: vi.fn() }),
}))
vi.mock('drizzle-orm', () => ({
and: vi.fn((...args) => ({ _op: 'and', args })),
asc: vi.fn((col) => ({ _op: 'asc', col })),
eq: vi.fn((col, val) => ({ _op: 'eq', col, val })),
inArray: vi.fn((col, vals) => ({ _op: 'inArray', col, vals })),
lte: vi.fn((col, val) => ({ _op: 'lte', col, val })),
}))
vi.mock('@sim/utils/id', () => ({
generateId: vi.fn(() => 'test-event-id'),
}))
import { enqueueOutboxEvent, processOutboxEvents } from './service'
function makePendingRow(overrides: Partial<OutboxRow> = {}): OutboxRow {
return {
id: 'evt-1',
eventType: 'test.event',
payload: { foo: 'bar' },
status: 'pending',
attempts: 0,
maxAttempts: 10,
availableAt: new Date(Date.now() - 1000),
lockedAt: null,
lastError: null,
createdAt: new Date(Date.now() - 5000),
processedAt: null,
...overrides,
}
}
function resetState() {
state.claimedRows = []
state.leaseHeld = true
state.reapedRowIds = []
state.inserts.length = 0
state.updates.length = 0
}
describe('enqueueOutboxEvent', () => {
beforeEach(() => {
vi.clearAllMocks()
resetState()
})
it('inserts a row with the given event type and payload', async () => {
const id = await enqueueOutboxEvent(mockDb, 'test.event', { foo: 'bar' })
expect(id).toBe('test-event-id')
expect(state.inserts[0].values).toMatchObject({
id: 'test-event-id',
eventType: 'test.event',
payload: { foo: 'bar' },
maxAttempts: 10,
})
})
it('respects maxAttempts override', async () => {
await enqueueOutboxEvent(mockDb, 'test.event', {}, { maxAttempts: 3 })
expect(state.inserts[0].values).toMatchObject({ maxAttempts: 3 })
})
it('respects availableAt override for delayed processing', async () => {
const future = new Date(Date.now() + 60_000)
await enqueueOutboxEvent(mockDb, 'test.event', {}, { availableAt: future })
expect((state.inserts[0].values as { availableAt: Date }).availableAt).toBe(future)
})
})
describe('processOutboxEvents — empty / no handler', () => {
beforeEach(() => {
vi.clearAllMocks()
resetState()
})
it('returns zero counts when no events are due', async () => {
const result = await processOutboxEvents({})
expect(result).toEqual({
processed: 0,
retried: 0,
deadLettered: 0,
leaseLost: 0,
reaped: 0,
})
})
it('dead-letters events with no registered handler', async () => {
state.claimedRows = [makePendingRow({ eventType: 'unknown.event' })]
const result = await processOutboxEvents({})
expect(result.deadLettered).toBe(1)
const terminal = state.updates.find((u) => u.set.status === 'dead_letter')
expect(terminal).toBeDefined()
expect(terminal?.set.lastError).toMatch(/No handler registered/)
})
})
describe('processOutboxEvents — handler success and retry', () => {
beforeEach(() => {
vi.clearAllMocks()
resetState()
})
it('transitions to completed on handler success and passes context to handler', async () => {
const handlerCalls: Array<{ payload: unknown; eventId: string; attempts: number }> = []
const handler = vi.fn(async (payload: unknown, ctx: { eventId: string; attempts: number }) => {
handlerCalls.push({ payload, eventId: ctx.eventId, attempts: ctx.attempts })
})
state.claimedRows = [makePendingRow()]
const result = await processOutboxEvents({ 'test.event': handler })
expect(result.processed).toBe(1)
expect(handlerCalls).toEqual([{ payload: { foo: 'bar' }, eventId: 'evt-1', attempts: 0 }])
const completeUpdate = state.updates.find((u) => u.set.status === 'completed')
expect(completeUpdate).toBeDefined()
})
it('schedules retry with exponential backoff on handler failure below maxAttempts', async () => {
const handler = vi.fn(async () => {
throw new Error('transient failure')
})
state.claimedRows = [makePendingRow({ attempts: 2 })]
const before = Date.now()
const result = await processOutboxEvents({ 'test.event': handler })
expect(result.retried).toBe(1)
const retryUpdate = state.updates.find((u) => u.set.status === 'pending' && 'attempts' in u.set)
expect(retryUpdate).toBeDefined()
expect(retryUpdate?.set.attempts).toBe(3)
expect(retryUpdate?.set.lastError).toBe('transient failure')
// Backoff after nextAttempts=3: 1000 * 2^3 = 8000ms
const scheduledAt = retryUpdate?.set.availableAt as Date
expect(scheduledAt.getTime()).toBeGreaterThan(before + 7500)
expect(scheduledAt.getTime()).toBeLessThan(before + 10_000)
})
it('dead-letters on failure when attempts reaches maxAttempts', async () => {
const handler = vi.fn(async () => {
throw new Error('permanent failure')
})
state.claimedRows = [makePendingRow({ attempts: 9, maxAttempts: 10 })]
const result = await processOutboxEvents({ 'test.event': handler })
expect(result.deadLettered).toBe(1)
const deadUpdate = state.updates.find((u) => u.set.status === 'dead_letter')
expect(deadUpdate).toBeDefined()
expect(deadUpdate?.set.attempts).toBe(10)
expect(deadUpdate?.set.lastError).toBe('permanent failure')
})
it('caps exponential backoff at 1 hour', async () => {
const handler = vi.fn(async () => {
throw new Error('transient')
})
state.claimedRows = [makePendingRow({ attempts: 20, maxAttempts: 100 })]
const before = Date.now()
await processOutboxEvents({ 'test.event': handler })
const retryUpdate = state.updates.find((u) => u.set.status === 'pending' && 'attempts' in u.set)
expect(retryUpdate).toBeDefined()
const scheduledAt = retryUpdate?.set.availableAt as Date
// 1hr = 3,600,000ms
expect(scheduledAt.getTime()).toBeLessThan(before + 3_600_000 + 1000)
expect(scheduledAt.getTime()).toBeGreaterThan(before + 3_599_000)
})
})
describe('processOutboxEvents — lease CAS / reaper race', () => {
beforeEach(() => {
vi.clearAllMocks()
resetState()
})
it('reports leaseLost when completion UPDATE affects zero rows', async () => {
const handler = vi.fn(async () => {
// "succeeds" but terminal write will fail the lease CAS
})
state.claimedRows = [makePendingRow()]
state.leaseHeld = false
const result = await processOutboxEvents({ 'test.event': handler })
expect(result.leaseLost).toBe(1)
expect(result.processed).toBe(0)
})
it('reports leaseLost on retry-schedule UPDATE when row was reclaimed', async () => {
const handler = vi.fn(async () => {
throw new Error('transient')
})
state.claimedRows = [makePendingRow({ attempts: 2 })]
state.leaseHeld = false
const result = await processOutboxEvents({ 'test.event': handler })
expect(result.leaseLost).toBe(1)
expect(result.retried).toBe(0)
})
})
describe('processOutboxEvents — handler timeout', () => {
beforeEach(() => {
vi.clearAllMocks()
resetState()
vi.useFakeTimers()
})
afterEach(() => {
vi.useRealTimers()
})
it('times out a stuck handler without releasing it for overlapping retry', async () => {
const neverResolves = vi.fn(() => new Promise<void>(() => {}))
state.claimedRows = [makePendingRow({ attempts: 0 })]
const promise = processOutboxEvents({ 'test.event': neverResolves })
// Must exceed DEFAULT_HANDLER_TIMEOUT_MS (90s).
await vi.advanceTimersByTimeAsync(90 * 1000 + 1)
const result = await promise
expect(result.leaseLost).toBe(1)
const timeoutUpdate = state.updates.find(
(u) => !('status' in u.set) && 'attempts' in u.set && 'lockedAt' in u.set
)
expect(timeoutUpdate?.set.attempts).toBe(1)
expect(timeoutUpdate?.set.lastError).toMatch(/timed out/)
})
})
describe('processOutboxEvents — reaper recovery', () => {
beforeEach(() => {
vi.clearAllMocks()
resetState()
})
it('reaps stuck processing rows back to pending and reports count', async () => {
state.reapedRowIds = ['stuck-1', 'stuck-2', 'stuck-3']
const result = await processOutboxEvents({})
expect(result.reaped).toBe(3)
expect(result.processed).toBe(0)
// The reaper's UPDATE sets status='pending' with NO attempts / availableAt
// fields — that's how runHandler's retry update is distinguished from it.
const reaperUpdate = state.updates.find(
(u) => u.set.status === 'pending' && !('attempts' in u.set) && !('availableAt' in u.set)
)
expect(reaperUpdate).toBeDefined()
expect(reaperUpdate?.set.lockedAt).toBeNull()
})
it('returns zero reaped when no rows are stuck', async () => {
const result = await processOutboxEvents({})
expect(result.reaped).toBe(0)
})
})
+580
View File
@@ -0,0 +1,580 @@
import { db } from '@sim/db'
import { outboxEvent } from '@sim/db/schema'
import { createLogger } from '@sim/logger'
import { toError } from '@sim/utils/errors'
import { generateId } from '@sim/utils/id'
import { and, asc, eq, inArray, lte, sql } from 'drizzle-orm'
const logger = createLogger('OutboxService')
const DEFAULT_MAX_ATTEMPTS = 10
const STUCK_PROCESSING_THRESHOLD_MS = 10 * 60 * 1000 // 10 minutes
const MAX_BACKOFF_MS = 60 * 60 * 1000 // 1 hour
const BASE_BACKOFF_MS = 1000 // 1 second, doubled per attempt
// Kept below the serverless route `maxDuration` (120s) so our in-process
// timeout fires before the platform kills the invocation and leaves the
// row stranded in `processing` for the 10-minute reaper window. Also well
// under `STUCK_PROCESSING_THRESHOLD_MS` so the reaper cannot steal a row
// a worker is still actively processing.
const DEFAULT_HANDLER_TIMEOUT_MS = 90 * 1000 // 90 seconds
class OutboxHandlerTimeoutError extends Error {
constructor(timeoutMs: number) {
super(`Outbox handler timed out after ${timeoutMs}ms`)
this.name = 'OutboxHandlerTimeoutError'
}
}
/**
* Context passed to every outbox handler. Use `eventId` as the Stripe
* (or any external service) idempotency key so that handler retries
* collapse on the external side: a second execution of the same event
* lands on the same Stripe invoice id / charge id rather than creating
* a duplicate. The outbox lease CAS handles our DB side.
*/
interface OutboxEventContext {
eventId: string
eventType: string
/** How many times this event has been attempted (zero on first run). */
attempts: number
}
/**
* A handler invoked by the outbox worker for events of a given type.
* Throwing bumps `attempts` and schedules a retry via exponential
* backoff; a successful return transitions the event to `completed`.
*/
export type OutboxHandler<T = unknown> = (payload: T, context: OutboxEventContext) => Promise<void>
/**
* Map of `eventType` → handler. Register all handlers in one place
* and pass them to `processOutboxEvents`.
*/
export type OutboxHandlerRegistry = Record<string, OutboxHandler>
export interface EnqueueOptions {
/** Total attempts before the event moves to `dead_letter`. Default 10. */
maxAttempts?: number
/** Earliest time a worker may pick up this event. Default now. */
availableAt?: Date
}
export interface ProcessOutboxResult {
processed: number
retried: number
deadLettered: number
leaseLost: number
reaped: number
}
export type ProcessSingleOutboxResult =
| 'completed'
| 'pending'
| 'dead_letter'
| 'lease_lost'
| 'not_found'
| 'processing'
/**
* Transactional outbox for reliable "DB write + external system" flows.
*
* Callers enqueue an event *inside* a `db.transaction` alongside the
* primary write; the event row commits or rolls back with the business
* data. A polling worker (invoked via the cron endpoint) claims pending
* rows with `SELECT ... FOR UPDATE SKIP LOCKED`, marks them as
* `processing`, runs the registered handler outside the transaction,
* and transitions the event to `completed` / `pending` (retry) /
* `dead_letter` (max attempts exceeded).
*
* Two-phase claim-then-process keeps external API calls out of DB
* transactions. A reaper at the top of each run reclaims `processing`
* rows whose worker died mid-operation (stale `lockedAt`).
*
* Enqueue must be called with a `tx` from `db.transaction` so atomicity
* with the primary write is preserved. `db` itself is also accepted but
* then the caller must guarantee the enqueue and the primary write share
* a transaction some other way (or none at all).
*/
export async function enqueueOutboxEvent<T>(
executor: Pick<typeof db, 'insert'>,
eventType: string,
payload: T,
options: EnqueueOptions = {}
): Promise<string> {
const id = generateId()
await executor.insert(outboxEvent).values({
id,
eventType,
payload: payload as never,
maxAttempts: options.maxAttempts ?? DEFAULT_MAX_ATTEMPTS,
availableAt: options.availableAt ?? new Date(),
})
logger.info('Enqueued outbox event', { id, eventType })
return id
}
/** Cap on how many dead-lettered rows a single reconciler scan materializes. */
const DEAD_LETTER_SCAN_LIMIT = 100
/**
* Return events currently in `dead_letter` for the given event types (capped at
* `DEAD_LETTER_SCAN_LIMIT`). Used by periodic reconcilers to surface stuck work
* that exhausted its retries and now needs operator attention — the cap keeps a
* runaway backlog from materializing unboundedly into memory each run.
*/
export async function findDeadLetteredEvents(
eventTypes: string[]
): Promise<(typeof outboxEvent.$inferSelect)[]> {
if (eventTypes.length === 0) return []
return db
.select()
.from(outboxEvent)
.where(and(eq(outboxEvent.status, 'dead_letter'), inArray(outboxEvent.eventType, eventTypes)))
.limit(DEAD_LETTER_SCAN_LIMIT)
}
/**
* True when an event of the given type whose JSON payload has
* `payload->>payloadKey === payloadValue` is still `pending` or `processing`.
* Lets a competing writer detect that a DB→external sync is already in flight
* for a subject and avoid clobbering the not-yet-pushed DB value.
*/
export async function hasInflightOutboxEvent(
eventType: string,
payloadKey: string,
payloadValue: string
): Promise<boolean> {
const [row] = await db
.select({ id: outboxEvent.id })
.from(outboxEvent)
.where(
and(
eq(outboxEvent.eventType, eventType),
inArray(outboxEvent.status, ['pending', 'processing']),
sql`${outboxEvent.payload} ->> ${payloadKey} = ${payloadValue}`
)
)
.limit(1)
return Boolean(row)
}
/**
* Process one batch of outbox events. Safe to call concurrently from
* multiple workers — `SELECT FOR UPDATE SKIP LOCKED` serializes claims.
*/
export async function processOutboxEvents(
handlers: OutboxHandlerRegistry,
options: { batchSize?: number; maxRuntimeMs?: number; minRemainingMs?: number } = {}
): Promise<ProcessOutboxResult> {
const batchSize = options.batchSize ?? 10
const deadline = options.maxRuntimeMs ? Date.now() + options.maxRuntimeMs : undefined
const minRemainingMs = options.minRemainingMs ?? DEFAULT_HANDLER_TIMEOUT_MS + 5000
const reaped = await reapStuckProcessingRows()
let processed = 0
let retried = 0
let deadLettered = 0
let leaseLost = 0
for (let i = 0; i < batchSize; i++) {
if (deadline && Date.now() + minRemainingMs > deadline) break
const [event] = await claimBatch(1)
if (!event) break
const result = await runHandler(event, handlers)
if (result === 'completed') processed++
else if (result === 'dead_letter') deadLettered++
else if (result === 'lease_lost') leaseLost++
else retried++
}
return { processed, retried, deadLettered, leaseLost, reaped }
}
/**
* Process a specific outbox event immediately after its surrounding
* transaction commits. Safe to race with the cron worker: the claim uses
* `FOR UPDATE SKIP LOCKED`, and non-pending rows are left alone.
*/
export async function processOutboxEventById(
eventId: string,
handlers: OutboxHandlerRegistry
): Promise<ProcessSingleOutboxResult> {
const now = new Date()
const event = await db.transaction(async (tx) => {
const [row] = await tx
.select()
.from(outboxEvent)
.where(eq(outboxEvent.id, eventId))
.limit(1)
.for('update', { skipLocked: true })
if (!row) return null
if (row.status !== 'pending') return row.status as ProcessSingleOutboxResult
if (row.availableAt > now) return 'pending' as const
await tx
.update(outboxEvent)
.set({ status: 'processing', lockedAt: now })
.where(eq(outboxEvent.id, eventId))
return {
...row,
status: 'processing' as const,
lockedAt: now,
}
})
if (!event) {
const [current] = await db
.select({ status: outboxEvent.status })
.from(outboxEvent)
.where(eq(outboxEvent.id, eventId))
.limit(1)
return current ? (current.status as ProcessSingleOutboxResult) : 'not_found'
}
if (typeof event === 'string') return event
return runHandler(event, handlers)
}
/**
* Reaper: move `processing` rows whose worker died (stale `lockedAt`)
* back to `pending` so another worker can pick them up. Without this,
* a SIGKILL between claim and result-write would permanently strand
* the row in `processing`.
*/
async function reapStuckProcessingRows(): Promise<number> {
const stuckBefore = new Date(Date.now() - STUCK_PROCESSING_THRESHOLD_MS)
const result = await db
.update(outboxEvent)
.set({ status: 'pending', lockedAt: null })
.where(and(eq(outboxEvent.status, 'processing'), lte(outboxEvent.lockedAt, stuckBefore)))
.returning({ id: outboxEvent.id })
if (result.length > 0) {
logger.warn('Reaped stuck outbox processing rows', {
count: result.length,
thresholdMs: STUCK_PROCESSING_THRESHOLD_MS,
})
}
return result.length
}
/**
* Phase 1: claim a batch of due pending events.
*
* `SELECT ... FOR UPDATE SKIP LOCKED` atomically picks rows that no
* other worker is currently looking at. We then flip those rows to
* `processing` inside the same tx so the claim survives the lock
* release — the status change becomes the out-of-band mutual exclusion.
*/
async function claimBatch(batchSize: number): Promise<(typeof outboxEvent.$inferSelect)[]> {
const now = new Date()
return db.transaction(async (tx) => {
const rows = await tx
.select()
.from(outboxEvent)
.where(and(eq(outboxEvent.status, 'pending'), lte(outboxEvent.availableAt, now)))
.orderBy(asc(outboxEvent.createdAt))
.limit(batchSize)
.for('update', { skipLocked: true })
if (rows.length === 0) return []
await tx
.update(outboxEvent)
.set({ status: 'processing', lockedAt: now })
.where(
inArray(
outboxEvent.id,
rows.map((r) => r.id)
)
)
// Return rows with the claim state we just committed. `lockedAt`
// on this object is the authoritative lease timestamp used by the
// terminal-update lease CAS (see `runHandler`).
return rows.map((row) => ({
...row,
status: 'processing' as const,
lockedAt: now,
}))
})
}
/**
* Phase 2: invoke the handler for a claimed event, outside any DB
* transaction, then transition the row to its terminal or retry state.
*
* Every terminal UPDATE is guarded by a lease CAS (`WHERE status =
* 'processing' AND locked_at = event.lockedAt`). This defends against
* the "slow handler + reaper" race: if our handler takes longer than
* `STUCK_PROCESSING_THRESHOLD_MS`, the reaper will have reset the row
* to `pending` and another worker may have reclaimed it with a fresh
* `locked_at`. Our stale terminal write's WHERE clause won't match —
* rowCount is 0 — and we log+skip instead of clobbering the new lease.
*/
async function runHandler(
event: typeof outboxEvent.$inferSelect,
handlers: OutboxHandlerRegistry
): Promise<'completed' | 'pending' | 'dead_letter' | 'lease_lost'> {
const handler = handlers[event.eventType]
if (!handler) {
logger.error('No handler registered for outbox event type', {
eventId: event.id,
eventType: event.eventType,
})
await updateIfLeaseHeld(event, {
status: 'dead_letter',
lastError: `No handler registered for event type '${event.eventType}'`,
processedAt: new Date(),
lockedAt: null,
})
return 'dead_letter'
}
try {
await runHandlerWithTimeout(handler, event)
const updated = await updateIfLeaseHeld(event, {
status: 'completed',
processedAt: new Date(),
lockedAt: null,
})
if (!updated) {
logger.warn('Outbox event completion skipped — lease lost (reaped + reclaimed)', {
eventId: event.id,
eventType: event.eventType,
})
return 'lease_lost'
}
logger.info('Outbox event processed', {
eventId: event.id,
eventType: event.eventType,
attempts: event.attempts + 1,
})
return 'completed'
} catch (error) {
if (error instanceof OutboxHandlerTimeoutError) {
return recordTimedOutAttempt(event, error.message)
}
const nextAttempts = event.attempts + 1
const isDead = nextAttempts >= event.maxAttempts
const errMsg = toError(error).message
if (isDead) {
const updated = await updateIfLeaseHeld(event, {
attempts: nextAttempts,
status: 'dead_letter',
lastError: errMsg,
processedAt: new Date(),
lockedAt: null,
})
if (!updated) {
logger.warn('Outbox event dead-letter skipped — lease lost', {
eventId: event.id,
eventType: event.eventType,
})
return 'lease_lost'
}
logger.error('Outbox event dead-lettered after max attempts', {
eventId: event.id,
eventType: event.eventType,
attempts: nextAttempts,
error: errMsg,
})
return 'dead_letter'
}
return scheduleRetry(event, errMsg)
}
}
async function recordTimedOutAttempt(
event: typeof outboxEvent.$inferSelect,
errMsg: string
): Promise<'dead_letter' | 'lease_lost'> {
const nextAttempts = event.attempts + 1
const isDead = nextAttempts >= event.maxAttempts
if (isDead) {
const updated = await updateIfLeaseHeld(event, {
attempts: nextAttempts,
status: 'dead_letter',
lastError: errMsg,
processedAt: new Date(),
lockedAt: null,
})
if (!updated) return 'lease_lost'
logger.error('Outbox event dead-lettered after handler timeout max attempts', {
eventId: event.id,
eventType: event.eventType,
attempts: nextAttempts,
error: errMsg,
})
return 'dead_letter'
}
const updated = await updateProcessingIfLeaseHeld(event, {
attempts: nextAttempts,
lastError: errMsg,
lockedAt: new Date(),
})
if (!updated) return 'lease_lost'
logger.warn('Outbox event handler timed out; leaving lease for stuck-row reaper', {
eventId: event.id,
eventType: event.eventType,
attempts: nextAttempts,
reaperThresholdMs: STUCK_PROCESSING_THRESHOLD_MS,
error: errMsg,
})
return 'lease_lost'
}
async function scheduleRetry(
event: typeof outboxEvent.$inferSelect,
errMsg: string,
minimumBackoffMs = 0
): Promise<'pending' | 'dead_letter' | 'lease_lost'> {
const nextAttempts = event.attempts + 1
const isDead = nextAttempts >= event.maxAttempts
if (isDead) {
const updated = await updateIfLeaseHeld(event, {
attempts: nextAttempts,
status: 'dead_letter',
lastError: errMsg,
processedAt: new Date(),
lockedAt: null,
})
if (!updated) {
logger.warn('Outbox event dead-letter skipped — lease lost', {
eventId: event.id,
eventType: event.eventType,
})
return 'lease_lost'
}
logger.error('Outbox event dead-lettered after max attempts', {
eventId: event.id,
eventType: event.eventType,
attempts: nextAttempts,
error: errMsg,
})
return 'dead_letter'
}
const backoffMs = Math.max(
minimumBackoffMs,
Math.min(MAX_BACKOFF_MS, BASE_BACKOFF_MS * 2 ** nextAttempts)
)
const nextAvailableAt = new Date(Date.now() + backoffMs)
const updated = await updateIfLeaseHeld(event, {
attempts: nextAttempts,
status: 'pending',
lastError: errMsg,
availableAt: nextAvailableAt,
lockedAt: null,
})
if (!updated) {
logger.warn('Outbox event retry-schedule skipped — lease lost', {
eventId: event.id,
eventType: event.eventType,
})
return 'lease_lost'
}
logger.warn('Outbox event failed, scheduled retry', {
eventId: event.id,
eventType: event.eventType,
attempts: nextAttempts,
backoffMs,
nextAvailableAt: nextAvailableAt.toISOString(),
error: errMsg,
})
return 'pending'
}
async function updateProcessingIfLeaseHeld(
event: typeof outboxEvent.$inferSelect,
patch: {
attempts: number
lastError: string
lockedAt: Date
}
): Promise<boolean> {
const whereClauses = [eq(outboxEvent.id, event.id), eq(outboxEvent.status, 'processing')]
if (event.lockedAt) {
whereClauses.push(eq(outboxEvent.lockedAt, event.lockedAt))
}
const result = await db
.update(outboxEvent)
.set(patch)
.where(and(...whereClauses))
.returning({ id: outboxEvent.id })
return result.length > 0
}
function runHandlerWithTimeout(
handler: OutboxHandler,
event: typeof outboxEvent.$inferSelect,
timeoutMs: number = DEFAULT_HANDLER_TIMEOUT_MS
): Promise<void> {
const context: OutboxEventContext = {
eventId: event.id,
eventType: event.eventType,
attempts: event.attempts,
}
return new Promise((resolve, reject) => {
const timeout = setTimeout(() => {
reject(new OutboxHandlerTimeoutError(timeoutMs))
}, timeoutMs)
handler(event.payload, context)
.then((value) => {
clearTimeout(timeout)
resolve(value)
})
.catch((err) => {
clearTimeout(timeout)
reject(err)
})
})
}
/**
* Conditional terminal update scoped to the lease acquired at claim
* time. Returns true if the UPDATE affected a row, false if the row's
* lease was revoked (reaped, reclaimed by another worker). Callers
* treat `false` as a "lease lost" signal and skip without retrying —
* the newer owner is responsible for the row now.
*/
async function updateIfLeaseHeld(
event: typeof outboxEvent.$inferSelect,
patch: {
status: 'completed' | 'pending' | 'dead_letter'
attempts?: number
lastError?: string | null
availableAt?: Date
lockedAt: Date | null
processedAt?: Date | null
}
): Promise<boolean> {
const whereClauses = [eq(outboxEvent.id, event.id), eq(outboxEvent.status, 'processing')]
if (event.lockedAt) {
whereClauses.push(eq(outboxEvent.lockedAt, event.lockedAt))
}
const result = await db
.update(outboxEvent)
.set(patch)
.where(and(...whereClauses))
.returning({ id: outboxEvent.id })
return result.length > 0
}