@empyria/restate 0.4.13 → 0.4.14

package/.github/dependabot.yml

@@ -5,7 +5,7 @@
 
 version: 2
 updates:
-    - package-ecosystem: '' # See documentation for possible values
-      directory: '/' # Location of package manifests
-      schedule:
-          interval: 'daily'
+  - package-ecosystem: "" # See documentation for possible values
+    directory: "/" # Location of package manifests
+    schedule:
+      interval: "daily"

package/README.md

@@ -118,6 +118,10 @@ endpoint, then create jobs with either:
   starts with the application.** The job has the stable `id` you give it, so calling it on every start never adds a second
   job: the same request changes nothing, and a changed request (another schedule, target or payload) replaces the pending
   run under the same ID. Call it once the endpoint has registered, e.g. from `setupRestate`'s `registrationCallbackFn`.
+  It also repairs a job whose chain broke: each run schedules the next, so a pending run that was killed, cancelled or
+  lost with a deleted deployment would stop the job for good. When the same request finds its pending run overdue by
+  more than `OVERDUE_GRACE_MS` (60 s), `ensure` schedules a fresh one; a replaced run that still arrives does nothing,
+  because a run only works when it is the job's current one.
 
 Cron expressions are `croner` patterns (five fields, or six with seconds for tests). The target handler is sent the
 `payload` as its input, or no input when there is none.

package/lib/Cron.js

@@ -27,6 +27,15 @@ export const RETRY_FOREVER = {
 	exponentiationFactor: 2,
 }
 
+/**
+ * How late a job's pending run may be before `ensure` treats it as lost and schedules a fresh one. A run is normally
+ * late only by the time the services were down; a run that is late by more than this, when the application starts and
+ * calls `ensure`, has most likely been killed, cancelled or lost with a deleted deployment, and nothing else would ever
+ * schedule the next one. Replacing a run that is merely slow is harmless: `execute` ignores a run that is no longer
+ * the job's current one.
+ */
+export const OVERDUE_GRACE_MS = 60_000
+
 export const cronJobInitiator = restate.service({
 	name: 'CronJobInitiator',
 	handlers: {
@@ -56,7 +65,8 @@ export const cronJob = restate.object({
 		 * code that starts with the application and cannot know whether the job was created before.
 		 *
 		 * - no job yet: schedules it, as `initiate` does;
-		 * - the same request already scheduled: changes nothing and returns the current state;
+		 * - the same request already scheduled: changes nothing and returns the current state, unless its pending run
+		 *   is overdue by more than {@link OVERDUE_GRACE_MS} (killed, cancelled or lost): then it schedules a fresh one;
 		 * - a different request (another schedule, target or payload): cancels the pending run and
 		 *   schedules the new request under the same job ID.
 		 */
@@ -64,7 +74,11 @@ export const cronJob = restate.object({
 			const wanted = normalizeRequest(request)
 			const jobState = await ctx.get(JOB_STATE_NAME)
 
-			if (jobState && sameRequest(jobState.request, wanted)) return jobState
+			if (jobState && sameRequest(jobState.request, wanted)) {
+				const dueAt = jobState.next_execution_at ?? Date.parse(jobState.next_execution_time)
+				const now = await ctx.date.now()
+				if (!(Number.isFinite(dueAt) && now > dueAt + OVERDUE_GRACE_MS)) return jobState
+			}
 			if (jobState) ctx.cancel(jobState.next_execution_id)
 
 			return await scheduleNextExecution(ctx, wanted)
@@ -74,6 +88,15 @@ export const cronJob = restate.object({
 			if (!jobState) {
 				throw new TerminalError('Job not found.')
 			}
+			// Only the job's current run works and schedules the next one. A run that `ensure` replaced (or that a
+			// changed schedule superseded) and that still arrives does nothing, so a job never runs as two chains.
+			const self = ctx.request?.().id
+			if (self && jobState.next_execution_id && self !== jobState.next_execution_id) {
+				ctx.console?.log?.(
+					`CronJob ${ctx.key}: run ${self} is not the current one (${jobState.next_execution_id}), skipped`,
+				)
+				return
+			}
 
 			// execute the task
 			const { service, method, key, payload } = jobState.request
@@ -156,6 +179,8 @@ const scheduleNextExecution = async (ctx, request) => {
 	const jobState = {
 		request,
 		next_execution_time: next.toString(),
+		// The same instant as a number, so `ensure` can tell an overdue run without parsing a date string.
+		next_execution_at: next.getTime(),
 		next_execution_id: await handle.invocationId,
 	}
 	ctx.set(JOB_STATE_NAME, jobState)

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@empyria/restate",
-	"version": "0.4.13",
+	"version": "0.4.14",
 	"description": "Restate.dev helpers for the Empyria nanoservice framework",
 	"license": "MIT",
 	"author": "Imre Fazekas <imre.fazekas@icloud.com>",

package/test/Cron.test.js

@@ -5,6 +5,7 @@ import {
 	ensureCronJob,
 	setJobStateName,
 	JOB_STATE_NAME,
+	OVERDUE_GRACE_MS,
 } from '../lib/Cron.js'
 import { TerminalError } from '@restatedev/restate-sdk'
 
@@ -34,7 +35,7 @@ describe('cronJobInitiator.create', () => {
 
 // Every cronJob handler that schedules a next run needs the same "date/objectSendClient/set"
 // wiring; this factory keeps the mocked context consistent across tests.
-function makeSchedulingCtx({ existing } = {}) {
+function makeSchedulingCtx({ existing, now = Date.now(), invocationId } = {}) {
 	const sets = []
 	const cancelled = []
 	return {
@@ -42,7 +43,8 @@ function makeSchedulingCtx({ existing } = {}) {
 		cancelled,
 		cancel: (invocationId) => cancelled.push(invocationId),
 		get: async () => existing,
-		date: { now: async () => Date.now() },
+		date: { now: async () => now },
+		...(invocationId ? { request: () => ({ id: invocationId }) } : {}),
 		key: 'job-1',
 		objectSendClient: () => ({
 			execute: () => ({ invocationId: Promise.resolve('exec-inv-1') }),
@@ -109,6 +111,37 @@ describe('cronJob.execute', () => {
 		expect(ctx.sets.length).toBe(1)
 	})
 
+	test("a run that is not the job's current one does nothing and schedules nothing", async () => {
+		const sent = []
+		const ctx = makeSchedulingCtx({
+			invocationId: 'inv-replaced',
+			existing: {
+				request: { service: 'Svc', method: 'do', cronExpression: '* * * * *' },
+				next_execution_id: 'inv-current',
+			},
+		})
+		ctx.genericSend = (opts) => sent.push(opts)
+		await cronJob.object.execute(ctx)
+		expect(sent).toEqual([])
+		expect(ctx.sets).toEqual([])
+	})
+
+	test('the current run works and schedules the next one', async () => {
+		const sent = []
+		const ctx = makeSchedulingCtx({
+			invocationId: 'inv-current',
+			existing: {
+				request: { service: 'Svc', method: 'do', cronExpression: '* * * * *' },
+				next_execution_id: 'inv-current',
+			},
+		})
+		ctx.genericSend = (opts) => sent.push(opts)
+		await cronJob.object.execute(ctx)
+		expect(sent.length).toBe(1)
+		expect(ctx.sets.length).toBe(1)
+		expect(ctx.sets[0][1].next_execution_at).toBeNumber()
+	})
+
 	test('throws when there is no job state to execute', async () => {
 		const ctx = makeSchedulingCtx({ existing: undefined })
 		await expect(cronJob.object.execute(ctx)).rejects.toThrow(TerminalError)
@@ -179,6 +212,42 @@ describe('cronJob.ensure', () => {
 		expect(ctx.cancelled).toEqual([])
 	})
 
+	test('the same request with a pending run that is long overdue (lost) schedules a fresh run', async () => {
+		const now = Date.parse('2026-10-05T08:00:00Z')
+		const existing = {
+			request,
+			next_execution_time: new Date(now - OVERDUE_GRACE_MS - 1000).toString(),
+			next_execution_at: now - OVERDUE_GRACE_MS - 1000,
+			next_execution_id: 'inv-killed',
+		}
+		const ctx = makeSchedulingCtx({ existing, now })
+		const state = await cronJob.object.ensure(ctx, request)
+		expect(state.next_execution_id).toBe('exec-inv-1')
+		expect(state.next_execution_at).toBeGreaterThan(now)
+		expect(ctx.cancelled).toEqual(['inv-killed'])
+	})
+
+	test('a state written before next_execution_at existed is judged by its date string', async () => {
+		const now = Date.parse('2026-10-05T08:00:00Z')
+		const existing = {
+			request,
+			next_execution_time: new Date(now - 10 * 60_000).toString(),
+			next_execution_id: 'inv-killed',
+		}
+		const ctx = makeSchedulingCtx({ existing, now })
+		expect((await cronJob.object.ensure(ctx, request)).next_execution_id).toBe('exec-inv-1')
+	})
+
+	test('a pending run that is due soon, or only a little late, is left alone', async () => {
+		const now = Date.parse('2026-10-05T08:00:00Z')
+		for (const dueAt of [now + 30_000, now - OVERDUE_GRACE_MS + 1000]) {
+			const existing = { request, next_execution_at: dueAt, next_execution_id: 'inv-pending' }
+			const ctx = makeSchedulingCtx({ existing, now })
+			expect(await cronJob.object.ensure(ctx, request)).toBe(existing)
+			expect(ctx.cancelled).toEqual([])
+		}
+	})
+
 	test('unknown fields and field order do not make a request "different"', async () => {
 		const existing = { request, next_execution_time: 'T', next_execution_id: 'inv-old' }
 		const ctx = makeSchedulingCtx({ existing })