How to check if a cron job ran

Read the record the job leaves, not the scheduler's own log. Count runs and successes per job in the run table: a job that fired and threw shows runs 8, ok 0, and the scheduler still reports it as having run. Windows Task Scheduler states the same fact as Last Result: 3.

Why check this

A scheduled job runs when nobody is watching. The tester signing off a nightly export sees the next morning's data, by which time a failed run is eight hours old.

Run it on staging sign-off, after a deploy that changes a job's dependencies, and after a schedule change. It prevents this: a price sync fires on time, throws ECONNREFUSED 127.0.0.1:5432 on its first line, and counts as a run on every dashboard that counts runs.

A real system leaves a run row, a last-success timestamp, an exit code and an output file. This page reads all four, and keeps two facts apart: the scheduler fired, and the work completed.

Prerequisites

• Node 22. The scheduler uses node:sqlite, which prints an ExperimentalWarning on start.
• A SQLite client. The run below used sqlite3 3.50.6.
• Windows Task Scheduler for steps 7 and 8. See the schtasks reference.
• The target, saved as scheduler.mjs. Three jobs on a 5 second interval: export-csv writes a file, sync-prices throws, archive-logs returns from a guard and writes nothing. Each job inserts a row on start and updates it on finish.

import { DatabaseSync } from 'node:sqlite';
import { createServer } from 'node:http';
import { writeFileSync, rmSync, mkdirSync } from 'node:fs';

const INTERVAL_MS = 5000;
rmSync('jobs.db', { force: true });
rmSync('out', { recursive: true, force: true });
mkdirSync('out');

const db = new DatabaseSync('jobs.db');
db.exec(`CREATE TABLE job_run (
  id INTEGER PRIMARY KEY, job TEXT NOT NULL, started_at TEXT NOT NULL,
  finished_at TEXT, status TEXT NOT NULL, detail TEXT)`);
const start = db.prepare("INSERT INTO job_run (job, started_at, status) VALUES (?, ?, 'running')");
const finish = db.prepare('UPDATE job_run SET finished_at = ?, status = ?, detail = ? WHERE id = ?');
const now = () => new Date().toISOString().replace('T', ' ').slice(0, 19);

const jobs = {
  'export-csv': () => writeFileSync('out/export.csv', `id,amount\n1,${Date.now()}\n`),
  'sync-prices': () => { throw new Error('ECONNREFUSED 127.0.0.1:5432'); },
  'archive-logs': () => { const pending = 0; if (pending === 0) return; writeFileSync('out/archive.tar', ''); },
};

function runAll() {
  for (const [name, fn] of Object.entries(jobs)) {
    const id = start.run(name, now()).lastInsertRowid;
    try {
      fn();
      finish.run(now(), 'ok', null, id);
    } catch (e) {
      finish.run(now(), 'error', e.message, id);
    }
  }
}
setInterval(runAll, INTERVAL_MS);
runAll();

// The heartbeat a monitor reads: last successful finish per job, and its age.
createServer((req, res) => {
  const rows = db.prepare(`SELECT job, MAX(finished_at) AS last_success FROM job_run
    WHERE status = 'ok' GROUP BY job`).all();
  const body = Object.keys(jobs).map((job) => {
    const r = rows.find((x) => x.job === job);
    const age = r ? Math.round((Date.now() - Date.parse(r.last_success + 'Z')) / 1000) : null;
    return { job, last_success: r ? r.last_success : null, age_seconds: age, interval_seconds: INTERVAL_MS / 1000 };
  });
  res.writeHead(200, { 'content-type': 'application/json' });
  res.end(JSON.stringify(body) + '\n');
}).listen(8931, '127.0.0.1', () => console.log(`scheduler on 8931, every ${INTERVAL_MS / 1000} s, pid ${process.pid}`));

• Two Windows tasks, plus one created and never started.

schtasks /create /tn h2check-sync /tr "cmd /c exit 3" /sc once /st 23:59 /f

Steps

1. 1

   Step 1.

   Start the scheduler and keep the pid it prints.

   node scheduler.mjs
   

   (node:39848) ExperimentalWarning: SQLite is an experimental feature and might change at any time
   (Use `node --trace-warnings ...` to show where the warning was created)
   scheduler on 8931, every 5 s, pid 39848

   The pid stops it in step 9. A 5 second interval makes the page reproducible in two minutes.

2. 2

   Step 2.

   Ask the naive question first.

   sqlite3 -header -column jobs.db "SELECT job, COUNT(*) AS started_recently, MAX(started_at) AS last_start FROM job_run WHERE started_at > datetime('now', '-15 seconds') GROUP BY job;"
   

   job           started_recently  last_start
   ------------  ----------------  -------------------
   archive-logs  3                 2026-09-12 18:06:24
   export-csv    3                 2026-09-12 18:06:24
   sync-prices   3                 2026-09-12 18:06:24

   Three jobs, three starts each. Nothing here is wrong and nothing is useful: it is the answer a green dashboard rests on. Both datetime('now') and started_at are UTC.

3. 3

   Step 3.

   Count successes, not runs.

   sqlite3 -header -column jobs.db "SELECT job, COUNT(*) AS runs, SUM(status = 'ok') AS ok, MAX(CASE WHEN status = 'ok' THEN finished_at END) AS last_success, MAX(CASE WHEN status = 'error' THEN detail END) AS last_error FROM job_run GROUP BY job;"
   

   job           runs  ok  last_success         last_error
   ------------  ----  --  -------------------  ---------------------------
   archive-logs  8     8   2026-09-12 18:06:29
   export-csv    8     8   2026-09-12 18:06:29
   sync-prices   8     0                        ECONNREFUSED 127.0.0.1:5432

   sync-prices ran eight times and succeeded zero times. runs beside ok is the smallest query that tells the two facts apart, and last_error names the dependency that is down.

4. 4

   Step 4.

   Run the query a monitoring rule carries, and watch a job vanish.

   sqlite3 -header -column jobs.db "SELECT job, MAX(finished_at) AS last_success FROM job_run WHERE status = 'ok' GROUP BY job;"
   

   job           last_success
   ------------  -------------------
   archive-logs  2026-09-12 18:06:34
   export-csv    2026-09-12 18:06:34

   Two rows for three jobs. Filtering on success before grouping drops the job that never succeeded, so a staleness alert has nothing to fire against.

5. 5

   Step 5.

   Read the heartbeat the way a monitor does, over HTTP.

   curl -s http://127.0.0.1:8931/jobs
   

   [{"job":"export-csv","last_success":"2026-09-12 18:06:34","age_seconds":5,"interval_seconds":5},{"job":"sync-prices","last_success":null,"age_seconds":null,"interval_seconds":5},{"job":"archive-logs","last_success":"2026-09-12 18:06:34","age_seconds":5,"interval_seconds":5}]

   This endpoint walks the list of jobs that should exist, so sync-prices reports null instead of vanishing. Compare age_seconds against interval_seconds: 5 against 5 is one interval behind, the freshest possible.

6. 6

   Step 6.

   Check the artifact. An exit status is not an output file.

   ls -l --time-style=full-iso out/
   

   total 1
   -rw-r--r-- 1 khark 197121 26 2026-09-12 21:06:39.043345900 +0300 export.csv

   archive-logs reported ok eight times and there is no archive.tar. The file time carries +0300 while the run rows are UTC, so an unconverted comparison calls a fresh file three hours stale.

7. 7

   Step 7.

   Read the same fact from Windows Task Scheduler.

   schtasks /query /tn h2check-sync /fo LIST /v
   

   …
   TaskName:                             \h2check-sync
   Next Run Time:                        12.09.2026 23:59:00
   Status:                               Ready
   …
   Last Run Time:                        12.09.2026 21:03:44
   Last Result:                          3
   …
   Task To Run:                          cmd /c exit 3
   …

   Status: Ready means the task is not running now, not that it worked. Last Run Time is the scheduler firing; Last Result is the exit code of what it fired.

8. 8

   Step 8.

   List every task at once.

   Get-ScheduledTask -TaskName 'h2check-*' | Get-ScheduledTaskInfo | Format-Table TaskName, LastRunTime, LastTaskResult -AutoSize
   

   TaskName        LastRunTime         LastTaskResult
   --------        -----------         --------------
   h2check-missing 12.09.2026 21:04:04     2147942402
   h2check-archive 30.11.1999 0:00:00          267011
   h2check-export  12.09.2026 21:05:05              0
   h2check-sync    12.09.2026 21:03:03              3

   Four states in one table. h2check-missing points at a path that does not exist, so a run is recorded and nothing ran. The two readers also disagree on the clock: 21:03:03 here against 21:03:44 in step 7.

9. 9

   Step 9.

   Stop the scheduler by the pid from step 1, then ask again.

   powershell -Command "Stop-Process -Id 39848 -Force"
   curl -s http://127.0.0.1:8931/jobs; echo "curl exit $?"
   

   curl exit 7

   Exit 7 is a refused connection. A monitor reading only the heartbeat body cannot tell a failing job from an absent scheduler. Stop by pid: a stop by image name takes every node on the machine.

How to read the result

| What you see | What it means | What to do | | --- | --- | --- | | runs 8, ok 8 and a fresh last_success | The job fired and finished on every interval | Keep the pair in the check. One number cannot say this. | | runs 8, ok 0 with a last_error | The schedule works and the job does not | Fix the dependency in last_error, here a database on 127.0.0.1:5432. | | A job missing from the result entirely | The query filtered on success before grouping | Drive the query from the job list, as step 5 does. A missing row raises no alert. | | status ok and no output file | The job returned early and reported success | Assert on the artifact and its mtime as well as the status. | | Last Result: 267011 with a 1999 date | Windows has never started this task | Check the trigger, not the job. See Why is my cron job not running. | | Last Run Time recent, Last Result non-zero | The scheduler fired and the program failed or was not found | Decode the code. 2147942402 is 0x80070002, file not found. | | curl exit 7 on the heartbeat | Nothing is scheduling at all | Alert on the endpoint being absent, separately from what it reports. |

Common mistakes

What to check next

• Why is my cron job not running: no run row at all.
• How to check if a cron job is still running: a run that never released its slot.
• How to validate cron expression: the schedule you wrote against the one you meant.
• How to check if a queue worker is running: the same question for a worker that pulls.
• How to test graceful shutdown: a deploy during a run.

FAQ

How do I test a cron job without waiting for the schedule?

Trigger the entry on demand. schtasks /run /tn h2check-sync started the task here and set Last Result to 3 within four seconds. For a timer, shorten the interval: this run used 5 seconds.

How do I check that a cron job ran successfully, not only that it started?

Compare two numbers from one row. runs 8, ok 0 is a job the scheduler fired eight times that finished none of them. A check counting only starts calls it healthy.

How do I test a scheduled task on Windows?

Create it with schtasks /create, start it with schtasks /run, then read Last Run Time and Last Result with schtasks /query /fo LIST /v. Get-ScheduledTaskInfo reports both fields for several tasks at once.

Where do cron job logs live?

That depends on the daemon and the distribution, and it answers a different question. A log path says what the scheduler did; the run table says what your code did.

Verified