# How to run a cron job in the cloud Give a scheduler a cron expression, a time zone and a command, and it starts the command on a machine at each time. That machine can be one you keep running, a CI service's scheduled workflow, or a managed scheduler that starts a fresh machine for each run. They differ in three ways that matter more than the syntax. A run can fail with nobody watching. A run can still be going when the next time comes. And daylight saving time can move your job by an hour. **On Runtime (withruntime.com), a job starts a fresh Linux machine at each time of its cron expression, in the time zone you name. It runs your command, keeps the exit code and output, and bills only while the command runs.** Nothing stays up between runs. Each run is a new machine, so one run's files and processes never reach the next. ## Where a cron job can run | Where it runs | What you keep running | What a missed or failed run does | | --------------------------------- | -------------------------------------------- | ----------------------------------------------------------------------------------- | | `crontab` on a server you run | The server, all month, and its updates | Nothing tells you. Output goes to local mail or is lost | | A CI service's scheduled workflow | Nothing, but the repository must stay active | GitHub may delay a scheduled run under load, and disables schedules in an idle public repository after 60 days | | A managed scheduler with machines | Nothing | The run's state, exit code and output are kept, and a failed run can be retried | `crontab` is the right tool when the job must touch that one server's own disk. For anything else, a server that waits all day for a two-minute job is a cost and a machine to patch. ## Schedule a job on Runtime The schedule below runs at 06:30 London time on weekdays. The command runs without a shell unless you give it one, so it starts with `bash -lc`. ```ts import { Runtime } from "withruntime"; const runtime = new Runtime(); const job = await runtime.jobs.create({ name: `daily-report-${Date.now()}`, schedule: { cron: "30 6 * * 1-5", timezone: "Europe/London" }, // 06:30 London time, weekdays command: ["bash", "-lc", "date -u +%FT%TZ && echo report sent"], timeoutSeconds: 600, retry: { maxAttempts: 3, backoffSeconds: 60 }, }); try { console.log(job.state, new Date(job.nextRunAt!).toISOString()); // active, the next weekday 06:30 London await runtime.jobs.trigger(job.id); // one run now, besides the schedule let run; for (let i = 0; i < 240 && !run?.endedAt; i++) { await new Promise((r) => setTimeout(r, 500)); [run] = (await runtime.jobs.runs(job.id, { limit: 1, order: "newest" })).data; } const logs = await runtime.jobs.logs(run!.id); console.log(run!.state, run!.exitCode, JSON.stringify(logs).includes("report sent")); // succeeded 0 true if (run!.exitCode !== 0) throw new Error(JSON.stringify(run)); } finally { await runtime.jobs.cancel(job.id); } ``` The same job from the command line: ```bash no-run runtime jobs create daily-report --cron "30 6 * * 1-5" --timezone Europe/London \ --attempts 3 --backoff 60 -- bash -lc 'date -u +%FT%TZ && echo report sent' runtime jobs trigger # one run now; the schedule does not move runtime jobs logs -f # the latest run's output runtime jobs cancel # end it for good ``` `trigger` is how to test a job before its first time comes. Delete the `cancel` line, and the job keeps its schedule. ## Settings that decide whether it works | Setting | What it does on Runtime | | ------------- | ------------------------------------------------------------------------------------------------------------- | | `timezone` | An IANA name such as `Europe/London`. The job stays at 06:30 local time through daylight saving. UTC when omitted | | `retry` | `maxAttempts` from 1 to 5, the first attempt included, with `backoffSeconds` between attempts | | `timeoutSeconds` | How long one run may take before it is stopped. 30 minutes when omitted | | Overlap | Runs of one job never overlap. A time that comes while a run is still going is not started beside it | | Output | Each run keeps its exit code and its last 256 KiB of output | ## Mistakes to avoid - **Writing the schedule in UTC for a local-time task.** A job meant for 07:00 in New York runs at 06:00 or 07:00 local time depending on the season. Name the time zone instead of converting by hand. - **A job that is not safe to run twice.** A retry runs the command again from the start. Make the command check whether its work is already done, such as a file or row for today, before it writes anything. - **Trusting `*/7` to mean every seven minutes.** A step counts from the start of its field, so `*/7` in the minute field runs at :00, :07 and so on to :56, then at :00 again, four minutes later. See [cron expressions](/glossary/cron-expression). - **No owner for failures.** Read the run's state each morning, or have the command post to a chat or a webhook when it fails. ## Related - [What is a cron expression?](/glossary/cron-expression) for the five fields and common schedules. - [Scheduled AI agents](/blog/scheduled-ai-agents), a job that runs an agent each night and opens a pull request. - [How to speed up GitHub Actions](/how-to/speed-up-github-actions), for checks that run on each push rather than on a clock. - [Jobs](/docs/jobs) in the guides. ## Sources - [crontab(5), Linux manual page](https://man7.org/linux/man-pages/man5/crontab.5.html), checked 17 November 2026 - [GitHub Docs, Events that trigger workflows: schedule](https://docs.github.com/en/actions/reference/workflows-and-actions/events-that-trigger-workflows#schedule), checked 17 November 2026 Facts on this page were checked on 17 November 2026.