Blog
Why scheduled jobs miss runs or fail silently, and how to catch it in the schedulers you already use.
Every process that imports your app and calls start() runs its own scheduler. Count them, see what --preload and the dev reloaders really do, then pick one scheduler process and accept its trade-off.
A Better Stack heartbeat expects a request to a secret URL every period plus grace. What it can and cannot express, what heartbeat blocks and responder licenses cost, and where CronRadar differs.
What self.retry() and autoretry_for actually compute, the default delay the docs misstate, why a killed worker is a separate path, how retries collide with beat, and which signal fires only when Celery gives up.
The common crontab error messages, each traced to the source line that prints it in cronie, Debian cron, and BusyBox: what triggers each one, the retry prompt, the missing-newline errors, and what the daemon logs when nothing checked the file.
GitLab queues scheduled pipelines only when its schedule worker runs, and a schedule goes inactive when its owner is blocked or removed. Start delays, rules, invalid configs, notifications, and a pipeline that checks in.
The LocalRequestsOnlyAuthorizationFilter default, sync and async dashboard filters, MapHangfireDashboardWithAuthorizationPolicy, IsReadOnlyFunc, reverse proxies and path bases, and what the dashboard cannot tell you on its own.
Hangfire resolves every job from a fresh DI scope and disposes it when the job returns. Here is what that means for DbContext, why ObjectDisposedException repeats on every retry, and the patterns that avoid it.
The @nestjs/schedule 12.0.2 decorators and options checked against source, the try/catch that swallows handler errors, the waitForCompletion and threshold skips, and the official @nestjs/locks way to run a job on one instance.
Six Node.js schedulers side by side, every cell sourced: seconds fields and cron extensions, time zones, what survives a restart, overlap options, multi-instance behavior, latest releases, and weekly downloads.
Opsgenie access ends April 5, 2027. What a heartbeat interval actually checks, where heartbeats live in Jira Service Management, how to convert one to a cron schedule plus grace, and a parallel-run checklist.
What run_pending does, why a job exception ends the loop and where to catch it, how a long job delays the rest, what a restart forgets, how .at() handles time zones and DST, and how to report every run.
The exact Quartz.NET 3.22.9 cron rules from CronExpression.cs: why one day field must be ?, what L, W, LW, L-n, and # mean, what the builder helpers return, how daylight saving moves a run, and how to rewrite a Quartz string as a five-field UTC schedule.
Both coding agents run unattended, and both can exit 0 without doing the work: Claude Code after its tool calls were denied, either one after changing nothing. The flags that bound a run, the output that says what happened, and a wrapper that reports it.
Point a cron job at an UptimeRobot heartbeat URL: how the interval and grace period decide down, why a failed run looks like a missing one, which plans and prices include heartbeats, and how it compares to CronRadar.
Claude Code, Cursor, or any coding agent can wire monitoring into your scheduled jobs. Scope its access, keep the deployment key out of the chat, review the diff, and confirm a real run yourself.
Paused DAGs, import errors, full pools, and a stopped scheduler all look the same in Airflow: nothing happens. How Airflow 3 schedules runs, what Deadline Alerts see, and an outside check per DAG.
Anacron runs daily, weekly, and monthly jobs once per period, even on machines that were switched off. When each distribution starts it, the anacrontab format, and why cron.daily skipped today.
The six-field cron() format, rate() and at(), time zones, flexible windows, retry policies, the execution role, and alerts for scheduled Lambda jobs that fail or never run.
NCRONTAB syntax, WEBSITE_TIME_ZONE limits by plan, the blob lease that keeps one run at a time, IsPastDue catch-up after downtime, and alerts for timer runs that fail or never start.
upsertJobScheduler and the end of repeatable jobs in BullMQ 6, missed runs that collapse into one, stalled and hung processors, Redis eviction, and check-ins from the worker.
A renewal job can exit 0 having charged nobody. The outcomes, expectations, and deadlines to monitor for billing, email, report, export, and deletion jobs, plus reconciliation and routing.