A scheduler that only mails on failure answers one question. This release separates a task that failed from a check that found something, sends both where operators already look, and lets a monitor ask whether the scheduler itself is alive.
Alerts apart from failures
- Check tasks A task whose result carries the alert columns is a check: its run ends well and records each alert, with a level, a key, a text and a link, in one transaction.
- A fluent API A script returns
Ax.ext.core.cron.alerts().critical(key, text)rather than building a result set; an empty return is a check that found nothing. - Who is told One SQL statement resolves which mail and channel rows receive a run's alerts.
Channels other than mail
- Seven targets Google Chat, Slack, Mattermost, Microsoft Teams, Discord, Telegram, or any system that accepts a JSON webhook, each with a guided setup and a Send a test button.
- An outbox Notices are queued and sent with fixed timeouts and no redirects; a failure is retried only when it is temporary, and the channel's secret is decrypted in memory and kept out of every error.
Liveness and control
- JMX Every node holding the scheduler publishes its status, heartbeat age, tasks running and failing, overdue tasks and longest run, for an external monitor to read.
- Run now Starts a task once on the node where it is pressed, queued behind a scheduled run of the same task rather than beside it.
The backup health check is the first task built on it: a failed run is the alert, and an OK run is the ping.