I now mannequin every scheduled run with greater than success and failure. A run can have succeeded, failed, arrived late, been skipped by coverage or be unknown as a result of an enter couldn’t be noticed. That final state is the awkward one, and it’s the one which earns its maintain. Every anticipated run has a freshness deadline. Every necessary knowledge supply has an availability end result separate from its merchandise rely. A heartbeat is accepted solely after the job produces the artifact the schedule exists to provide.
This creates extra yellow on the dashboard. Good. I’d relatively examine an sincere yellow than belief a fabricated inexperienced. The fleet just isn’t much less dependable than it was earlier than I began monitoring this. It is strictly as dependable because it all the time was, and I can lastly see it.
Read the operating state
None of that is new pc science, which is strictly why it’s straightforward to overlook. Agent methods arrive wrapped in new language — planners, device use, reminiscence, reflection — however they nonetheless depend upon schedulers, shells, paths, credentials, clocks and exit codes. The new layer doesn’t repeal the outdated ones. It offers the outdated ones extra methods to fail quietly.
