← all releases

v1.9.2

on GitHub

Not the newest. v1.10.0 is out →

Release notes

Changed

  • A negative warn_after, restart_delay or graceful_stop in [defaults] is rejected with an error naming [defaults] instead of loading and being ignored.
  • A reload restarts a running service only when a change reaches its process (command, shell, working directory, environment, user, limits, instance count). Edits such as its description, retention or restart policy apply without a restart.
  • failures output patterns match the text without colour codes, so ^ERROR also matches a coloured line.

Fixed

  • A failures output pattern also matches lines printed after the log reached log_max_size, so a run that printed the error past the cap is marked failed.
  • Retention and the trash purge delete more than 32,766 runs at once instead of failing on every pass.
  • A notifier URL that redirects with 301, 302 or 303 is reported as a failed delivery naming the new address, instead of counting as delivered while the message was lost. 307 and 308 redirects are still followed.
  • In-app notifications on a host outside UTC are kept for keep_for and grouped over the right window, instead of being shifted by the time zone offset.
  • The log viewer numbers lines correctly when the rotated .prev log file is empty.
  • Stopping a task, or removing it in a reload, cancels a retry still waiting out its delay.
  • [storage] max_size no longer deletes all finished runs when the logs of active runs and stray files alone exceed it. It keeps the history and logs a warning instead.
  • Deleting more than 32,766 selected runs at once works instead of failing.
  • The web UI catches up after the daemon restarts, after a long disconnect, and when another tab takes over the live connection, instead of missing runs and notifications from the gap. The live event stream sends a new resync event when a client must refetch its state.
  • After the daemon is killed outside a service manager (kill -9 from a shell), the next start stops the runs it left running, using each task’s stop_signal and graceful_stop, and notes this in the run’s log. The next firing no longer starts a second copy next to them.
All releases