Troubleshooting • Re: watchdog process restart

A systemd-aware service daemon sends a periodic sd_notify(3) "WATCHDOG=1" message to systemd, allowing systemd to detect a service which has hung but not crashed (theoretically, at least, it depends on a good implementation in the service daemon). See WatchdogSec= in systemd.service(5).

I’ve never seen a “systemd-aware service daemon” in the wild. Are there any?

Besides, watchdog has test/repair facilities that can run an arbitrary executable such as a client to a daemon in order to perform custom tests. The only requirement for the client is to exit with a non-zero signal in case the test failed.

TBH I’ve found systemd-watchdogd to be a bit of a head scratcher.

• It boasts advantages that only exist if developers implement its API, and 10 years on, about nobody did.
  The traditional linux watchdog can also be effectively extended thanks to Unix signals, a free API, so to speak.
• Out of the box without integration, systemd-watchdogd has lesser capability than the Linux watchdog which includes multiple standard tests.

I’m not particularly looking for an argument but in general I think watchdog is a fairly misunderstood concept, and if systemd-watchdogd comes to the fore its advantages should be made clear.
Right now, I don’t see any.

Statistics: Posted by epoch1970 — Wed Sep 04, 2024 6:44 am

---