On Mon, 2024-02-05 at 16:20 +0200, Yanko Kaneti wrote:
On Mon, 2024-02-05 at 14:31 +0100, Florian Weimer wrote:
> >
> > Running with rtkit-daemon stoped and masked prevents the main evo
> > process being prioritized and consequently killed with higher load.
> >
> > The fact that the kernel kills the (main)thread without a beep anywhere
> > in the logs is strange to me.
>
> I think it's not so unusual for the kernel not to log the SIGKILL. This
> does not seem to be purely an emergency measure like the OOM killer—it's
> more like something that an application (set of processes) told the
> kernel to do to itself. As such, it's not unreasonable for the
> application to do the logging. On the other hand, I think it's
> difficult for rtkit-daemon to arrange for that logging because it can't
> monitor the target process directly.
Logical but leaves users like me scratching there head with mishaps like
these. Given that print-fatal-signals kernel param (which I just learned
about) is off by default.
FWIW, I've been running for the last couple days with print-fatal-
signals=1 and the only additional logs I get are exactly the ones that
would've helped in this case.
kernel: RT Watchdog Timeout (hard): evolution[4604]
kernel: RT Watchdog Timeout (hard): liferea[5045]
i.e. RT watchdog kills do seem like an out-of-the-ordinary events and
there is case to be made that logging in should be on by default.
-Yanko
>