Hi, On 2023-02-15 09:57:41 -0800, Nathan Bossart wrote: > On Tue, Feb 14, 2023 at 04:20:59PM -0800, Andres Freund wrote: > > On 2023-02-14 14:23:32 -0800, Nathan Bossart wrote: > >> On Tue, Feb 14, 2023 at 12:47:12PM -0800, Andres Freund wrote: > >> > Not really related: I do wonder how often we end up self deadlocking in > >> > quickdie(), due to the ereport() not beeing reentrant. We'll "fix" it > >> > soon > >> > after, due to postmasters SIGKILL. Perhaps we should turn on > >> > send_abort_for_kill on CI? > >> > >> +1, this seems like it'd be useful in general. I'm guessing this will > >> require a bit of work on the CI side to generate the backtrace. > > > > They're already generated for all current platforms. It's possible that > > debug > > info for some system libraries is missing, but the most important one (like > > libc) should be available. > > > > Since yesterday the cfbot website allows to easily find the coredumps, too: > > http://cfbot.cputube.org/highlights/core-7.html > > Oh, that's nifty. Any reason not to enable send_abort_for_crash, too?
I think it'd be too noisy. Right now you get just a core dump of the crashed process, but if we set send_abort_for_crash we'd end up with a lot of core dumps, making it harder to know what to look at. We should never need the send_abort_for_kill path, so I don't think the noise issue applies to the same degree. Greetings, Andres
