💾 Archived View for bbs.geminispace.org › u › byte › 11575 captured on 2024-08-18 at 20:32:09. Gemini links have been rewritten to link to archived content
⬅️ Previous capture (2024-07-09)
-=-=-=-=-=-=-
Re: "help debugging likely disk write issue?"
@tjp did it work out?
2023-11-08 · 9 months ago
☕️ Morgan · 2023-11-09 at 14:16:
Usually with this kind of issue I would try to add any kind of alternative log output possible ... aka "printf debugging" :) ... to learn more next time it happens. In cases where printf is not visible I would append to a file somewhere.
🌊 tjp [OP] · 2023-11-09 at 15:59:
Thanks for the suggestions @byte. I'd be really surprised if there was a deadlock issue in svlogd as that's a heavily tested and used program as it's the logger in runit. I also don't think it should run out of file descriptors (it'd have to leak them), but is there a way I can see file descriptor counts for a process in FreeBSD? That's the kind of debugging I'm looking for.
🌊 tjp [OP] · 2023-11-09 at 16:02:
Nothing new to report as it hasn't happened again since I first posted here (it generally runs for a few weeks before this lock-up occurs). I'm hoping to gain a little knowledge for when it does happen again: how can I inspect a running process, for example to see how many file descriptors it has open? Or whether it's currently blocked on a file write call, or has some other syscall open but not completed?
just strip everything from the server part, and write some logging calls in a loop there, this way you can reproduce it without waiting for a week and try to debug it.
🌊 tjp [OP] · 2023-11-13 at 18:21:
I finally figured this out.
Based on the frequency of the problem arising I thought it might be connected with svlogd's log rotation, and the svlogd manpage refers to a possibility of it locking up when an external log file processor fails. So to induce it more quickly I reduced the max log file size and generated a bunch of traffic to my server with a shell script.
I was using the !processor functionality of svlogd to run rotated log files through a script (`!gzip`), but also starting svlogd with `chpst -p 1`, which sets a soft limit of 1 process. I removed that limit and it works fine now.
help debugging likely disk write issue? — I have a server service running in a freebsd server I manage, which is connected via a pipe to a logging service which writes to disk. Periodically (generally after a few weeks), the server entirely stops doing anything. It doesn't respond to any requests, and even after a restart it doesn't bind to any ports and none of the usual startup log messages come through. The only way I've found to get it un-stuck is to restart the logging service in the...