Bug#2552: syslogd and race conditions
Martin Schulze writes ("Bug#2552: syslogd and race conditions"):
> Raul or Ian wrote that sometimes syslogd doesn't start correct. I was
> able to reproduce this for some time, but was unable to debug it. The
> delay came from the part where all file descriptors were closed. This
> routine (the library- or kernel-routine) hang.
>
> For a long time I don't have this problem anymore. I think this is a
> race condition where we can't do anything.
Race conditions should be fixed.
> I will put a kill -HUP syslogd in the startup script.
Surely there is some better solution, like actually fixing the real
problem ? I suppose this might do for a workaround.
> I, too, have the problem that I can't login into my server for about 30secs
> after booting. I, too, think that this is a syslog-problem, but can't find
> out why. This problem doesn't occur on my desktop machine which is
> Debian 1.1 (1.3.95) and sysklogd 1.3-2 ELF. It only happenes on the server
> which is still a.out.
I get this every time my machine boots, and it takes about a minute
and a half.
It seems to get stuck logging the end of the kernel bootup messages -
it gets most of them and then goes very very slowly (one line every
10-30 seconds) for the last few lines.
I suspect a bug in klogd or the kernel, or something where syslogd is
spinning in a system call loop and prevenging klogd from getting
enough CPU.
Ian.
Reply to: