httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Arkadi Shishlov" <>
Subject piped log bug?
Date Sun, 27 Feb 2005 10:52:53 GMT
I believe there is a bug in Apache which makes some child processes
stuck at graceful restart.

On my medium busy FreeBSD and Linux servers I noticed that sometimes
apache leaves previous instance children in non-working state after
restart. Examination with ps/strace/truss shows that all of them
are writing to piped (cronolog backed) access log, but by that time
all cronolog processes are already gone.

I quickly examined the source, both 1.3 and 2.0, and seems that the
reading side of logging pipe is not closed in Apache process. I think
the server must close reading fd in child process, or else all children
are stuck in pipewr when there are no one to read the log.

Or I'm missing something?


View raw message