http://issues.apache.org/SpamAssassin/show_bug.cgi?id=4703
------- Additional Comments From [EMAIL PROTECTED] 2005-11-23 21:34 -------
'I really don't know if there is something around the spamassassin code that
once in a while loops over all dead children and waits for them'
yep, there is, in spamd.
I think the fix from bug 4518 is better than patch 3266; with that, the
processes are killed, which is the most important thing, the FDs leak is
plugged, and while the leftover zombie processes are not cleaned up
immediately, they do get cleaned up eventually by spamd.
Patch 3271 may be better, but unfortunately *includes* the bug 4518 fix instead
of building from that, so it's quite hard to read, and will not apply to 3.1.x
or trunk. This is bad. bear in mind 3.0.x is not a supported distro, so
going to that much effort before fixing the 2 supported ones is going
overboard!
I'm -1 on applying these patches as a result.
I would prefer to see a new patch, building on top of the patch from bug 4518,
which adds:
- the same code, or calls to a shared Util subroutine, in the other modules
that use helper processes, of course
- a waitpid() call after SIGTERM
- possibly, a SIGKILL, but I'm really not sure if this is really necessary
(ps: response to a question -- we cannot use CHLD=IGNORE, due to other bugs on
other platforms, iirc; so we always need to wait for children somewhere.)
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
