Date: Sun, 04 Apr 2010 20:56:28 -0700 From: Doug Barton <dougb@FreeBSD.org> To: Jilles Tjoelker <jilles@stack.nl> Cc: svn-src-stable@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org, svn-src-stable-8@freebsd.org Subject: Re: svn commit: r205806 - stable/8/etc Message-ID: <4BB95F6C.7070202@FreeBSD.org> In-Reply-To: <20100329173214.GA17249@stack.nl> References: <201003282019.o2SKJfPg033857@svn.freebsd.org> <4BAFBBFA.7020701@FreeBSD.org> <20100328210630.GA2086@stack.nl> <4BAFE1EE.9040908@FreeBSD.org> <20100329173214.GA17249@stack.nl>
next in thread | previous in thread | raw e-mail | index | archive | help
This is a multi-part message in MIME format. --------------070707050009090200000503 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit On 03/29/10 10:32, Jilles Tjoelker wrote: > On Sun, Mar 28, 2010 at 04:10:38PM -0700, Doug Barton wrote: >> On 03/28/10 14:06, Jilles Tjoelker wrote: >>> On Sun, Mar 28, 2010 at 01:28:42PM -0700, Doug Barton wrote: >>>> Probably my fault for not saying something sooner, but there is a >>>> problem with the code in head that sometimes causes it to loop >>>> repeatedly even though pwait exits successfully. I am trying to track it >>>> down, but since it only happens about once every 10 shutdowns it's been >>>> difficult. > >>> There is a difference between the two methods in what is waited for >>> exactly. pwait(1) will wait for the process to terminate; if it is >>> applied to a zombie it will return immediately (printing the exit status >>> if -v was given). On the other hand, kill(1) will continue to return >>> success until the process has been waited for by its parent. > >> The process that I see this with most often is devd, does that fit the >> model you're describing? > > Possibly. This would mainly happen because init has been busy, I think > (or if the parent isn't init). > >> What are the implications of moving on after a >> successful pwait even though there is still a zombie process? > > For shutdown/stop, nothing. > > For restart, there may be problems if a restarted daemon checks the > validity of the pid in the pidfile using kill(). Ok, in that case I'm not really comfortable with the idea of ignoring the results of kill -0, however I've come up with what I think is a good solution in the attached patch. Based on your description and my ongoing analysis the problem I was seeing with the long string of the same pid repeated over and over seems to be a side effect of pwait returning successfully (thus the || sleep 2 never kicks in) but 'kill -0' still being able to see the pid. So, I've moved the sleep up so that if we're not in the first pass but kill -0 is still seeing the pid that we sleep for 1 second, then proceed. I think that'll handle both the problem I saw, and the odd case where pwait doesn't return successfully. Sound good? Doug -- ... and that's just a little bit of history repeating. -- Propellerheads Improve the effectiveness of your Internet presence with a domain name makeover! http://SupersetSolutions.com/ --------------070707050009090200000503 Content-Type: text/plain; name="wait_for_pids.diff" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="wait_for_pids.diff" Index: rc.subr =================================================================== --- rc.subr (revision 206117) +++ rc.subr (working copy) @@ -359,12 +359,14 @@ if [ -z "$_list" ]; then return fi - _prefix= + + local _prefix= while true; do _nlist=""; for _j in $_list; do if kill -0 $_j 2>/dev/null; then _nlist="${_nlist}${_nlist:+ }$_j" + [ -n "$_prefix" ] && sleep 1 fi done if [ -z "$_nlist" ]; then @@ -373,7 +375,7 @@ _list=$_nlist echo -n ${_prefix:-"Waiting for PIDS: "}$_list _prefix=", " - pwait $_list 2>/dev/null || sleep 2 + pwait $_list 2>/dev/null done if [ -n "$_prefix" ]; then echo "." --------------070707050009090200000503--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4BB95F6C.7070202>