Il mar, 2003-12-02 alle 12:36, Fabio Massimo Di Nitto ha scritto: > Ciao Roberto, > > On Tue, 2 Dec 2003, Roberto Sebastiano wrote: > > > > No. otherwise we would have told you before. I cannot reproduce it on any > > > of my servers.. some of them with almost 2 years of uptime... > > > > > Maybe only on server with an high number of apache childs > > I have 500 childs on the machines that exposed the problem > > > > I think that the problem is clearly that /etc/init.d/apache reload fails > > sometimes with an high numbers of childs > > Can you try using these init script?: > > http://cvs.raw.no/cgi-bin/viewcvs.cgi/debian-apache/debian/apache.init?view=markup > > you might have to remove the should_start check since it was introduced > recently. > I don't think is a init script problem I think there's a bug when apache receives SIGUSR1 with an high number of childs. Sometimes the master process dies leaving the child up > If you have a high traffic box in loadbalacing you might perhaps try a > reload on one them when it has a lot (>100) childs running.. of course i > can't promise it will fix the issue since as you saw yourself it doesn't > happen on a regular base. > These box are so high loaded that I don't use dynamic child startup. I have instead always a fixed numebr of child (the maximum the servers can handle). Dynamic childs loading make the server not respond well, as request may arrive all at once (and the server should have to fork() too fast) Those are task specific machines, there's no need to saving resources having less childs around Cheers, -- Roberto Sebastiano <robs@multiplayer.it>
Attachment:
signature.asc
Description: Questa parte del messaggio =?ISO-8859-1?Q?=E8?= firmata