I like thie idea of the high priority getty. The problem is that Zimbra is very memory intensive, and spawns a lot of Java processes when it receives a lot of messages at one (ie 12 nagios alerts at once sent to 4 different people). However, that's a symptom not the problem. Those 48 nagios alerts could just as easy be 48 spam messages or a DOA attack.
I'll do some more research, but at least I'm on the right track now. Thanks again! Doug On Tue, Aug 26, 2008 at 12:57 AM, Lars Ellenberg <[EMAIL PROTECTED]>wrote: > On Mon, Aug 25, 2008 at 11:50:28PM -0400, Doug Eubanks wrote: > > I've reconfigured watchdog, and set it to 100, 90, 80. > > > > I tried ridiculously large values (500), but one of the systems went > > completely unresponsive. > > to keep your servers responsive > > * reconfigure your application > to not behave like an io intensive fork bomb. > * use deadline io-scheduler > * mount noatime > * put a (ulimit-ted) memlocked high priority > (busybox) getty on the console > > it is my strong believe that triggering failover by rebooting > due to too much application load > is the wrong approach. > > something is causing the load, > typically that something is client requests. > > during failover (and coming up cache cold on the other node), > more client requests pile up. > > the strategy "reboot and failover" is > likely to worsen any load problem. > > -- > : Lars Ellenberg > : LINBIT HA-Solutions GmbH > : DRBD(R)/HA support and consulting http://www.linbit.com > > DRBD(R) and LINBIT(R) are registered trademarks > of LINBIT Information Technologies GmbH > _______________________________________________ > Linux-HA mailing list > [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems > _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
