Forum: Cfengine Help Subject: Re: db errors Author: sauer Link to topic: https://cfengine.com/forum/read.php?3,22382,22472#msg-22472
Mikhail Gusarov Wrote: ------------------------------------------------------- > On 06/07/2011 07:20 PM, Michael Stevens wrote: > > We're only running cf3 on a subset of our > machines for now, so I'm inclined to wait for > 3.1.6, as functionality doesn't seem to be > impaired. Any adverse impact from this beyond > warning messages? > > Well if you don't see any problems, then you're > lucky -- given it's only > locks database which is seriously affected, the > only side effect I can > think of is verifying some promises more often > than expected. > > The problem manifests itself badly on high-loaded > cf-serverd instances > where different threads stomp on each other toes > trying to verify and > write to same database often. With several thousand machines running, I'm pretty frequently seeing cf-agent processes which will just hang up, leaving 2-10 cf-agent processes taking 100% CPU and just running "forever" since moving to from 3.0.5 to 3.1.5. This is with a promise set which takes about a minute to run; it's not complicated, but it still explodes seemingly randomly. Sometimes it just dies, other times it hangs up and blocks. On a couple of machines where I have really long-running promises (~20 minutes) the cf-agent processes build up until the system is completely brought down - usually within 24-36 hours. In both cases, I'm actually only seeing problems which appear related to db manifesting in cf-agent; I haven't upgraded any of the policy servers to 3.1.5 yet, and probably won't. But since the question was asked "how does this show up"... :) I'm pretty sure we'll have to stop using the community RPM and start building our own, since the fixes are always in SVN. This issue really seems like it justifies a patch-level package be built, but I guess time will tell what happens with that. Unfortunately, it's not a trivial task (mostly politically) to deploy a new binary across a very large network, so I'm screwed either way. :) _______________________________________________ Help-cfengine mailing list Help-cfengine@cfengine.org https://cfengine.org/mailman/listinfo/help-cfengine
