On Sun, 22 Nov 2009, Ask Bj?rn Hansen wrote: > Hi everyone, > > Today our monitoring system started sending messages that our primary > beanstalk box was upset. beanstalkd was using all available memory > (~12GB). > > We have the binlog enabled. After restarting the box (and thus > beanstalkd) it's using a few hundred KB for 2400 jobs. > > The memory usage had increased steadily over the last ~5-6 days since > we updated to 1.4.2 (plus patches: git = ee671f).
We've seen this behavior as well with 1.4.2 on our production beanstalkd instance, unfortunately it's been practically unreproducible in a development environment. Running beanstalkd under valgrind doesn't give me the impression that beanstalkd is leaking memory so much as it's being over-ambitious in its allocation of memory for its internal job management. Haven't gotten any good idea of what it might be, our load is on average quite low but has periodic spikes, so I've not tracked down anything entirely useful :( Cheers, -R. Tyler Ballance -------------------------------------- GitHub: http://github.com/rtyler Twitter: http://twitter.com/agentdero Blog: http://unethicalblogger.com
pgpgd6QPvO4O9.pgp
Description: PGP signature
