Dang. That does look an awful lot like my issue. I am in fact using directory environments (mostly because of the screaming deprecation warnings telling me I was a bad man if I didn't).
:/ Shame on me for using a .0 release. On Thursday, May 22, 2014 10:17:24 AM UTC-7, Ellison Marks wrote: > > Hey, I think I remember another thread that mentioned that there were some > performance issues with directory environments. Basically, the next 3.6 > release will add a caching option that mostly alleviated the problem for > the OP from that thread. > > https://groups.google.com/forum/#!topic/puppet-users/wzy8NPWauu4 > > On Thursday, May 22, 2014 9:59:25 AM UTC-7, Tristan Smith wrote: >> >> After much hacking to get directory environments settled and the manifest >> directory in place, I rolled Puppet 3.6 to our puppetmasters last night. >> >> One of our puppetmasters has nearabouts 1000 clients, runs passenger >> under apache 2.2 (ruby 1.8.7, sadly, thanks CentOS), and normally doesn't >> really notice puppet running - basically peaks out at 20% CPU usage. >> >> Under 3.6, even doubling the passenger worker count, it couldn't keep up >> with the load - I started running out of apache procs due to workers stuck >> in waiting mode and they were all just hanging waiting for a passenger >> worker to free up. CPU usage on the system capped out. >> >> Strace -c on the passenger workers had them spending 30% of their time in >> clone() and 60% in wait4(), fwiw. >> >> I'm going to be digging to figure out what in hell changed to cause this, >> but has anyone else experienced a significant change in performance under >> 3.6? >> >> --Triss >> > -- You received this message because you are subscribed to the Google Groups "Puppet Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/puppet-users/38226246-7171-49dd-8d30-1782c061156b%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
