https://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=20582
--- Comment #30 from David Cook <[email protected]> --- (In reply to Jerome Charaoui from comment #29) > In order to deal with this I modified the config by adding "accepts => 200" > so that workers get recycled more quickly. If you have any other ideas as to > how we could deal with memory usage, please let us know! I think we'd have to do more profiling, but I have a couple of thoughts. 1) I think memory usage will always go up, because when the worker first starts, it's only pre-loading a small selection of modules. As more work gets done, I think more modules get loaded into memory, and they stay there for the life of the worker. But how much memory is used by loading modules... that's something I don't know. I have looked a bit at memory use based on modules being loaded and noticed big differences. 2) I'm guessing maybe we have memory leaks. My first guess would DBIx::Class objects hanging around longer than they should. Some of that might be scoping and some of that might be caching in DBIx::Class. Mind you that's just a guess. This is something that could maybe be investigated and changed. But yeah... both just guesses. I'd say it would need application level profiling to really know. But considering how a multi-threaded Tomcat process can eat up GBs of memory, a multi-process hyponotoad eating up GBs of memory collectively doesn't really shock me either. -- Oh one thought... if we pre-loaded more modules into the master hypnotoad process, in theory there would be less in the unshared memory, since they'd be using the shared memory for the modules. I once encountered a legacy mod_perl app which wasn't preloading enough modules, and it caused performance problems when new workers were forked because they had to reload lots of modules. I preloaded everything important into the master, and performance went way up. Memory wasn't really a big deal for the app as we had lots to spare, but I assume copy-on-write was used so that the preloaded modules were in shared memory. Just my 2 cents. -- You are receiving this mail because: You are watching all bug changes. _______________________________________________ Koha-bugs mailing list [email protected] http://lists.koha-community.org/cgi-bin/mailman/listinfo/koha-bugs website : http://www.koha-community.org/ git : http://git.koha-community.org/ bugs : http://bugs.koha-community.org/
