Closed pmclanahan closed 9 years ago
@peterbe, can you give this one a shot and see if it fixes your issues?
@pmclanahan I like it. Let me have a play.
@jsocol the sad thing is that I'm not able to reproduce the threading problem. Most likely because I'm not setting up my apache on EC2 the same way as Mozilla IT does it.
All I know is Apache 2.2, MPM-Worker, ModWsgi in daemon mode is what bedrock used when we saw issues. I think the wsgi daemon process was set to 14 threads and 2 procs, though the number of procs shouldn't matter. Set it to a bunch of threads and 1 proc and hit it with a bunch of simultaneous requests using something like bees with machine guns. Might work.
I think the reason why I can't reproduce this on EC2 is because it's hard to install mod_wsgi for python 2.6 and I have too little patience (considering the ROI) to bother.
If I continue to get lots of problems on airmozilla I'm going to ask IT to switch to multiple processes instead of multiple threads.
But any reduction of global
is a friend of mine.
I'm not merging this now because we have no reliable info that it actually fixes anything. I realize we might not be able to test it in the test suite (which... ugh, but OK) but we need to have some evidence that it's a helpful switch.
Agreed. Peter and I were confident we'd get some info from our load testing, but we really didn't. If I find an opportunity to reproduce this and test I'll come back to it. Thanks James.
I'm going to close this out. Anyone runs into it again, please re-open.
This is my proposal for fixing the threading issues noted by @peterbe in #55. Tox is happy, but I'm not sure how well covered an issue with this is. I have also broken backward compat if
jingo.load_helpers
is considered public API.cc @jsocol @jezdez @jbalogh