launchpad-reviewers team mailing list archive

[Martin Pool <mbp@xxxxxxxxxxxxxx>]

Martin Pool has proposed merging lp:~mbp/launchpad/690021-rlimit into lp:launchpad.

Requested reviews:
  Launchpad code reviewers (launchpad-reviewers)
Related bugs:
  #690021 scan_branches terminated for excessive memory abuse
  https://bugs.launchpad.net/bugs/690021


This change limits the address space of scan_branches.py to 2GB.  If it tries to allocate more, it will crash.  See <https://bugs.launchpad.net/launchpad-code/+bug/690021>

After some discussion with spm and mwh on irc this seemed like a good idea.  It is consistent with what we do in other scripts.

2Gb is a compromise between I hope being small enough to stop the machine bogging down, but not so small that scans that would otherwise succeed will now fail.

It would be nice if the limit was not hard coded, but this seems like the most pragmatic thing for now.  I added a note on https://dev.launchpad.net/LEP/FeatureFlags that we could potentially do it there.

Obviously this will not actually fix whatever the bad scaling behaviour is in bug 690021; we should probably split that to a separate bug.

I would welcome opinions on how to test this.  Do these jobs run on staging or qastaging?  Perhaps we should just let it get to production and then watch for it to fail cleanly.
-- 
https://code.launchpad.net/~mbp/launchpad/690021-rlimit/+merge/43733
Your team Launchpad code reviewers is requested to review the proposed merge of lp:~mbp/launchpad/690021-rlimit into lp:launchpad.

=== modified file 'cronscripts/scan_branches.py'
--- cronscripts/scan_branches.py	2010-04-27 19:48:39 +0000
+++ cronscripts/scan_branches.py	2010-12-15 05:38:55 +0000
@@ -9,6 +9,8 @@
 
 import _pythonpath
 
+import resource
+
 from lp.services.job.runner import JobCronScript
 from lp.code.interfaces.branchjob import IBranchScanJobSource
 
@@ -21,5 +23,8 @@
 
 
 if __name__ == '__main__':
+
+    resource.setrlimit(resource.RLIMIT_AS, (2L << 30, 2L << 30))
+
     script = RunScanBranches()
     script.lock_and_run()