Bump the open file ulimit on OSX.

Review Request #3733 - Created April 20, 2016 and submitted

Information
John Sirois
pants
jsirois/osx-ci/bump-open-fh-ulimit
3239
Reviewers
pants-reviews
patricklaw, stuhood
Tests showed the default was 256 previously which appears to be too low
for modern pants as evidenced by CI failures caused by too many open
file handles.

 .travis.osx.yml | 3 +++
 1 file changed, 3 insertions(+)
OSX CI ran here:
  https://travis-ci.org/pantsbuild-osx/pants/builds/124555939

That had 1 shard failure in a test, but apparently no open-fh issues
any longer.

That confirms the 256 initial setting and successful bump to 8192.

Still not sure if this will do the trick when it hits master, but
256 is a bit low.

Normal CI cancelled since it does not test the file edited here.
John Sirois
John Sirois
Stu Hood
John Sirois
Review request changed

Status: Closed (submitted)

Change Summary:

Now on master:

git log -1 origin/master
commit 14b54722bd1f32c5ffb9575f5870203e4ac9f6d7
Author: John Sirois <john.sirois@gmail.com>
Date:   Wed Apr 20 13:59:36 2016 -0600

    Bump the open file ulimit on OSX.
    
    Tests showed the default was 256 previously which appears to be too low
    for modern pants as evidenced by CI failures caused by too many open
    file handles.
    
    Testing Done:
    OSX CI ran here:
      https://travis-ci.org/pantsbuild-osx/pants/builds/124555939
    
    That had 1 shard failure in a test, but apparently no open-fh issues
    any longer.
    
    That confirms the 256 initial setting and successful bump to 8192.
    
    Still not sure if this will do the trick when it hits master, but
    256 is a bit low.
    
    Normal CI cancelled since it does not test the file edited here.
    
    Bugs closed: 3239
    
    Reviewed at https://rbcommons.com/s/twitter/r/3733/
Eric Ayers

I think there is something going wrong in Pants unit testing that is keeping too many files open - leaking file descriptors. when I run build-support/bin/unit-test.sh it can't make it all the way through. Yet, if I run those tests that fail in isolation, tbe tests work fine.

  1. I went too high in this change by using 8192 since ci.sh works for me on linux with 1024 for `ulimit -n`.  The OSX `256` was clearly too low, but that does seem high and I wouldn't doubt fd leaks.  Since you can repro a failure Eric, it would be bonus if you find time to hunt this down.
Loading...