#general
- r
rough-minister-58256
04/06/2016, 1:12 AMhaha - b
bored-art-40741
04/06/2016, 1:12 AMYeah that's what I'd expect - b
bored-art-40741
04/06/2016, 1:12 AMLike you know cache hits in pants? - r
rough-minister-58256
04/06/2016, 1:13 AMcache hits? - b
bored-art-40741
04/06/2016, 1:13 AMIt took David Taylor and I running down a bug in the CPython 2.7.x zlib module to discover why they performed so badly ~2 years ago - r
rough-minister-58256
04/06/2016, 1:13 AMsounds about right heh - b
bored-art-40741
04/06/2016, 1:13 AMIf you fork a worker process pool while the zlib (global static!) lock is taken, the workers inherit the lock in the taken state - b
bored-art-40741
04/06/2016, 1:13 AMBut naturally it never gets released in their memory space - b
bored-art-40741
04/06/2016, 1:14 AMSo you end up with workers that just hang when you give them work, which is FAR worse than just crashing and getting restarted - r
rough-minister-58256
04/06/2016, 1:14 AM - r
rough-minister-58256
04/06/2016, 1:14 AMheh nice - b
bored-art-40741
04/06/2016, 1:15 AMOh and that bug was closed at wontfix for 2.7.x, despite that being TEN YEARS AGO - b
bored-art-40741
04/06/2016, 1:15 AMThe solution was to upgrade to python3 - b
bored-art-40741
04/06/2016, 1:15 AMOne of the few times I've read that and truly wanted to table flip - r
rough-minister-58256
04/06/2016, 1:15 AMits a crying shame how things turned out for 2.7 vs 3.x - b
bored-art-40741
04/06/2016, 1:15 AMhmmm - b
bored-art-40741
04/06/2016, 1:15 AMIt looks like it just failed to talk to pypi I guess? - b
bored-art-40741
04/06/2016, 1:16 AMI've never seen that one specifically - r
rough-minister-58256
04/06/2016, 1:16 AMah I see - b
bored-art-40741
04/06/2016, 1:16 AMYeah, unfortunately the breaking features of py3 were combined with the breaking rewrite of the core C backend - b
bored-art-40741
04/06/2016, 1:17 AMThey should have done the backend rewrite as python 2.8 and then brought in all future syntax changes with
, which is an otherwise wonderful feature__future__ - m
mammoth-pharmacist-54215
04/06/2016, 1:17 AMrage telling people to upgrade to python 3.x is not a reasonable solution - b
bored-art-40741
04/06/2016, 1:17 AMYeah - b
bored-art-40741
04/06/2016, 1:17 AMI'm like, one of the biggest py3 apologists you will find out there. And sometimes I find py2 bugs closed as wontfix that make me properly rage4 - b
bored-art-40741
04/06/2016, 1:21 AMAlso @rough-minister-58256 unfortunately while you can restart jobs on jenkins, you can't restart individual shards - b
bored-art-40741
04/06/2016, 1:21 AMMy next change (aiming for this weekend) is to just codegen one job per shard - b
bored-art-40741
04/06/2016, 1:21 AMIt will greatly simplify the UI of this - b
bored-art-40741
04/06/2016, 1:22 AMAlso @rough-minister-58256 I hate to be the one to bear this news, but... http://jenkins.pantsbuild.org/job/pants_ci.unit_tests/SHARD=0/185/console - f
full-kilobyte-93429
04/06/2016, 1:22 AMswitching to in_memory didn't help, will need to disable more tests @rough-minister-58256 - r
rough-minister-58256
04/06/2016, 1:22 AMk, trying to get a repro under travis - it passed 1x