pantsbuild #general

Channels

welcome

github-notifications

announce

witty-crayon-22786

12/08/2016, 1:41 AM

should grow to a bounded number and then stop

witty-crayon-22786

12/08/2016, 1:41 AM

approximately 1 per jvm tool that is used

rough-minister-58256

12/08/2016, 1:48 AM

here’s the email diff between assigned vs reviewers:

rough-minister-58256

12/08/2016, 1:48 AM

@User uploaded a file:

Pasted image at 2016-12-07, 5:48 PM▾

rough-minister-58256

12/08/2016, 1:48 AM

using Reviewers seems better

rough-minister-58256

12/08/2016, 1:49 AM

@thankful-library-32113 would be good to collect full process entries from ps and look at their fingerprints/identities etc

thankful-library-32113

12/08/2016, 1:49 AM

ok, i'll keep an 👀 out

rough-minister-58256

12/08/2016, 1:49 AM

the only case I can think of where “crazy #s” would happen would be if someone were obliterating their .pids (e.g. a

git clean

could do this) between pants runs and even then clean/ng-killall should enumerate the ps table and excise them

rough-minister-58256

12/08/2016, 1:51 AM

but it’d be good to collect the full ps entries w/ cli args + the state of the users $repo/.pids

enough-analyst-54434

12/08/2016, 2:15 AM

The ngs are still per clone, so if the workflow has folks, say using 5 clones of 1 repo, then its 5 x 1 nailgun per tool used., so maybe 25 or so nailguns total in this example.

limited-country-20626

12/08/2016, 6:14 PM

@fancy-queen-20734 - you should be able to fix by setting encoding

limited-country-20626

12/08/2016, 6:14 PM

we set

fancy-queen-20734

12/08/2016, 6:14 PM

on travis?

limited-country-20626

12/08/2016, 6:14 PM

export PYTHONIOENCODING="utf-8"

limited-country-20626

12/08/2016, 6:15 PM

I did not hit the link to get details so I didn't see the source of the exception

limited-country-20626

12/08/2016, 6:15 PM

okay

fancy-queen-20734

12/08/2016, 6:15 PM

ah, so isort.sh on travis occasionally fails because a spec starts with 0xe2

limited-country-20626

12/08/2016, 6:15 PM

I would think that to be a reasonable thing to add

limited-country-20626

12/08/2016, 6:16 PM

since we attempt to enforce through headers. But maybe leaving unset is useful in that it surfaces weird cases like that

astonishing-jelly-60479

12/08/2016, 6:35 PM

should pex chroots by default be isolated from the system site packages?

astonishing-jelly-60479

12/08/2016, 6:36 PM

seeing some strange behavior crop up where system site packages are added to the pythonpath for tests; there's an systemwide install of six at an older version that gets imported instead of the new version we specify

astonishing-jelly-60479

12/08/2016, 6:39 PM

also, on the nailgun discussion yesterday... in our CI environment we run pants with multiple clones per host (8 to be exact) and end up with lots of nailgun processes eating several GB RAM each after a while

rough-minister-58256

12/08/2016, 7:01 PM

@astonishing-jelly-60479 sounds like https://github.com/pantsbuild/pex/issues/302

rough-minister-58256

12/08/2016, 7:23 PM

also, you guys can always use

--kill-nailguns

or add a

./pants clean-all

./pants ng-killall

at the end of your CI jobs

fancy-queen-20734

12/08/2016, 7:23 PM

my hypothesis for the ascii error that’s popping up now is that the ./pants changed call in isort.sh is the first pants call, so it sets up rust. The changed call is teed into a file that is supposed to only contain target specs. But if any portion of the Rust setup writes to stdout, it’ll corrupt that file.

fancy-queen-20734

12/08/2016, 7:24 PM

after the cache is available, it’s no longer an issue

fancy-queen-20734

12/08/2016, 7:24 PM

which is why it’s intermittent

rough-minister-58256

12/08/2016, 7:24 PM

hrm interesting

fancy-queen-20734

12/08/2016, 7:24 PM

if another shard runs first, then it will pass.

astonishing-jelly-60479

12/08/2016, 7:41 PM

@rough-minister-58256 didn't know about the

--kill-nailguns

flag; thanks