https://pantsbuild.org/ logo
a

average-vr-56795

10/08/2019, 5:01 PM
@hundreds-father-404 Do we expect v2 test output to stream on travis?
h

hundreds-father-404

10/08/2019, 5:03 PM
No. It only outputs once every test has finished
a

average-vr-56795

10/08/2019, 5:04 PM
Ok, thanks! Is that our desired end-goal, or a temporary state?
h

hundreds-father-404

10/08/2019, 5:05 PM
Not sure! I personally would like it to be much more interactive. It’s frustrating that it takes so long to get any feedback beyond a binary “success” vs “failed” lot Cc @red-balloon-89377 who has been thinking about V2 UX iirc
w

witty-crayon-22786

10/08/2019, 5:05 PM
because
travis_wait
(the superficial issue is that
travis_wait
doesn't stream)
a

average-vr-56795

10/08/2019, 5:06 PM
We wouldn’t need it to stream if pants streamed, though, right?
h

hundreds-father-404

10/08/2019, 5:06 PM
Ah yes. I think I saw someone using Travis wait enhanced? That would be great to make a PR to always use
w

witty-crayon-22786

10/08/2019, 5:07 PM
the need to use travis wait is... odd. i haven't investigated, because we seem to be emitting log messages every 10 seconds or so, on average
haven't looked for outliers
h

hundreds-father-404

10/08/2019, 5:08 PM
I think it’s a vestige of when we only had 1 remote worker. Maybe we can safely remove it!
e

early-needle-54791

10/08/2019, 5:55 PM
ive been using travis wait enhanced to get around this issue
if/when my current PR lands we can use it everywhere.
(the issue of travis_wait buffering streams)
w

witty-crayon-22786

10/08/2019, 5:59 PM
@hundreds-father-404: didn't you try removing it at one point though? that wasn't in the "one worker" days, was it?
h

hundreds-father-404

10/08/2019, 6:01 PM
I might have. I think we stuck with it because if multiple PRs are up and we’re running ITs, there’s a risk that it takes more than ten minutes for an IT to complete. Some ITs take 15 minutes just to run.. How about we use enhanced Travis wait?
w

witty-crayon-22786

10/08/2019, 6:06 PM
fine with that.
it's not something to jump into without deeper thought, BUT, fixing this more fundamentally might mean incorporating some amount of logging (even with the "UI" disabled) here: https://github.com/pantsbuild/pants/blob/8fac40d5bb7b75da9ba26d7fa34e30c18b954e33/src/rust/engine/src/scheduler.rs#L311-L344 ...in case of long running tasks
in particular, the best we could do right now would be the
str(..)
of a node. which has been discussed elsewhere as being not the best representation.