https://pantsbuild.org/ logo
Powered by
#development
  • w

    witty-crayon-22786

    02/03/2017, 10:15 PM
    ah
  • w

    witty-crayon-22786

    02/03/2017, 10:16 PM
    wait, what is the second fork while daemonizing?
  • r

    rough-minister-58256

    02/03/2017, 10:16 PM
  • w

    witty-crayon-22786

    02/03/2017, 10:16 PM
    ...but yea, that would do it.
  • r

    rough-minister-58256

    02/03/2017, 10:17 PM
    standard unix daemonization strategy described in the docstring there
  • w

    witty-crayon-22786

    02/03/2017, 10:17 PM
    and the graph is created between forks? interesting.
  • r

    rough-minister-58256

    02/03/2017, 10:18 PM
    ah, so it’s pantsd_launcher->fork->fork->pantsd->fork->fork->pantsd-runner
  • r

    rough-minister-58256

    02/03/2017, 10:18 PM
  • w

    witty-crayon-22786

    02/03/2017, 10:18 PM
    yea
  • r

    rough-minister-58256

    02/03/2017, 10:19 PM
    we may not actually need that
  • w

    witty-crayon-22786

    02/03/2017, 10:19 PM
    yea, it hadn't occurred to me that the fork from the daemon to the backgrounded runner might matter too, but it does
  • w

    witty-crayon-22786

    02/03/2017, 10:20 PM
    because anything that runs synchronously in the runner can kick off tasks on the engine.
  • r

    rough-minister-58256

    02/03/2017, 10:20 PM
    since we’ve already double forked once, we could probably switch daemon_pants_runner to daemon_spawn which is single-fork: https://github.com/pantsbuild/pants/blob/master/src/python/pants/pantsd/process_manager.py#L450
  • w

    witty-crayon-22786

    02/03/2017, 10:20 PM
    yea.
  • w

    witty-crayon-22786

    02/03/2017, 10:20 PM
    don't see why not.
  • w

    witty-crayon-22786

    02/03/2017, 10:21 PM
    but i don't think that will fix this, because i think any fork will destroy the threads
  • r

    rough-minister-58256

    02/03/2017, 10:21 PM
    ah
  • r

    rough-minister-58256

    02/03/2017, 10:21 PM
    yea
  • r

    rough-minister-58256

    02/03/2017, 10:21 PM
    only the main thread is preserved during fork
  • w

    witty-crayon-22786

    02/03/2017, 10:21 PM
    yea. ok.
  • w

    witty-crayon-22786

    02/03/2017, 10:22 PM
    i'll look into making the pool more durable
  • w

    witty-crayon-22786

    02/03/2017, 10:23 PM
    alternatively, i guess we could explicitly re-init.
  • f

    fancy-queen-20734

    02/03/2017, 10:23 PM
    @witty-crayon-22786 cleaned up non-porting related comments in my PR. Fixed the broken tests, switched to enum. Mind taking another look? https://github.com/pantsbuild/pants/pull/4227
  • w

    witty-crayon-22786

    02/03/2017, 10:23 PM
    sure
  • f

    fancy-queen-20734

    02/03/2017, 10:25 PM
    I haven’t completely pulled the enum thread, but it’s very promising
  • w

    witty-crayon-22786

    02/03/2017, 10:42 PM
    on the bright side of this integration, unified logging ended up being hella easy
  • r

    rough-minister-58256

    02/03/2017, 10:45 PM
    heh
  • w

    witty-crayon-22786

    02/03/2017, 10:50 PM
    (haven't pushed that yet, but: added 
    externs.log(Level, msg)
    )
  • w

    witty-crayon-22786

    02/03/2017, 10:56 PM
    @rough-minister-58256 : ok, took the easy way out and just created a pool per 
    execute
    call.
  • w

    witty-crayon-22786

    02/03/2017, 10:56 PM
    threads are cheapish.
1...222223224...906Latest