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

    happy-kitchen-89482

    12/23/2019, 9:22 PM
    AFAICT it occurs when developing in the pants repo itself, I'm not sure if we've seen it happen in other repos. So it may be some pathology related to running pants in the pants repo, and doesn't affect users at large. But it's still becoming a showstopper for many of us.
    h
    d
    • 3
    • 3
  • h

    happy-kitchen-89482

    12/24/2019, 8:00 AM
    Also can someone take a look at https://github.com/pantsbuild/pants/issues/8861
  • h

    happy-kitchen-89482

    12/24/2019, 8:00 AM
    thanks
  • h

    happy-kitchen-89482

    12/24/2019, 8:00 AM
    (someone with engine internals experience that is)
  • h

    happy-kitchen-89482

    12/24/2019, 1:39 PM
    Also, note an issue with engine rule validation errors being non-deterministic: https://github.com/pantsbuild/pants/issues/8862
    a
    • 2
    • 1
  • a

    aloof-angle-91616

    12/25/2019, 7:23 AM
    https://github.com/pantsbuild/pants/pull/8859 <= use all your favorite 
    BinaryToolBase
    instances in v2! appears to have some errors with rule resolution (not being able to locate the 
    OptionsBootstrapper
    Param maybe??) which i will hopefully figure out during the flight tomorrow!
  • a

    aloof-angle-91616

    12/25/2019, 7:26 AM
    also have a neat as HECK diff to make use of bash scripts in v2 process executions really easily (to do things like move files around after calling another process, to avoid having to create special intrinsics to do those sorts of small things). it adds a 
    Bash
    instance of 
    NativeTool
    and is therefore remotable! going to add testing on the flight tomorrow too (that's three things i've promised to do on this flight so far, so apply a few grains of salt)
  • h

    hundreds-breakfast-49010

    12/27/2019, 6:17 PM
    is there any reason why binaries run with 
    ./pants run
    have to have their working directory set to the build root? I built the 
    InteractiveProcessRunner
    to do this because v1 pants did this, but is there any actual requirement that programs have to run there?
    e
    • 2
    • 5
  • h

    hundreds-breakfast-49010

    12/27/2019, 11:20 PM
    can anyone confirm that the 
    No space left on device
    error that various people were seeing when running pants in the pants repo is still happening for them?
  • h

    happy-kitchen-89482

    12/28/2019, 9:21 PM
    I saw it yesterday, not since then
  • h

    happy-kitchen-89482

    12/30/2019, 2:41 PM
    And again yesterday (Sunday)
    h
    • 2
    • 2
  • h

    happy-kitchen-89482

    12/30/2019, 2:41 PM
  • h

    happy-kitchen-89482

    12/30/2019, 2:42 PM
    I think this gets us very close to feature parity with the v1 python pipeline!
    đź’Ż 3
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:33 PM
    okay, re: 
    No space left on device
    error, I think I know what's going on here
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:35 PM
    when we run python tests with the v2 engine, we're including 
    native_engine.so
    in the 
    input_files
    Digest of the ExecuteProcessRequest that runs pytest, and we spawn many of these as we're running tests
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:35 PM
    native_engine.so
    is 159MB
  • a

    aloof-angle-91616

    12/30/2019, 10:35 PM
    is this not deduplicated?
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:35 PM
    which is by far the biggest file in any of these 
    process-execution-*
    directories
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:36 PM
    @aloof-angle-91616 no, I'm not sure how that deduplication would work given our current process execution model
    đź‘Ť 1
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:36 PM
    each instance of EPR is separate from every other and has a Digest materialized to disk for its own work while it's running
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:36 PM
    so native_engine.so gets materialized separately for each EPR just like any other file. the only problem is that it's pretty big
  • a

    aloof-angle-91616

    12/30/2019, 10:37 PM
    i was under the impression that the filesystem abstraction was a merkle tree? we could symlink?
  • w

    witty-crayon-22786

    12/30/2019, 10:37 PM
    by default it should be deleted after the process runs.
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:38 PM
    @witty-crayon-22786 I think that is happening, but we're running many of these EPRs at once, so if enough get created before any of them are done and the tmpdir is deleted, we run out of disk space in /tmp
    đź‘Ť 1
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:38 PM
    this is consistent with what I was seeing on my linux system with /tmp filling up, I'm not sure if Macs treat their /tmp dir differently in a relevant way
  • w

    witty-crayon-22786

    12/30/2019, 10:38 PM
    you'd need to run many thousands of concurrent processes to run out of temp space though...?
  • a

    aloof-angle-91616

    12/30/2019, 10:39 PM
    is there not a way to tell the os to actually delete it? this seems like we’re not doing anything too absurd and it feels like something pathological is happening
  • w

    witty-crayon-22786

    12/30/2019, 10:39 PM
  • a

    aloof-angle-91616

    12/30/2019, 10:40 PM
    are we actually clear on what leads to the “no space left on device” error? i am not familiar enough with osx to understand why this error message is happening when the files are being deleted
  • h

    hundreds-breakfast-49010

    12/30/2019, 10:41 PM
    @witty-crayon-22786 on my system the size of the tmpfs mounted at /tmp is 4GB
1...682683684...906Latest