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

    rough-minister-58256

    03/02/2017, 8:28 PM
    aha
  • r

    rough-minister-58256

    03/02/2017, 8:30 PM
    on a quick instruments sample I took yesterday I noticed some weird patterns in the allocation call tree
  • r

    rough-minister-58256

    03/02/2017, 8:31 PM
    @User uploaded a file:

    Pasted image at 2017-03-02, 12:30 PM▾

  • r

    rough-minister-58256

    03/02/2017, 8:31 PM
    seems to be hopping up by 10s of MB at a time in the future stacks
  • r

    rough-minister-58256

    03/02/2017, 8:31 PM
    not sure if futures are just obscuring some underlying work happening there or what - not sure how to interpret this
  • w

    witty-crayon-22786

    03/02/2017, 8:31 PM
    that call stack is what used to be expected when the 
    main
    thread was executing all scheduling
  • r

    rough-minister-58256

    03/02/2017, 8:32 PM
    ah
  • w

    witty-crayon-22786

    03/02/2017, 8:32 PM
    basically, the main thread would recurse through the entire graph until it encountered an IO operation, and then the IO would happen on an IO thread
  • r

    rough-minister-58256

    03/02/2017, 8:33 PM
    ah
  • w

    witty-crayon-22786

    03/02/2017, 8:33 PM
    with the change i made to avoid the 
    Shared
    bug (*edit), each Node runs on the CpuPool, and so they should have their own stacks now
  • w

    witty-crayon-22786

    03/02/2017, 8:33 PM
    it's less performant because context switching / locking, but it should be easier to interpret
  • r

    rough-minister-58256

    03/02/2017, 8:34 PM
    cool, will take another peek
  • w

    witty-crayon-22786

    03/02/2017, 9:36 PM
    @salmon-summer-38098 : how is it going?
  • s

    salmon-summer-38098

    03/02/2017, 10:53 PM
    @witty-crayon-22786: set HipChat to afk, forgot about here :/
  • s

    salmon-summer-38098

    03/02/2017, 10:53 PM
    going ok - just getting my bearings in the code right now
  • w

    witty-crayon-22786

    03/02/2017, 11:09 PM
    ok, let me know if you have questions.
  • w

    witty-crayon-22786

    03/02/2017, 11:09 PM
    @salmon-summer-38098 : @full-kilobyte-93429 is going to start by splitting 
    SelectDependencies(, transitive=true/false)
    into two different selectors
  • w

    witty-crayon-22786

    03/02/2017, 11:09 PM
    ie, split out 
    SelectTransitive
  • w

    witty-crayon-22786

    03/02/2017, 11:09 PM
    cc @fancy-queen-20734
  • w

    witty-crayon-22786

    03/02/2017, 11:10 PM
    that might touch what you are doing as well.
  • s

    salmon-summer-38098

    03/02/2017, 11:16 PM
    ah I see, makes sense @full-kilobyte-93429 - let's discuss in standup tomorrow and we can keep each other posted that way too
  • f

    full-kilobyte-93429

    03/02/2017, 11:18 PM
    sure, for me that's a separate task after the 
    SelectDependencies
    perf improvement
  • w

    witty-crayon-22786

    03/02/2017, 11:19 PM
    er
  • w

    witty-crayon-22786

    03/02/2017, 11:19 PM
    it's first, yea?
  • w

    witty-crayon-22786

    03/02/2017, 11:19 PM
    @full-kilobyte-93429 : ^
  • f

    full-kilobyte-93429

    03/02/2017, 11:24 PM
    that is ok too, i didn't realize the ordering change, now it probably makes sense
  • w

    witty-crayon-22786

    03/02/2017, 11:39 PM
  • r

    rough-minister-58256

    03/02/2017, 11:40 PM
    whoa nice
  • r

    rough-minister-58256

    03/02/2017, 11:40 PM
    wonder if this works w/ system alloc
  • r

    rough-minister-58256

    03/02/2017, 11:41 PM
    oh i guess so
1...260261262...906Latest