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

    rough-minister-58256

    03/20/2017, 9:35 PM
    @full-kilobyte-93429 sure, sec
  • r

    rough-minister-58256

    03/20/2017, 9:55 PM
    @full-kilobyte-93429
    Copy code
    RSS=4.0GB
real    13m57.684s
user    15m10.730s
sys     2m36.708s
  • f

    full-kilobyte-93429

    03/20/2017, 9:56 PM
    hmm...
  • w

    witty-crayon-22786

    03/20/2017, 9:57 PM
    big hmm
  • f

    full-kilobyte-93429

    03/20/2017, 9:57 PM
    i would hold off merging before we understand better
  • w

    witty-crayon-22786

    03/20/2017, 9:57 PM
    i have a theory, which is that the fact that the loop is batch-oriented affects concurrency
  • f

    full-kilobyte-93429

    03/20/2017, 9:58 PM
    doesn't explain memory
  • w

    witty-crayon-22786

    03/20/2017, 9:58 PM
    no, doesn't explain memory.
  • f

    full-kilobyte-93429

    03/20/2017, 9:58 PM
    if so the fs io could have similar perf implications?
  • w

    witty-crayon-22786

    03/20/2017, 10:00 PM
    yea.
  • w

    witty-crayon-22786

    03/20/2017, 10:01 PM
    which is interesting.
  • f

    full-kilobyte-93429

    03/20/2017, 10:02 PM
    can do a full heap profile now at least for the memory portion
  • r

    rough-minister-58256

    03/20/2017, 10:02 PM
    k
  • f

    full-kilobyte-93429

    03/20/2017, 10:03 PM
    it's non negligible 4G vs 3.5
  • w

    witty-crayon-22786

    03/20/2017, 10:04 PM
    yea, i have zero theories there. 1) fewer Nodes, 2) fewer python values
  • r

    rough-minister-58256

    03/20/2017, 10:11 PM
    if it helps, there seems to be a burst period just after all the build parsing warnings are printed where it jumps from 3.6G->4.0G in the last ~30 seconds of the run.
  • r

    rough-minister-58256

    03/20/2017, 10:12 PM
    too bad we don’t have workunit segments to attribute per-step timings to in the engine
  • r

    rough-minister-58256

    03/20/2017, 10:12 PM
    tho we could probably track some kind of timing metric there and spit it out at the end
  • r

    rough-minister-58256

    03/20/2017, 10:13 PM
    spent N seconds computing product X etc
  • w

    witty-crayon-22786

    03/20/2017, 10:13 PM
    @full-kilobyte-93429 : i'm going to prototype changing the fs loop into future recursion, and see what that looks like.
  • w

    witty-crayon-22786

    03/20/2017, 10:13 PM
    yea, in theory there is a ton of data available. just need to figure out how to summarize it
  • r

    rough-minister-58256

    03/20/2017, 10:13 PM
    yeh
  • f

    full-kilobyte-93429

    03/20/2017, 10:13 PM
    ok
  • r

    rough-minister-58256

    03/20/2017, 10:24 PM
    kicked off a heap profile, should have some results late tonight or in the morning
  • w

    witty-crayon-22786

    03/20/2017, 10:36 PM
    nothing odd about the visualization that i can see
  • r

    rough-minister-58256

    03/20/2017, 10:52 PM
    whoa
  • w

    witty-crayon-22786

    03/20/2017, 10:53 PM
    (it should look completely different post ff544118a5dff74da5e02c7fb2f4585bef1fb002)
  • r

    rough-minister-58256

    03/20/2017, 10:54 PM
    looks like I’m seeing closer to ~40% improvement for the shared race fix
  • r

    rough-minister-58256

    03/20/2017, 10:54 PM
    verifying
  • r

    rough-minister-58256

    03/20/2017, 10:57 PM
    looks like 4343 is good to merge?
1...284285286...906Latest