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

    rough-minister-58256

    02/03/2018, 5:48 AM
    ah nice
  • w

    witty-crayon-22786

    02/03/2018, 5:55 AM
    i believe that it eliminates the most interesting part of https://github.com/pantsbuild/pants/issues/4394, yea. after sketching it up further, i came to the conclusion that eliminating the actual file capture is by far the most important part... since all of the syscalls will be cached in the graph, the actual glob expansion is very unlikely to be a big deal.
  • w

    witty-crayon-22786

    02/03/2018, 5:56 AM
    ("actual file capture" meaning "file fingerprinting", which used to always operate at the "entire Snapshot" level.)
  • r

    rough-minister-58256

    02/03/2018, 5:56 AM
    word
  • r

    rough-minister-58256

    02/03/2018, 5:57 AM
    refresh my memory - did we ever end up implementing 
    Graph
    member eviction/lifetimes?
  • r

    rough-minister-58256

    02/03/2018, 5:57 AM
    getting slightly worried about memory util for the daemon in source
  • r

    rough-minister-58256

    02/03/2018, 5:57 AM
    have seen ~5GB sustained util
  • r

    rough-minister-58256

    02/03/2018, 5:58 AM
    even with OS compression it seems like that might be a tough squeeze on our majority-8GB userbase
  • r

    rough-minister-58256

    02/03/2018, 5:58 AM
    at least for anyone doing transitive dependees, transitive changed, list :: et al
  • w

    witty-crayon-22786

    02/03/2018, 6:05 AM
    no, we didn't implement eviction
  • r

    rough-minister-58256

    02/03/2018, 6:05 AM
    rgr
  • r

    rough-minister-58256

    02/03/2018, 6:07 AM
    the lease thing was for snapshot eviction I guess
  • w

    witty-crayon-22786

    02/03/2018, 6:07 AM
    yea.
  • w

    witty-crayon-22786

    02/03/2018, 6:07 AM
    and that did get implemented.
  • r

    rough-minister-58256

    02/03/2018, 6:07 AM
    yeh thats what I was thinking of
  • w

    witty-crayon-22786

    02/03/2018, 6:09 AM
    i know my machine has always been odd, but i still never see above 4GB sustained.
  • r

    rough-minister-58256

    02/03/2018, 6:09 AM
    still not that great
  • r

    rough-minister-58256

    02/03/2018, 6:09 AM
    git workspace
    + daemon + 
    ::
    could be pretty bad
  • w

    witty-crayon-22786

    02/03/2018, 6:10 AM
    so, remember that ~nobody does ::
  • r

    rough-minister-58256

    02/03/2018, 6:10 AM
    yeah, but they do it implicitly
  • w

    witty-crayon-22786

    02/03/2018, 6:10 AM
    but also, that most of the graph should be safe to page out if it isn't being actively used
  • w

    witty-crayon-22786

    02/03/2018, 6:10 AM
    hm, why...?
  • r

    rough-minister-58256

    02/03/2018, 6:10 AM
    transitive changed, transitive dependees, etc
  • r

    rough-minister-58256

    02/03/2018, 6:11 AM
    there are cases where full graph walks happen without the user logically thinking about addressing the whole repo
  • w

    witty-crayon-22786

    02/03/2018, 6:11 AM
    i highly doubt anyone is doing that.
  • r

    rough-minister-58256

    02/03/2018, 6:11 AM
    lets hope so
  • w

    witty-crayon-22786

    02/03/2018, 6:11 AM
    given that it would take ~2+ minutes pre-daemon
  • r

    rough-minister-58256

    02/03/2018, 6:11 AM
    but its a case that should be possible
  • w

    witty-crayon-22786

    02/03/2018, 6:11 AM
    and still ~30 seconds post daemon
  • r

    rough-minister-58256

    02/03/2018, 6:12 AM
    hmm
1...443444445...906Latest