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

    aloof-angle-91616

    04/16/2018, 4:29 PM
    yeah i see that being very feasible (interface-wise, no clue how to implement)
  • a

    aloof-angle-91616

    04/16/2018, 4:29 PM
    (yet)
  • a

    aloof-angle-91616

    04/16/2018, 4:29 PM
    async-hostile how?
  • a

    average-vr-56795

    04/16/2018, 5:11 PM
    The interface would basically be a Path -> Inode HashMap… Trivial for readonly filesystems, but managing concurrency in the face of deletes is hard.
    a
    • 2
    • 2
  • a

    average-vr-56795

    04/16/2018, 5:11 PM
    Async-hostile because the trait functions require that you return a value, and has no way for you to defer that
  • a

    average-vr-56795

    04/16/2018, 5:12 PM
    The fuse (not fuse-mt) layer uses callbacks, which you can defer calling into, but fuse-mt layers eager return values on top of that
  • a

    average-vr-56795

    04/16/2018, 5:12 PM
    See also the description on https://github.com/pantsbuild/pants/pull/5705 🙂
  • w

    witty-crayon-22786

    04/17/2018, 10:56 AM
    from my understanding, 
    Future::wait
    is cheaper than it might be with eagerly executed futures
  • w

    witty-crayon-22786

    04/17/2018, 10:57 AM
    with eager futures like scala's, you'd be guaranteeing an immediate atomic read and potential context switch
  • w

    witty-crayon-22786

    04/17/2018, 10:58 AM
    but with 
    wait
    , you'd begin eagerly executing the work with no atomics and no context switches (until the actual async work was running)
  • w

    witty-crayon-22786

    04/17/2018, 10:59 AM
    so... maybe not a big deal here.
  • w

    witty-crayon-22786

    04/17/2018, 11:00 AM
    (just need to guarantee that you're allowed to block the thread you're on, which presumably you can here)
  • w

    witty-crayon-22786

    04/17/2018, 11:00 AM
    haven't reviewed it yet, sorry! eager to
  • a

    average-vr-56795

    04/17/2018, 11:00 AM
    No worries! Totally understand busy-ness! How’s Poland? 🙂
  • w

    witty-crayon-22786

    04/17/2018, 11:03 AM
    great so far!
  • a

    average-vr-56795

    04/17/2018, 11:04 AM
    My concern with the 
    wait
    thing is that you’re blocking the wrong threadpool; fuse-mt has a threadpool which it uses for all FS operations. Say only 
    read
    actually needs Futures - it means if you have 4 threads, and 4 `read`s are happening, you can no longer 
    stat
    or 
    readdir
    or 
    getattr
    until one of those `read`s completes. Whereas by having the I/O Futures running on their own threadpool, it’s only 
    read
    operations which are blocked on other 
    read
    operations
  • w

    witty-crayon-22786

    04/17/2018, 11:05 AM
    but it's a pool?
  • w

    witty-crayon-22786

    04/17/2018, 11:05 AM
    presumably because they're expecting you to block
  • w

    witty-crayon-22786

    04/17/2018, 11:06 AM
    @witty-crayon-22786 uploaded a file: Scalasphere 2018 - I See What You Did There.pdf
  • w

    witty-crayon-22786

    04/17/2018, 11:06 AM
    cc @happy-kitchen-89482: ^
  • a

    average-vr-56795

    04/17/2018, 11:07 AM
    It’s a single pool used for different operations which may have significantly different costs. In particular, it’s a pool which needs to be scheduled on even if the operation doesn’t require any I/O; which means cheap things may get queued behind expensive things
  • a

    average-vr-56795

    04/17/2018, 11:08 AM
    Whereas by having our own pool just for expensive things, and using callbacks scheduled on the pool, we don’t need to block cheap things behind expensive things
  • w

    witty-crayon-22786

    04/17/2018, 11:08 AM
    that sounds like a problem with the pool rather than with "what" blocks it
  • w

    witty-crayon-22786

    04/17/2018, 11:08 AM
    in the sense that the library would need more pools if it didn't want writes to block reads
  • w

    witty-crayon-22786

    04/17/2018, 11:09 AM
    (anyway, sorry... let me review. i'm probably missing something)
  • u

    user

    04/17/2018, 11:11 AM
    @witty-crayon-22786 commented on @witty-crayon-22786’s file https://pantsbuild.slack.com/files/U06A03HV1/FA811NWPM/scalasphere_2018_-_i_see_what_you_did_there.pdf: i'm not posting these to twitter quite yet, because i'm hoping that the video is more useful to people, and want to wait until that is out
  • a

    average-vr-56795

    04/17/2018, 11:15 AM
    Sorry, got distracted by your slides for a bit - that talk looks really cool 🙂
  • a

    average-vr-56795

    04/17/2018, 11:17 AM
    On the pool thing - yeah 🙂 the fuse-mt thing currently uses one pool, which is what I was contrasting with - A solvable problem if we wanted to change fuse-mt or write a new abstraction layer 🙂 But also, probably not worth the hassle
  • w

    witty-crayon-22786

    04/17/2018, 11:18 AM
    ooh. yea, sorry.
  • w

    witty-crayon-22786

    04/17/2018, 11:18 AM
    (sidenote... are writes even supported?)
1...474475476...906Latest