remotecall_fetch hangs when CPU is busy on the remote host #16874
Closed
davidparks21 opened this Issue on Jun 10 · 1 comment
Projects
None yet
Labels
parallel
Milestone
No milestone
Assignees
No one assigned
3 participants
@davidparks21
@amitmurthy
@tkelman
Notifications

You’re not receiving notifications from this thread.
@davidparks21
davidparks21 commented on Jun 10 • edited

I posted this originally as a forum question, but after reproducing it on multiple machines, I'm confident that it's an issue that needs to be addressed.

https://groups.google.com/forum/#!topic/julia-users/pH-CuZlspic

remotecall_fetch (and generally remotecall) hangs indefinitely when the remote CPU is busy.

The code below demonstrates a simple example that can be run locally to reproduce the issue.

addprocs(1)

@everywhere module Test
       stop_flag = false
       function runner() global stop_flag; while(!stop_flag) x=rand(5000,5000); end end
       function stopper() global stop_flag; stop_flag = true; end
       end

remotecall(2, Test.runner)
remotecall_fetch(2, Test.stopper)

# Hangs here indefinitely

I've attempted to start the remote process asynchronously, but it doesn't change the results.

It seems likely that the remotecall listener does not have a high enough thread priority. Though I haven't actually looked into the mechanics of it myself.

julia> versioninfo()
Julia Version 0.4.5
Commit 2ac304d (2016-03-18 00:58 UTC)
Platform Info:
  System: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

@tkelman tkelman added the parallel label on Jun 10
@amitmurthy
The Julia Language member
amitmurthy commented on Jun 12

There is a single thread of execution. Tasks are light-weight coroutines scheduled co-operatively on the single OS thread. There is no pre-emption of running tasks. Tasks yield either on IO or by calling yield() explicitly. If a task does not yield (your example above), no other tasks can execute.

Changing function runner() global stop_flag; while(!stop_flag) x=rand(5000,5000); end end to

function runner()
    global stop_flag
    while(!stop_flag)
        yield()
        x=rand(5000,5000)
    end
end

will work as intended.
@amitmurthy amitmurthy closed this on Jun 12