Libuv and tasks are not thread safe

[ranjanan]

Doing
@threads for i = 1:10 sleep(2) end
regularly segfaults.

This is the backtrace. I understand that sleep() doesn't apply to threads and applies to tasks, but I just thought I'd post this here:

#0  0x00007ffff6e248c9 in jl_apply (nargs=0, args=0x0, f=0x0) at /home/ranjan/julia-threading5/src/julia.h:1375
#1  start_task () at /home/ranjan/julia-threading5/src/task.c:246
#2  0x0000000000000000 in ?? ()

[yuyichao]

Right. Task switches and IO's are the only two segfaults left with threading that I've reproduced.

[vtjnash]

You can use Libc.systemsleep(2) for a blocking pause. sleep also tries to access/modify libuv state, which is not threadsafe.

[jeff-regier]

Is this a Julia 0.6 issue? Or Julia 1.0?

[yuyichao]

It will not block 0.6

[Keno]

There's two issues here. This first is a fully thread-safe multi-threaded I/O system which is a big issue. The second is just the original report that sleep doesn't work right, which is related, but could probably be fixed differently (e.g. if on a thread sleep does systemsleep). Which one were you asking about?

[jeff-regier]

Thanks for the responses. I was wondering about the former--thread-safe IO in general. Seems like this issue has become a catch-all for libuv+threading issues, with others closed as duplicates.

[xiuliren]

just tested in Julia 1.0.2, still have the segfault error.

[ViralBShah]

This is not expected to work until PARTR multi-threading is officially announced as available.

[StefanKarpinski]

Specifically, this issue will be closed when #29706 is merged.

[StefanKarpinski]

@JeffBezanson: close this now?

[JeffBezanson]

Fixed by #32174 and #32309.