Assigning value on remote worker: SIGSEGV (regression, parallel)

[bermanmaxim]

I ran into this crash while testing the assignment of globals on remote workers

   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "help()" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.4.0-dev+6478 (2015-08-03 04:22 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 4d08b6d* (2 days old master)
|__/                   |  x86_64-linux-gnu

julia> addprocs(1);

julia> a = [1,2,3];

julia> @spawnat 2 global b = a;

julia> @everywhere println(b)

signal (11): Segmentation fault
unknown function (ip: 0x7f0ca8b36e86)
jl_f_isa at /home/maxim/julia/usr/bin/../lib/libjulia.so (unknown line)
typeinf_uncached at ./inference.jl:1481
unknown function (ip: 0x7f0ca59facc0)
jl_apply_generic at /home/maxim/julia/usr/bin/../lib/libjulia.so (unknown line)
typeinf at ./inference.jl:1340
unknown function (ip: 0x7f0ca59f5416)
jl_apply_generic at /home/maxim/julia/usr/bin/../lib/libjulia.so (unknown line)
typeinf_ext at ./inference.jl:1284
jl_apply_generic at /home/maxim/julia/usr/bin/../lib/libjulia.so (unknown line)
unknown function (ip: 0x7f0ca8b4a69d)
unknown function (ip: 0x7f0ca8b4b107)
jl_apply_generic at /home/maxim/julia/usr/bin/../lib/libjulia.so (unknown line)
sync_end at ./task.jl:406
anonymous at multi.jl:421
unknown function (ip: 0x7f0ca8bb4ceb)
jl_toplevel_eval_in at /home/maxim/julia/usr/bin/../lib/libjulia.so (unknown line)
eval_user_input at REPL.jl:63
jlcall_eval_user_input_21091 at  (unknown line)
jl_apply_generic at /home/maxim/julia/usr/bin/../lib/libjulia.so (unknown line)
anonymous at task.jl:92
unknown function (ip: 0x7f0ca8ba7580)
unknown function (ip: (nil))
fish: “julia” terminated by signal SIGSEGV (Address boundary error)

A git bisect narrowed the commit introducing this behavior to 4d08b6d ; before that, there is no SIGSEGV, only the expected output:

julia> @everywhere println(b)
exception on 1: ERROR: UndefVarError: b not defined
 in eval at sysimg.jl:14
 in anonymous at multi.jl:1321
 in run_work_thunk at multi.jl:584
 in remotecall_fetch at multi.jl:657
 in remotecall_fetch at multi.jl:672
 in anonymous at task.jl:365
    From worker 2:  [1,2,3]

[amitmurthy]

I get

  | | |_| | | | (_| |  |  Version 0.4.0-dev+6496 (2015-08-05 14:55 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 2171f80 (0 days old master)
|__/                   |  x86_64-linux-gnu

julia> addprocs(1);

julia> a = [1,2,3];

julia> @spawnat 2 global b = a;

julia> @everywhere println(b)
        From worker 2:  [1,2,3]
ERROR: UndefVarError(:b)
 in eval at sysimg.jl:14
 in anonymous at multi.jl:1350
 in run_work_thunk at multi.jl:643
 in remotecall_fetch at multi.jl:716
 in anonymous at task.jl:447
 in remotecall_fetch at multi.jl:717
 in anonymous at task.jl:447
 in sync_end at ./task.jl:413
 in anonymous at multi.jl:422

Could you do a git pull; make clean; make and try again?

[amitmurthy]

I don't have a Make.user

[amitmurthy]

Also can you print the output of versioninfo ?

Mine is

julia> versioninfo()
Julia Version 0.4.0-dev+6496
Commit 2171f80 (2015-08-05 14:55 UTC)
Platform Info:
  System: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i7-4770HQ CPU @ 2.20GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

[amitmurthy]

The underlying issue may be same - #12381

[bermanmaxim]

You were right, I don't see this problem anymore after make clean and rebuilding, I guess make alone wasn't enough (the bug seemed strangely consistent during my git bisect)... Thanks for the help !

julia> versioninfo()
Julia Version 0.4.0-dev+6511
Commit a2391b9 (2015-08-06 08:55 UTC)
Platform Info:
  System: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

[amitmurthy]

Please reopen if you see it again.