segfault when calling rmprocs   after  parallel for loop  7376 Closed felixjung opened this Issue on Jun 23, 2014 · 5 comments Projects None yet Labels parallel Milestone No milestone Assignees No one assigned 4 participants  felixjung  JeffBezanson  jiahao  ViralBShah Notifications    felixjung felixjung commented on Jun 23, 2014  Hi,  I've noted that Julia 0.3.0-prerelease+3789 segfaults when I try to remove workers after running the following parallelised code:    Add worker cores addprocs 10     Create distributed array d   dzeros  100, 100 , workers  ,  1, nworkers        Fill the DArray in a parallel loop  parallel for w   1:length d.chunks      Obtain local part of array on worker w   dl   localpart d       Fill localpart   for i   1:size dl, 1      for j  1:size dl, 2        dl i, j    randn       end   end end    Remove the worker cores rmprocs workers      This will cause julia to segfault  The code creates a DArray, where the second dimension is split between workers. During the parallel for loop, each worker fills its local part of the DArray with random numbers by iterating through all array indices. Finally, I try to remove worker nodes using rmprocs  .  The exact error I get is  julia  rmprocs workers    :ok  julia   From worker 2:   1:37,1:251,1:100       From worker 3:   1:37,1:251,101:200  From worker 13:  1:37,1:251,1101:1200       From worker 10:  1:37,1:251,801:900  From worker 7:   1:37,1:251,501:600     From worker 11:  1:37,1:251,901:1000  From worker 19:  1:37,1:251,1701:1800   From worker 20:  1:37,1:251,1801:1900  From worker 6:   1:37,1:251,401:500 From worker 16:  1:37,1:251,1401:1500  From worker 5:   1:37,1:251,301:400  From worker 15:  1:37,1:251,1301:1400  1     51187 segmentation fault  julia  I've tried whether this problem also occurs when just summing up random numbers in a  parallel for loop, i.e.:  addprocs 10  foo    parallel  +  for 1:100   randn   end rmprocs workers     However, I don't get a segfault in this case. Seems like the segmentation fault has something to do with the way I write to the DArray, or the DArray itself.  The example code runs fine in Julia 0.21. I've installed Julia through Homebrew using brew install --HEAD julia --64bit. This is my Hardware setup. screen shot 2014-06-23 at 09 41 53  Any ideas?  JeffBezanson The Julia Language member JeffBezanson commented on Jun 23, 2014  Why is there text output from the workers? The code doesn't seem to do any printing.  jiahao The Julia Language member jiahao commented on Jun 23, 2014  I can reproduce the segfault, but not with julia-debug, lldb julia or gdb julia.  felixjung felixjung commented on Jun 24, 2014   JeffBezanson Not sure. However, I just realised that the text output are the local indices  myindexes    of each worker in the computation I initially noticed the problem in. I don't know, if I accidentally pasted the wrong output, or if the bug is actually related to workers already having local parts of another DArray.  I also just noticed, that my example code does no longer produce a segfault on my machine. Not sure why though.  JeffBezanson The Julia Language member JeffBezanson commented on Jul 27, 2014  Reopen if this turns out to be reproducible.  JeffBezanson JeffBezanson closed this on Jul 27, 2014  ViralBShah ViralBShah added the parallel label on Jul 27, 2014  ViralBShah The Julia Language member ViralBShah commented on Jul 27, 2014  Cc:  amitmurthy