ThomasDelteil commented on issue #10042: [MXNET-86] Gluon dataloader crash on speech recognition training URL: https://github.com/apache/incubator-mxnet/issues/10042#issuecomment-374316037 When using num_workers > 0 I get after a few hundreds/thousands of batches (the higher the number of workers, the sooner the segfault): I am using mxnet-cu90 1.1.0: ``` Segmentation fault: 11 Stack trace returned 10 entries: [bt] (0) /home/ec2-user/anaconda3/lib/python3.6/site-packages/mxnet/libmxnet.so(+0x276938) [0x7fe86492c938] [bt] (1) /home/ec2-user/anaconda3/lib/python3.6/site-packages/mxnet/libmxnet.so(+0x28c53ae) [0x7fe866f7b3ae] [bt] (2) /lib64/libc.so.6(+0x353a0) [0x7fe8e4fe33a0] [bt] (3) /home/ec2-user/anaconda3/lib/python3.6/site-packages/mxnet/libmxnet.so(+0x28c2703) [0x7fe866f78703] [bt] (4) /home/ec2-user/anaconda3/lib/python3.6/site-packages/mxnet/libmxnet.so(+0x28c46d8) [0x7fe866f7a6d8] [bt] (5) /home/ec2-user/anaconda3/lib/python3.6/site-packages/mxnet/libmxnet.so(MXNDArrayCreateFromSharedMem+0x5f5) [0x7fe866a4f4c5] [bt] (6) /home/ec2-user/anaconda3/lib/python3.6/lib-dynload/../../libffi.so.6(ffi_call_unix64+0x4c) [0x7fe8d9299ec0] [bt] (7) /home/ec2-user/anaconda3/lib/python3.6/lib-dynload/../../libffi.so.6(ffi_call+0x22d) [0x7fe8d929987d] [bt] (8) /home/ec2-user/anaconda3/lib/python3.6/lib-dynload/_ctypes.cpython-36m-x86_64-linux-gnu.so(_ctypes_callproc+0x2ce) [0x7fe8d94ae82e] [bt] (9) /home/ec2-user/anaconda3/lib/python3.6/lib-dynload/_ctypes.cpython-36m-x86_64-linux-gnu.so(+0x12265) [0x7fe8d94af265] *** Error in `/home/ec2-user/anaconda3/bin/python': malloc(): memory corruption: 0x00007fe8380111f0 *** ``` Running this code: https://github.com/ThomasDelteil/CNN_NLP_MXNet/blob/master/Crepe-Gluon.ipynb and changing this line: `curr_loss = nd.mean(loss).asscalar()` to `curr_loss = nd.mean(loss)` Sometimes, not all the times, I also get the workers filling up 100% of my /dev/shm after the segfault. I am running the code in jupyter lab. Is this the same issue? Should I open a new one? The issue does not happen without multi-processing
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
