Open
Description
I keep running into Segmentation fault errors.
This happens most probably during calls to the OpenMM Python API.
Here is my "error message"
Stacktrace
[13736] signal 11 (1): Segmentation fault
in expression starting at none:0
_PyInterpreterState_GET at /usr/local/src/conda/python-3.12.8/Include/internal/pycore_pystate.h:133 [inlined]
get_state at /usr/local/src/conda/python-3.12.8/Objects/obmalloc.c:866 [inlined]
_PyObject_Free at /usr/local/src/conda/python-3.12.8/Objects/obmalloc.c:1850 [inlined]
PyObject_Free at /usr/local/src/conda/python-3.12.8/Objects/obmalloc.c:830
_buffer_info_free at /data/numerik/people/bzfsikor/conda/envs/conda_jl/lib/python3.12/site-packages/numpy/core/_multiarray_umath.cpython-312-x86_64-linux-gnu.so (unknown line)
array_dealloc at /data/numerik/people/bzfsikor/conda/envs/conda_jl/lib/python3.12/site-packages/numpy/core/_multiarray_umath.cpython-312-x86_64-linux-gnu.so (unknown line)
pydecref_ at /data/numerik/people/bzfsikor/software/julia_depot/packages/PyCall/1gn3u/src/PyCall.jl:118
pydecref at /data/numerik/people/bzfsikor/software/julia_depot/packages/PyCall/1gn3u/src/PyCall.jl:123
jfptr_pydecref_4550 at /data/numerik/people/bzfsikor/software/julia_depot/compiled/v1.11/PyCall/GkzkC_ddiUX.so (unknown line)
run_finalizer at /cache/build/builder-demeter6-3/julialang/julia-release-1-dot-11/src/gc.c:299
jl_gc_run_finalizers_in_list at /cache/build/builder-demeter6-3/julialang/julia-release-1-dot-11/src/gc.c:389
run_finalizers at /cache/build/builder-demeter6-3/julialang/julia-release-1-dot-11/src/gc.c:435
enable_finalizers at ./gcutils.jl:161 [inlined]
unlock at ./lock.jl:178 [inlined]
macro expansion at ./lock.jl:275 [inlined]
#282 at /home/htc/bzfsikor/.julia/juliaup/julia-1.11.3+0.x64.linux.gnu/share/julia/stdlib/v1.11/REPL/src/LineEdit.jl:2851
jfptr_YY.282_9263 at /data/numerik/people/bzfsikor/software/julia_depot/compiled/v1.11/REPL/u0gqU_dovaC.so (unknown line)
jl_apply at /cache/build/builder-demeter6-3/julialang/julia-release-1-dot-11/src/julia.h:2157 [inlined]
start_task at /cache/build/builder-demeter6-3/julialang/julia-release-1-dot-11/src/task.c:1202
Allocations: 775171170 (Pool: 775164164; Big: 7006); GC: 5349
fish: Job 1, 'env JULIA_HISTORY=./.history.jl…' terminated by signal SIGSEGV (Address boundary error)I have no idea how to investigate this further.
Metadata
Metadata
Assignees
Labels
No labels
Type
Projects
Milestone
Relationships
Development
No branches or pull requests
Participants
Activity
[-]SegmentationFault[/-][+]Segmentation Fault[/+]axsk commentedon Jan 24, 2025
And here is another one, which I run into more often. This puzzles me especially since it somehow involves CUDA as well..
Stacktrace
axsk commentedon Jan 24, 2025
I switched to a single threaded instance and have not yet observed this issue.
However, I don't make any (explicit) use of multi-threading nowhere in my code..
axsk commentedon Mar 4, 2025
Whereas above examples happened randomly in my training loop, I can now reproduce the problem by tab-completing a PyObjets fields:
Stacktrace
Again, this only happens with multiple threads, when starting Julia with
--threads=1it works fine.bhawkins commentedon Apr 8, 2025
Here's a simple script that reliably triggers a similar segfault for me:
Output on Linux with Julia 1.11.4 and Python 3.12.9
Output on macos with Julia 1.11.3 and Python 3.13.2
On my mac this crashes about 90% of the time. If I comment out the
pyimportthen the probability of crashing goes down somewhat, maybe 70%. If I comment both the first two lines then it doesn't crash. I'm not sure how important the function body is, but it doesn't seem to crash if I have a parallel loop with justsleep(1)in it.bhawkins commentedon Apr 8, 2025
Using the
pylock()suggestion here makes the crash go away. My example isn't actually calling any Python, but I guess gc can run anyway.