segmentation fault (SIGSEGV) when computing neighbors

[Hrovatin]

Please make sure these conditions are met

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of scanpy.
• (optional) I have confirmed this bug exists on the main branch of scanpy.

What happened?

Issue in computing neighbors. Likely env dependent, but given env is fresh I was hoping you may provide some guidance

Fresh env:

python==3.10
numba==0.61.0
numpy==1.26.4
scanpy==1.11.0
scipy==1.15.2
# using CONDA_SUBDIR=osx-64 for mamba env creation on M3 chip
# Other packages omitted

Try running neighbors:

sc.pp.neighbors(emb,n_pcs=30)
/Users/karinhrovatin/miniforge3/envs/protein/lib/python3.10/site-packages/threadpoolctl.py:1226: RuntimeWarning: 
Found Intel OpenMP ('libiomp') and LLVM OpenMP ('libomp') loaded at
the same time. Both libraries are known to be incompatible and this
can cause random crashes or deadlocks on Linux when loaded in the
same Python program.
Using threadpoolctl may cause crashes or deadlocks. For more
information and possible workarounds, please see
    https://github.com/joblib/threadpoolctl/blob/master/multiple_openmp.md
  warnings.warn(msg, RuntimeWarning)
OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.
Process finished with exit code 139 (interrupted by signal 11: SIGSEGV)

The packages libiomp and libomp are not within conda env (at least not seen via pip freeze).
Already tried export OMP_NUM_THREADS=1 that was mentioned in another issue.

Minimal code sample

see above

Error output

Versions

See above

[ilan-gold]

But @Hrovatin what system are you on? Mac? Linux? Could you give a clean reproducer? If this is truly a dependency issue, it should be as simple as (a) install some packages and hten (b) attempt to run the code on a very small anndata (i.e., no need for data sharing)

[Hrovatin]

Yes, M3 Mac.

FYI: I am currently busy with another project so it may take me a while before I get back to this and generate a reproducible example in fresh env. Sorry for the delay

[LspongebobJH]

I met the same problem:

import scanpy as sc
data = sc.datasets.pbmc3k()
sc.pp.neighbors(data) # where error occurs.

Error message: zsh: segmentation fault python test.py
System: MacOS 14.3 with Apple M3
Scanpy version: 1.11.1
python version: 3.11.7

[ilan-gold]

In the meantime check out https://scanpy.readthedocs.io/en/stable/how-to/knn-transformers.html but I have asked if anyone has an M3 mac I can use to debug.

[Hrovatin]

I tried to reproduce again and can no longer achieve it.

I created conda env with yml scanpy_test.yml:

name: scanpy_test
channels:
  - conda-forge
dependencies:
  - python==3.10
  - numba==0.61.0
  - numpy==1.26.4
  - scanpy==1.11.0
  - scipy==1.15.2 

and comand CONDA_SUBDIR=osx-64 mamba env create -f scanpy_test.yml as we before needed the osx-64 version due to other dependencies.
Then I ran neighbours on pbmc3k as suggested by LspongebobJH.
The error did not occur this time.

The difference from a few months ago could be that my MacOS was updated a few times in between.

[LspongebobJH]

Thanks! I just found a workaround. I upgrade python version from 3.11.7 to 3.11.11 then bug disappears.

[flying-sheep]

This might be related scikit-learn/scikit-learn#31884