siibra cache is not thread-safe

[ch-schiffer]

I just encountered an error while running parallel siibra calls from many different processes (through snakemake):

[Errno 2] No such file or directory: '/home/cschiffer/.cache/siibra.retrieval/56458555fc977a3234ba8523b554bc2eced8d2594e2efe52462e52e23638de6b_temp.lock'
  File "/mnt/ssd1/code/bigbrain_dl_feature_analysis/snakemake/bigbrain_features/Snakefile", line 172, in __rule_cortical_layer_meshes
  File "/mnt/ssd1/code/siibra-python/siibra/__init__.py", line 26, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/core/__init__.py", line 16, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/core/atlas.py", line 16, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/core/space.py", line 20, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/locations/__init__.py", line 17, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/locations/point.py", line 16, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/locations/boundingbox.py", line 16, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/locations/pointset.py", line 18, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/retrieval/__init__.py", line 16, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/retrieval/repositories.py", line 16, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/retrieval/requests.py", line 16, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/retrieval/cache.py", line 136, in <module>
  File "/mnt/ssd1/code/siibra-python/siibra/retrieval/cache.py", line 66, in instance
  File "/mnt/ssd1/code/siibra-python/siibra/retrieval/cache.py", line 82, in run_maintenance
  File "/mnt/ssd1/code/siibra-python/siibra/retrieval/cache.py", line 82, in <listcomp>
  File "/home/cschiffer/mamba/envs/atlas/lib/python3.11/concurrent/futures/thread.py", line 58, in runRuleException:

The same error occurred for multiple parallel processes at the same time. This makes me assume that the file-based cache that siibra uses internally is probably not thread-safe. I had a quick look at the code of the cache, in particular the run_maintenance function. This function seems to delete some previously identified files from the cache. However, there is no check to handle the case where another process deletes a file after it was found in the folder. As I see it, this is indeed not safe to use from multiple processes/threads.
Checking whether a file still exists before deleting it, or putting a lock around cache modifications in general could solve this issue (although the latter probably would have some performance implications).

[ch-schiffer]

Hey @AhmetNSimsek , just wanted to check what the status here is. #496 was not merged as far as I understood, so was anything implemented to address thread safety? My ancient siibra version (0.4a77) is not thread-safe, but before (finally) going through the effort of updating, I wanted to be sure that it is worth it wrt. thread safety.

[ch-schiffer]

For future reference, setting export SKIP_CACHEINIT_MAINTENANCE=True to disable the cache maintenance altogether works as a workaround.