numba: readthedocs build failure: cannot cache function: no locator available for file

My project uses numba 0.43.1 (from conda-forge) and has functions decorated with @numba.jit('<signature>', nopython=True, nogil=True, cache=True). Everything works fine locally, on Travis, and AppVeyor. readthedocs fails as soon as it tries importing the modules, with the following log:

WARNING: autodoc: failed to import module 'copula' from module 'pyscenarios'; the following exception was raised:
Traceback (most recent call last):
  File "/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/sphinx/ext/autodoc/importer.py", line 232, in import_module
    __import__(modname)
  File "<frozen importlib._bootstrap>", line 983, in _find_and_load
  File "<frozen importlib._bootstrap>", line 967, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 668, in _load_unlocked
  File "<frozen importlib._bootstrap>", line 638, in _load_backward_compatible
  File "/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/pyscenarios-0.2.0-py3.7.egg/pyscenarios/copula.py", line 12, in <module>
    from .sobol import sobol
  File "<frozen importlib._bootstrap>", line 983, in _find_and_load
  File "<frozen importlib._bootstrap>", line 967, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 668, in _load_unlocked
  File "<frozen importlib._bootstrap>", line 638, in _load_backward_compatible
  File "/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/pyscenarios-0.2.0-py3.7.egg/pyscenarios/sobol.py", line 67, in <module>
    def _calc_v_kernel(directions: np.ndarray) -> np.ndarray:
  File "/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/numba/decorators.py", line 191, in wrapper
    disp.enable_caching()
  File "/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/numba/dispatcher.py", line 566, in enable_caching
    self._cache = FunctionCache(self.py_func)
  File "/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/numba/caching.py", line 614, in __init__
    self._impl = self._impl_class(py_func)
  File "/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/numba/caching.py", line 349, in __init__
    "for file %r" % (qualname, source_path))
RuntimeError: cannot cache function '_calc_v_kernel': no locator available for file '/home/docs/checkouts/readthedocs.org/user_builds/pyscenarios/conda/v0.2.0/lib/python3.7/site-packages/pyscenarios-0.2.0-py3.7.egg/pyscenarios/sobol.py'

I’ve worked around the problem with the following hack:

def jit(sig: str):
    def decorator(func):
        try:
            return numba.jit(sig, nopython=True, nogil=True, cache=True)(func)
        except RuntimeError:
            # Work around issues with permissions on readthedocs
            return numba.jit(sig, nopython=True, nogil=True, cache=False)(func)
    return decorator

About this issue

Original URL
State: closed
Created 5 years ago
Comments: 20 (8 by maintainers)

Commits related to this issue

Fix cannot write to numba_cache_dir Following error occured during execution of the nf-pipeline: "cannot cache function: no locator available for file" Fix from https://github.com/numba/numba/issues... — committed to vjmarteau/scrnaseq-calprotectin-organoids by vjmarteau 2 years ago
fix(cid): set NUMBA_CACHE_DIR to runner.temp without this ```pytb cannot cache function 'sparse_mean_var_minor_axis': no locator ``` https://github.com/numba/numba/issues/4032#issuecomment-48810270... — committed to pinellolab/pyrovelocity by cameronraysmith 6 months ago
fix(cid): set NUMBA_CACHE_DIR to runner.temp without this ```pytb cannot cache function 'sparse_mean_var_minor_axis': no locator ``` https://github.com/numba/numba/issues/4032#issuecomment-48810270... — committed to pinellolab/pyrovelocity by cameronraysmith 6 months ago
fix(cid): set NUMBA_CACHE_DIR to runner.temp without this ```pytb cannot cache function 'sparse_mean_var_minor_axis': no locator ``` https://github.com/numba/numba/issues/4032#issuecomment-48810270... — committed to pinellolab/pyrovelocity by cameronraysmith 6 months ago

Most upvoted comments

@SumNeuron thanks for the workaround. i found though that NUMBA_CACHE_DIR was not persisted to the container because it was in a RUN command. so i put it in an ENV and it worked beautifully. also found that i did not have to actualy create the numba_cache directory either, so i deleted the rest of the RUN command. so i ended up simply with:

ENV NUMBA_CACHE_DIR=/tmp/numba_cache

+13

bjarthur on Nov 18, 2020

@esc that may be… but I have other .env vars in that file that do get evaluated. I can double check this tomorrow.

I can add a layer to the Dockerfile:

RUN mkdir /tmp/numba_cache & chmod 777 /tmp/numba_cache & NUMBA_CACHE_DIR=/tmp/numba_cache

I am still of the opinion that in the next version, this is added as fallback by numba - with a warning readout. This will prevent individuals from having to rebuild dockerfiles just to make sure a cache dir is set up. Yes it is a quality of life feature and not strictly required, but I would sure appreciate it ^•^

SumNeuron on Oct 28, 2019

This error is very strange, as it suggests that the cache cannot write to any of the standard locations on readthedocs. Assuming that /tmp is writeable, you could try setting the environment variable NUMBA_CACHE_DIR to /tmp.

If that works, let us know and we can put that suggestion into the Numba FAQ.

seibert on Apr 30, 2019

This was a helpful thread for fixing with GCP functions, which just like AWS Lambda, has a read-only filesystem except for /tmp. The suggested fix of setting NUMBA_CACHE_DIR env variable to /tmp did the trick.

erikreed on Nov 24, 2022

@SumNeuron maybe you can set NUMBA_CACHE_DIR to $HOME or a subdirectory therof?

esc on Oct 22, 2019