You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
RHEL (and clones, such as Rocky) can be put in FIPS mode, which restricts non-FIPS-140 compliant cryptographic settings. This, it turns out, includes md5 hashes, and it's apparently disabled some very low level, so calls to joblib.hash() in umap_.py lead to errors like
File "/home/cluster/bernstei/.local/lib/python3.9/site-packages/umap/umap_.py", line 2822, in fit
self._input_hash = joblib.hash(self._raw_data)
File "/home/Software/python/system/extra/lib/python3.9/site-packages/joblib/hashing.py", line 263, in hash
hasher = NumpyHasher(hash_name=hash_name, coerce_mmap=coerce_mmap)
File "/home/Software/python/system/extra/lib/python3.9/site-packages/joblib/hashing.py", line 169, in __init__
Hasher.__init__(self, hash_name=hash_name)
File "/home/Software/python/system/extra/lib/python3.9/site-packages/joblib/hashing.py", line 59, in __init__
self._hash = hashlib.new(hash_name)
File "/usr/lib64/python3.9/hashlib.py", line 170, in __hash_new
return __get_builtin_constructor(name)(data)
File "/usr/lib64/python3.9/hashlib.py", line 86, in __get_builtin_constructor
raise ValueError('unsupported hash type ' + name + '(in FIPS mode)')
ValueError: unsupported hash type md5(in FIPS mode)
Using joblib.hash(..., hash_name="sha1") appears to be sufficient to fix it, but I really don't know what you'd consider a good mechanism for controlling this at the umap level, or even whether you think it's an important enough use case to be worth addressing.
The text was updated successfully, but these errors were encountered:
RHEL (and clones, such as Rocky) can be put in FIPS mode, which restricts non-FIPS-140 compliant cryptographic settings. This, it turns out, includes md5 hashes, and it's apparently disabled some very low level, so calls to
joblib.hash()
inumap_.py
lead to errors likeUsing
joblib.hash(..., hash_name="sha1")
appears to be sufficient to fix it, but I really don't know what you'd consider a good mechanism for controlling this at theumap
level, or even whether you think it's an important enough use case to be worth addressing.The text was updated successfully, but these errors were encountered: