I have a detailed question about a problem I am encountering using the
gzip filter in HDF5.
The problem manifests as a failure to allocate the output buffer in
H5Zdeflate (line 135) reporting unavailable resources. I have debugged
this quite a bit and have confirmed that the buffer trying to be
allocated is only a few MB and there is roughly 60GB of memory still
available on the machine at the time the allocation fails. I strongly
suspect it is due to heap fragmentation (I don't have a 1MB contiguous
block of memory available).
This is only happening on a particular machine, and I understand there's
a lot of important subtlety to chunk size and chunk cache size so I'm
wondering if anyone has encountered some similar behavior in that heap
fragmentation is likely to occur in H5Zdeflate. Also wondering if anyone
can provide any guidance on how to resolve this. I was thinking the next
thing to try would be to play with options for the chunk cache size
(H5Pset_cache) described here