swmr capability of hdf5

Hi,
I am suffering from a shortcoming of h5py its swmr (single write multiple read) capability as described in an issue on github. The issue is labelled with ‘bug-in-external-lib’ which suggests that the shortcoming is located at the underlying hdf5 library rather than in h5py itself. I wonder if there is any interest from the hdf5 maintainers to change that. In the bitbucket repository is a branch called ‘full_swmr’. Does anybody know how mature this branch is? Is there a chance to see full swmr in hdf5 in the nearer future, especially regarding the github issue I mentioned above?

Best,
Alex

PS: I posted this message also on stackoverflow but later found this place to be more relevant.

Hi Alexander,

Unfortunately, this design issue has not been addressed (I saw my responses on h5py FORUM from 2 years ago :smirk:)

To answer your question about our interest to address it - yes, we definitely want to address it.

Quincey (@koziol) will correct me, but I don’t think the full_swmr branch solved it either. This branch allows to modify the file structure (add groups, datasets, and attributes) vs. current SWMR that only adds raw data. The branch is stable. It would be great if h5py community tries full_swmr branch and see how it works.

The HDF Group has been working on an alternative full SWMR implementation, but it is in the design/early prototype stage right now. It will not have this limitation.

Thank you!

Elena

Hello Elena,

thank you very much for your response. I am happy to hear that the HDF Group is working on the issue and I understand that the implementation demands a lot work under the hood. Whenever there is a public branch with preliminary code to play around with let us know :slight_smile:.

Thank you as well for clarifying what the ‘full_swmr’ branch is implementing. The information saved me from trying to resolve my issue using it.

Best,
Alex