Section highlighting datalad-fuse and dandisets?

[yarikoptic]

@asmacdo showed interest to participate in ongoing handbook hackathon and I thought that it might be great to have a use case show case for dandisets (super dataset at https://github.com/dandi/dandisets, individual at https://github.com/dandisets, asyncio code to update those from the archive within the tools/ of dandisets) and https://github.com/datalad/datalad-fuse/ extension. Dandisets are "special" in that typical files are there large but for access to metadata etc, only small portion of data is needed to be accessed. In datalad-fuse we use https://github.com/fsspec/filesystem_spec/ with local caching, to provide efficient sparse access to remote annexed files which have an http* url associated with them.

In datalad core we had a request for streaming datalad/datalad#4003 -- so it might be useful to highlight how streaming could be implemented, via fsspec interface within datalad-fuse or directly via FUSE filesystem of that one.

WDYT datalad-handbook folks about such a section? (attn @adswa @mih)

[adswa]

Sure, sounds like a cool usecase to me. There is a rough structure that usecases are usually following: http://handbook.datalad.org/en/latest/contributing.html#use-cases

[yarikoptic]

more specific target for the use case could be

• Running dandi validate to validate one of the dandisets, show how much is downloaded (du -scm .git/datalad/cache or whatever that path is) from the total size of files which are validated.
• Reference https://pynwb.readthedocs.io/en/stable/tutorials/advanced_io/streaming.html#streaming-method-2-fsspec in the introduction as for how to achieve access to those files directly within python using fsspec. But then present datalad-fuse target as "we can run any external command, not just python code, which accesses those files. As we are doing in https://github.com/dandi/dandisets-healthstatus for MATLAB matnwb library"

[adswa]

Just ping me if you need any infos. You should add a new file in docs/usecases and place it somewhere in the docs/usecases/intro.rst toctree. Usecases do not need to have code that is executed and captured, so you can go with .. code-block::s instead of .. runrecords::. Looking forward to it!