Pin dandischema to require the latest schema version

[kabilar]

Fixes #1463

Edit:
We encountered this issue again today with dandi==0.66.4 so I suggest that we pin to the latest dandischema version.

Error message:
$ dandi upload -i linc <path_to_ome_zarr>
Error: Server requires schema version 0.6.8; client only supports 0.6.9. You may need to upgrade dandi and/or dandischema

cc @aaronkanzer @dstansby

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.44%. Comparing base (05f5691) to head (360455f).
Report is 5 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1570      +/-   ##
==========================================
+ Coverage   88.34%   88.44%   +0.10%     
==========================================
  Files          78       78              
  Lines       10735    10735              
==========================================
+ Hits         9484     9495      +11     
+ Misses       1251     1240      -11     

Flag	Coverage Δ
unittests	88.44% <ø> (+0.10%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[jwodder]

@kabilar This seems to be solving the wrong problem. If the client's schema version is newer than the server's — as the given error message states — then pinning to the latest schema version will just make it so users can't downgrade their dandischema to an older version to solve the problem.

[kabilar]

@kabilar This seems to be solving the wrong problem. If the client's schema version is newer than the server's — as the given error message states — then pinning to the latest schema version will just make it so users can't downgrade their dandischema to an older version to solve the problem.

Thanks @jwodder. Good catch. I misread the error message.

[kabilar]

We will still upgrade the linc-archive dandischema version.

@jwodder Would it be worth it to go ahead with this pull request to pin the dandischema version as the dandi-archive does to address #1463?

From dandi-archive setup.py line 45:

'dandischema==0.11.0',  # schema version 0.6.9

[jwodder]

@kabilar I don't like the idea of pinning dandischema. What if a new dandischema version is released that doesn't change the schema version but still fixes some bug? If you really need dandischema to only be for the latest schema version, make the requirement be a range with the lower end being the point when that schema version was added.

[kabilar]

@kabilar I don't like the idea of pinning dandischema. What if a new dandischema version is released that doesn't change the schema version but still fixes some bug?

Thats fair.

If you really need dandischema to only be for the latest schema version, make the requirement be a range with the lower end being the point when that schema version was added.

So in this case would dandischema>=0.11.0 work?

[jwodder]

@kabilar I think that should be dandischema>=0.11.0,<0.12.0.

[kabilar]

@kabilar I think that should be dandischema>=0.11.0,<0.12.0.

Sounds good. I have updated it.

[yarikoptic]

tests are failing, fomr a sample run

FAILED dandi/tests/test_dandiarchive.py::test_parse_api_url[https://gui.dandiarchive.org/#/dandiset/000001/0.201104.2302-parsed_url2] - requests.exceptions.ConnectionError: ('Connection aborted.', ConnectionResetError(104, 'Connection reset by peer'))
FAILED dandi/tests/test_dandiarchive.py::test_follow_redirect - requests.exceptions.ConnectionError: ('Connection aborted.', ConnectionResetError(104, 'Connection reset by peer'))

looks like unrelated but that should be figured out (e.g. separate PR showing the same fails), and ideally fixed... FWIW yesterday's run seems had failed only in a single run https://github.com/dandi/dandi-cli/actions/runs/13025543642 , so more analysis should be done... assuming it was connection issue, I would restart failing tests now, so see if that would be gone

[kabilar]

Not sure what changed, but it looks like all tests are passing now.

[yarikoptic]

@kabilar probably they just flaked out due to connection issues or other external factors (redirects etc... as we were observing similar ones in upptime checks)

[github-actions]

🚀 PR was released in 0.66.5 🚀