[yarikoptic-gitmate]

This removes two candidates from the "removable now?" table in #1963.

Changes

• get_instance() in dandi/utils.py: drop the check meant to say that the dandi-staging instance identifier was removed in favor of dandi-sandbox (made an error in 0.74.0, Add support for all known instances in short URL format, removed dandi-staging and renamed EMBER* ones into EMBER-DANDI* #1750).
  • It never ran for an identifier given as a string, because known_instances[dandi_id] raises KeyError: 'dandi-staging' first. That KeyError is what users already get, so nothing changes for them.
  • It could only run for a DandiInstance object named dandi-staging.
  • The rename (Rename dandiarchive staging => sandbox #1668) happened in August 2025.
• _check_attempts_and_sleep() in dandi/download.py: stop retrying downloads on HTTP 400.
  • The retry was added in 2020 because Girder returned 400 (download: we need to retry #87), and Girder support was removed in 2021.
  • Each retry requests the same archive download URL again, so a 400 would not turn into a success.
  • Downloads still retry on RETRY_STATUSES (429, 500, 502, 503, 504) and on errors without a response.
  • A new test checks that 400, 403 and 404 are not retried.

The registry PR, #1970, also drops the allowed_schema_versions TODO about dandi/dandi-archive#2624, since it rewrites that block; it is left out here to avoid a conflict.

Testing

• test_download.py -k check_attempts, test_utils.py and cli/tests/test_instances.py pass locally. The exception is test_server_info, which needs dandiarchive.org, and the sandbox this was prepared in cannot reach it.
• pre-commit and mypy are clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_014sJtxzEPbSxycxetp4DquU

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.90%. Comparing base (e65abd3) to head (d63290c).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1969      +/-   ##
==========================================
+ Coverage   90.82%   90.90%   +0.08%     
==========================================
  Files          92       92              
  Lines       14190    14196       +6     
==========================================
+ Hits        12888    12905      +17     
+ Misses       1302     1291      -11     

Flag	Coverage Δ
unittests	90.90% <100.00%> (+0.08%)	⬆️

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

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[yarikoptic-gitmate]

test (ubuntu-latest, 3.11, nfs) failed in the step "Run all tests except those involving obolibrary" (job). Every other job in the run passed. That includes test (ubuntu-latest, 3.11, normal, default), which runs the same tests on the same OS and Python without NFS.

I couldn't read which test failed. The log tools available to me return only the last 5000 lines of the job, which are the Docker Compose dump from the next step.

I don't think this PR caused it:

• The change is the same on any filesystem. The PR stops retrying downloads on HTTP 400 and drops a get_instance() branch that can never run. If that broke a test, the test should also fail without NFS.
• The server log shows no failed downloads. In the part of the log that is visible (the last ~3.5 min of the test step), every /download/ request got a 302. The only two 400s were on upload endpoints (/api/uploads/initialize/, /api/zarr/), and those don't go through _check_attempts_and_sleep().
• NFS mode has known flaky tests. test_dandi_file_zarr_with_excluded_dotfiles is marked flaky(reruns=10), and test_reextract_metadata is xfailed under NFS (test_reextract_metadata some times times out on NFS #1507).
• master is green on the same base (e65abd3).

No fix exists to port, since the failing test isn't known. I'm re-running the failed job once. If it fails again, I'll treat it as real.

---

Generated by Claude Code