Problem

The webui job in both BDB and LMDB test workflows is unreliable and can consume the full 90-minute job window. Follow-up investigation shows this is more than a timeout: both BDB and LMDB show a suite-wide browser/setup failure cascade before cancellation.

Evidence:

• BDB webui cancellation in Test workflow:
  https://github.com/389ds/389-ds-base/actions/runs/26794794932
• LMDB webui cancellation in LMDB Test workflow:
  https://github.com/389ds/389-ds-base/actions/runs/26797037399

Observed behavior:

• Both BDB and LMDB collect about 120 webui tests.
• BDB and LMDB fail broadly across Firefox and Chromium before hitting job cancellation around 90 minutes.
• Cleanup then reports missing or incomplete pytest artifacts in some canceled runs, making diagnosis harder.

Representative failing areas include:

• webui/backup/backup_test.py::test_no_backup_dir
• webui/create/create_instance_test.py::*
• webui/database/database_test.py::*
• webui/ldap_browser/ldap_browser_test.py::*
• webui/plugins/plugins_test.py::*
• webui/replication/replication_test.py::*
• webui/server/server_test.py::*

Additional evidence from follow-up investigation

• Latest sampled BDB webui job 78989711065 collected 120 tests, started failing at webui/backup/backup_test.py::test_no_backup_dir[firefox], then failed through create/database/ldap_browser/plugins/replication/server tests before The operation was canceled.
• Latest sampled LMDB webui job 78996570665 shows the same pattern: collected 120 tests, broad Firefox failures, then Chromium failures, then cancellation.
• This points to a suite-wide Cockpit/browser/session/setup issue or shared state failure, not a single slow webui test.

Proposed work

Investigate whether the failures are caused by Cockpit startup/auth/session setup, browser fixture setup, test ordering/state leakage, backend instance availability, or job timeout sizing.

Consider splitting the webui matrix by browser or subdirectory if the suite is inherently too long for one job.

Acceptance criteria

• Root cause is identified for the first failing webui tests instead of only seeing job cancellation.
• The suite produces pytest XML/HTML artifacts even on failure or cancellation where possible.
• webui CI either completes within the job timeout or is split/timeout-tuned to report actionable failures.