The daily CI check picked this flicker for an automated stabilisation attempt today and dropped it without a PR. What it found:
The test waits for the one auto-saved version, then asserts that no second version appears within 10 s. That assertion is the requirement, and apart from it the test has only its two timeouts. Changing the expectation or a timeout would hide the bug instead of fixing it, so there is no fix on the test side.
The second version is therefore a real, redundant save: the save election in Saver (getSavingClientId / afterSave rescheduling) together with XWikiFormSaveTarget lets a second client save again after the first save has landed. The fix belongs there, in production code, and needs a human.
Because no fix was possible, the repeat-run measurement was not run.
Updated Develocity numbers (28 days): 8 failures in 122 executions (6.6%), still only with database = mysql latest (p=2e-07), Chrome only:
