#110487Flaky e2e test: inline_1n sorts an inline child down
Description
Problem
The playwright end to end test “inline_1n sorts an inline child down” in
Build/tests/playwright/e2e/form-engine/inline-1n.spec.ts fails sporadically.
A retry of the CI job passes, so the pipeline turns green in the end and the
problem stays mostly invisible.
Example: job 5790947 of pipeline 106713 on branch 14.3. That job needed 636
seconds and reported “1 failed, 1 flaky, 90 passed”.
The failure is always the same, three times in a row:
bc. Test timeout of 30000ms exceeded.
Error: locator.click: Test timeout of 30000ms exceeded.
waiting for locator(…) button span[data-identifier=“actions-move-down”] first
locator resolved to span with data-identifier=“actions-move-down“
attempting click action
2 x waiting for element to be visible, enabled and stable
element is not visible
The element resolves but never becomes visible during the full 30 seconds, so
this is not a timing race, the state is stable and simply not the expected one.
Analysis
The retained job artifacts, page snapshots and playwright traces, show the
following.
The spec runs in serial mode and depends on shared state by design. A comment
in the file states that the sorting test depends on the inline child added by
the preceding “create” test.
The database is not reset between playwright attempts. At the failing click of
the second attempt the parent record carried three inline children instead of
two, with the field names ending on uid 1, uid 6 and uid 7. uid 1 is the
styleguide fixture child, uid 6 and uid 7 are the children added by the create
test during the first and the second attempt. A serial group is retried as a
whole, so every attempt appends another child and the documented precondition
of exactly one added child no longer holds.
Two selectors do not survive that. The create test picks the newly created
child with a “last” match over all child input fields, and the sorting test
picks the sort control with a “first” match over all child panels. Both mean
something different as soon as the number of children changes. In the failing
attempts the first match was a control which the form deliberately renders
invisible, which is why waiting could never succeed.
A retried CI job provisions a new instance with a new database, so its first
attempt starts from clean styleguide data. That is why the flakyness always
disappears on a job retry, and why it is not reproducible by simply running the
suite once.
The file already documents this class of problem for another selector, which
was anchored to uid 1 for exactly the same reason.
Solution
Address the child records explicitly instead of relying on positional matches,
and assert the preconditions the sorting step needs before clicking, so a
violated precondition reports as a failed expectation and not as a click which
runs into a timeout. A not yet persisted inline child can be addressed
reliably, since its field name carries a “NEW” identifier until the record is
saved.
Affects main and 14.3. Branch 13.4 does not contain this spec.
Reviews 2
History 13
-
Gerrit Code Review pushed patch set 1 for main to change 95335 status
NewUnder Review <span data-processor="relativetime" data-value="2026-08-14T20:10:43+00:00" > 08/14/26, 20:10 </span> -
Gerrit Code Review pushed patch set 1 for 14.3 to change 95336 <span data-processor="relativetime" data-value="2026-08-14T20:11:26+00:00" > 08/14/26, 20:11 </span>
-
Stefan Bürk commented and changed status
Under ReviewResolved done0 %100 % <span data-processor="relativetime" data-value="2026-08-14T20:40:12+00:00" > 08/14/26, 20:40 </span>
Applied in changeset commit:d93c6ff8d3c4730977b676d72f3385ac2a35a568.