>>> [!note] Migrated issue <!-- Drupal.org comment --> <!-- Migrated from issue #3586114. --> Reported by: [mably](https://www.drupal.org/user/3375160) Related to !61 >>> <h3 id="summary-problem-motivation">Problem/Motivation</h3> <p><code>Drupal\Tests\purge_users\Kernel\UserIdsTest::testUserIdsByUserValues</code> with data set <code>all_purge_conditions.notify</code> is flaky on 4.x: on some runs the query returns extra users whose <code>login</code> (or <code>created</code>, for never-logged-in users) sits exactly on the 1-year boundary, e.g. <code>created1y_loginNever</code>, <code>created2y_login1y</code>, <code>created3y_login1y</code>.</p> <p>The failing dataset configures all four purge conditions at 30 months and a pre-notification lead of 18 months, so the notify threshold is <code>30 &minus; 18 = 12 months</code>. Test users are seeded with <code>strtotime('-1 year')</code>, which sits exactly on that 12-month threshold. Small wall-clock drift between the seed and the subsequent query flips the strict <code>&lt;</code> comparison and leaks those users into the notification set.</p> <p>Example failed job (before retry): <a href="https://git.drupalcode.org/project/purge_users/-/jobs/9457996">job #9457996</a> on MR !60 pipeline 800958. Retried jobs <a href="https://git.drupalcode.org/project/purge_users/-/jobs/9458699">#9458699</a> and <a href="https://git.drupalcode.org/project/purge_users/-/jobs/9458700">#9458700</a> passed without code changes, which confirms the flake is timing-dependent rather than functional.</p> <p>5.x's equivalent dataset already uses <code>user_before_notification_value = 12</code> (introduced as part of the policy refactor in <a href="https://www.drupal.org/project/purge_users/issues/3427680">#3427680</a>), which moves the threshold to <code>30 &minus; 12 = 18 months</code> &mdash; safely between the seeded 1-year and 2-year marks. The 4.x <code>mixed</code> dataset was already adjusted in <a href="https://www.drupal.org/project/purge_users/issues/3411228">#3411228</a> (commit <code>7525178</code>), but <code>all_purge_conditions</code> was missed.</p> <h4 id="summary-steps-reproduce">Steps to reproduce</h4> <p>Run <code>UserIdsTest::testUserIdsByUserValues</code> repeatedly on 4.x. The failure rate depends on wall-clock drift between seeding and querying; on GitLab CI it reproduces often enough to block pipelines on unrelated MRs.</p> <h3 id="summary-proposed-resolution">Proposed resolution</h3> <p>In <code>tests/src/Kernel/UserIdsTest.php</code>, change <code>user_before_notification_value</code> from <code>18</code> to <code>12</code> in the <code>all_purge_conditions</code> dataset. This shifts the notify threshold to 18 months &mdash; midway between the seeded 1-year and 2-year marks, so the <code>&lt;</code> comparison is no longer at a boundary.</p> <p>The expected notify list remains unchanged: all six users currently listed (<code>created3y_login2y*</code>, <code>created2y_loginNever*</code>, <code>created2y_login2y*</code>) have their time component at 2 years, well above the new 18-month threshold, so the array_diff against the purge candidates yields the same result.</p> <h3 id="summary-remaining-tasks">Remaining tasks</h3> <ul> <li>Apply the one-line change in the dataset.</li> <li>Verify the pipeline stays green across current, previous major, and next minor matrices.</li> </ul> <h3 id="summary-release-notes">Release notes snippet</h3> <p>Fixes a flaky kernel test on the 4.x branch caused by the pre-notification threshold sitting exactly on the 1-year boundary used by seeded test fixtures.</p>