Fix QAN search page-object race with debounced search

[claude]

Failures fixed (investigator)

• source: Nightly E2E tests Matrix run 31916469379 (main @ 37a003d, server perconalab/pmm-server:3-dev-latest), job test execution / @qan|@menu|@valkey-nightly|@permissions-nightly|@pt-summary-nightly|@pbm-nightly. Same two tests also failed in the previous night's run (31852756475) — not a one-off.
• tests:
  • codeceptjs-e2e/tests/QAN/details_explain_test.js / @qan — PMM-T1790 "Verify that there is any no error on Explains after switching between queries from different DB servers"
  • codeceptjs-e2e/tests/QAN/overview_test.js / @qan — PMM-T1061 "Verify Plan and PlanID with pg_stat_monitor"

What failed

Both tests died on the same page-object step, with the same error:

Clickable element "//input[contains(@name, "search")]" was not found by text|CSS|XPath
  ✖ I.click("//input[contains(@name, "search")]")  at QueryAnalyticsData.searchByValue (queryAnalyticsData.js:158:7)
  ✔ I.clearField("//input[contains(@name, "search")]")                              (:157)
  ✔ I.wait(1)                                                                        (:156)
  ✔ I.waitForVisible("//input[contains(@name, "search")]", 30)                        (:155)

The input is visible at :155 and gone by :158.

Root cause — a race in our page object, not the product

QAN Overview search is search-as-you-type (debounced) since PMM-14848 / percona/pmm#5537.

searchByValue(value) {
  I.waitForVisible(this.elements.queryRow(0), 30);
  I.waitForVisible(this.fields.searchBy, 30);
  I.wait(1);
  I.clearField(this.fields.searchBy);   // -> empty search -> Overview panel reloads
  I.click(this.fields.searchBy);        // <- panel is mid-reload; input is unmounted
  ...

I.clearField fires a change on the debounced field, which kicks off an empty-search reload of the Overview panel. The panel unmounts the search input while it reloads. I.click (proceedClick → assertElementExists) asserts existence without waiting for it, so if it lands inside that window it fails immediately.

Whether it lands there is purely a matter of how long the reload takes. Codecept's waitForAction: 500 sits between the two steps, so a reload shorter than ~500 ms is invisible and a longer one is fatal — which is why this is deterministic on the nightly (14 client shards feeding one server) and invisible on a small setup.

Evidence — throwaway Linode VM, PMM 3.9.1-v3-0c89674a1 (3-dev-latest), Chrome 151.0.7922.34 (same as CI)

Setups: --database ps --database psmdb --database pdpgsql (the services these two specs filter on: ps-dev, mongodb, pdpgsql_pmm).

1. The unmount is real. Instrumented the DOM across a clearField on an idle server — the input detaches for ~200 ms:

search inputs before clear: 1
search input EVER detached after clear: true (~200ms)

~200 ms is under waitForAction: 500, so both specs pass individually on an idle box — exactly why this doesn't show up in light environments.

2. With a CI-like reload duration it fails deterministically. Same page, same session, POST /v1/qan/metrics:getReport delayed 3 s, replaying both step sequences verbatim (clearField = fill(''), click = existence assert with no wait, waitForAction: 500 between steps):

OLD  clearField -> click     : found=0 -> FAIL: ElementNotFound (== CI)
NEW  click -> fillField      : found=1 -> pass (value="SELECT * FROM")

3. Unforced reproduction + no regressions. Full @qan tag run twice on the same VM, back to back, main vs this branch:

	passed	failed
main (37a003d)	45	18
this branch	47	16

Failing-test diff:

• fixed by this change: PMM-T1790 (@qan), PMM-T215 (@qan)
• regressions: none
• unchanged: 16, all from the 11 client setups this VM doesn't have (pgsql_pgss_pmm, mysql_pmm_, pxc_node__1, pxc-dev-cluster, pagination counts) — identical on both runs

The main run failed PMM-T1790 with the CI error verbatim, at queryAnalyticsData.js:158, with no throttling involved.

The fix

Don't clear before clicking. Click the already-loaded grid first and let I.fillField clear-and-type (Playwright fill clears the field) — one debounce cycle instead of two, and no click against a reloading panel. Settle with waitForLoaded() afterwards, and apply the same settle to clearSearchValue().

Notes

• This restores commit 0b2c040, written during the 3.9.0-rc dry-run and described in RC dry-run test fixes: QAN search race + RC CLI version resolution #1184 — that branch was later rewritten and the commit is no longer in it, so nothing on main or in any open PR carries the fix.
• PMM-T1061 passed in both full runs on this VM (it is the more load-sensitive of the two: its failing search comes after resetAllFilters()). It fails in CI with the identical error at the identical page-object line, and evidence 2 reproduces its exact failure deterministically. Only the next real nightly can confirm it end-to-end under the full 14-shard load.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01Dtd6vCp2nyWfx2T8VdfnJj

---

Generated by Claude Code