Mute PMM-T506 behind an expiring skip

[travagliad]

Failures fixed (investigator)

• source: nightly CI run https://github.com/percona/pmm-qa/actions/runs/33222594797 (Nightly E2E tests Matrix (remote PMM Server), job test execution / @nightly) — also red Aug 30 and Aug 31 on the same test
• tests:
  • codeceptjs-e2e/tests/verifyVMDashboards_test.js:10 / @nightly — PMM-T506 Verify metrics on VictoriaMetrics dashboard

What this is

A noise mute that expires by construction — an alternative to #1275, which is not being merged as it stands.

PMM-T506 expects a Time Series Information row with Top 10 metrics by time series count and
Top 10 hosts by time series count. percona/pmm#5822
removed both deliberately — in the author's words, "they were literally taking PMM Server down
under high load, i.e. 300+ clients." So the test fails every night and will keep failing until the
expectation drops (#1275) or the panels come back.

Two parts:

1. The mute (verifyVMDashboards_test.js)

+// skip-until: 2026-09-29 -- cardinality panels removed by percona/pmm#5822; unskip via pmm-qa#1275, or when the panels return.
-Scenario(
+Scenario.skip(

2. A guard that makes the date real (lint-changed.sh, lint.yml)

A skip parked with skip-until: YYYY-MM-DD becomes a lint failure on that date.

Why the guard, and not just a TODO

This repo carries 69 skipped scenarios across 32 files, and the convention already in use is
exactly a TODO with a ticket link — // TODO: unskip after …PMM-13544, …PMM-12956 (twice),
PMM-14002, PMM-13750. Two of those references are already dead: one points at
jira.percona.com, the Jira instance the project migrated off, and one at a grafana-dashboards
PR for a repo since folded into the monorepo. Dead links guarding dormant tests.

So a comment and a ticket is the mechanism with a demonstrated 69× failure rate here. A date the
build enforces does not depend on anyone remembering.

Where the teeth are: lint.yml runs git ls-files | xargs lint-changed.sh, i.e. every
tracked file on every PR. Once 2026-09-29 arrives, every PR in the repo fails lint until the
skip is revisited or the date is deliberately moved. A weekly schedule was added so it also
surfaces during a quiet week with no PR traffic.

The check lives in lint-changed.sh rather than as a workflow step because that script already
owns the linter mapping for both CI and the local PreToolUse commit gate — its own header warns
that re-listing linters in the workflow is how the two drift apart.

Opt-in by design: only skips that carry the marker are checked, so the existing 69 are
untouched and CI stays green today. Adopting it for them can be a deliberate, separate pass.

Relationship to #1275 — please don't merge both

They are alternatives and touch different files, so nothing stops both landing and leaving the test
skipped and the expectations dropped — the skip would then mask a test already made correct.

• Merge this → nightly quiet, no VictoriaMetrics dashboard coverage, forced review on 2026-09-29.
• Merge Drop the cardinality panel expectations and unskip PMM-T506 #1275 → nightly quiet, coverage kept. Then close this PR, or revert it if it landed first.

What the mute costs — worth being explicit

Scenario.skip disables the whole scenario, not just the two stale expectations: the ~40 other
panel titles it checks, and verifyThereAreNoGraphsWithoutData(1). Until it is unskipped, nothing
catches a genuinely broken or empty VictoriaMetrics dashboard. That is the reason for the expiry
rather than an open-ended skip.

I checked whether the test could leave the nightly but stay alive elsewhere — it cannot. Its other
two tags are dead: no workflow references @dashboards or @gssapi-nightly, and @nightly
appears in exactly one place (nightly-e2e-tests-matrix.yml:140). Skipping and untagging cost
identical coverage, and Scenario.skip at least shows as S in the run output rather than leaving
a test nothing silently runs.

Given that cost, #1275 remains the better outcome if its objection can be resolved.

Verification

Skip behaviour:

$ npx codeceptjs run -c pr.codecept.js --grep 'PMM-T506'
  S PMM-T506 - Verify metrics on VictoriaMetrics dashboard @nightly  @dashboards @gssapi-nightly
  OK  | 0 passed, 1 skipped   // 12ms

No server contact, and PMM-T507 in the same file is untouched and still runs.

Guard, both directions:

case	result
date in the future (2026-09-29, today 2026-09-01)	==> skip-until expiry, rc=0
date in the past (2026-08-15)	FAIL: skip-until date reached … :9:skip-until: 2026-08-15, rc=1
date is today (2026-09-01) — boundary	rc=1 (the date means review on it, not after)
all 393 tracked .js/.ts files	exactly one marker — the 69 existing skips do not trip it

Repo linters on the changed files (yamllint, actionlint, shellcheck, plus the new check) — all
clean, rc=0.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01SKRuqWTQVW4p6DDSYcFQEf

[claude]

🟡 Scope: this and #1275 are mutually exclusive but both are open, non-draft and unlabelled, so "please don't merge both" is enforced only by whoever reads the body — mark this one draft with Depends on pointing at the #1275 outcome, so it cannot land by accident.

Gate gap: lint.yml dispatches through .claude/hooks/lint-changed.sh, which has no .js group, so no linter reads codeceptjs-e2e/**; and no check on this PR runs @nightly, the only tag a workflow references. The green checks here say nothing about the change.