Resolve the ClickHouse config symlink in PMM-T2237

[claude]

Failures fixed (investigator)

• source: Percona-Lab/pmm-submodules#4516 — run 31586770693, check E2E / Docker configuration tests / e2e tests: @docker-configuration (job 94467124725)
• tests:
  • e2e_tests/tests/dockerConfiguration/clickhouseConfig.test.ts:32 / @docker-configuration — PMM-T2237, config pmm-server-default-clickhouse-config
  • e2e_tests/tests/dockerConfiguration/clickhouseConfig.test.ts:32 / @docker-configuration — PMM-T2237, config pmm-server-low-memory-clickhouse-config
  • e2e_tests/tests/dockerConfiguration/clickhouseConfig.test.ts:32 / @docker-configuration — PMM-T2237, config pmm-server-no-flag-clickhouse-config

What failed

All three PMM-T2237 cases, deterministically, on all three retries each. The Playwright step itself
ends in || true, so the job went red one step later on launchable gate
(Actionable Failures | 3):

Error: Config name should be: default-config but actual value is: Processing configuration file '/etc/clickhouse-server/config.xml'.

expect(received).toContain(expected)
Expected substring: "default-config.xml"
Received string:    "Processing configuration file '/etc/clickhouse-server/config.xml'."

Root cause — the assertion, not the product

The test read the ClickHouse startup log and expected the line to name the config
PMM_CLICKHOUSE_CONFIG selected.

percona/pmm#5747 (PMM-15309, the PR this FB builds)
changes how that config is handed to ClickHouse. Server is now always started with the fixed
config.xml name:

-command = /usr/bin/clickhouse-server --config-file=/etc/clickhouse-server/{{ .ClickHouseConfig }}-config.xml
+command = /usr/bin/clickhouse-server --config-file=/etc/clickhouse-server/config.xml

and pmm-managed-init repoints config.xml / users.xml at the selected config
(clickhouse.LinkClickHouseConfig), so drop-ins in config.d / users.d keep working for every
config. The env var still does exactly what PMM-T2237 is there to check — but the startup log now
names config.xml for every config, so a comparison on that file name can no longer see it.

The fix

Take the path ClickHouse reports and resolve it before comparing. That holds for both shapes:

Build	logged path	readlink -f
main today, PMM_CLICKHOUSE_CONFIG=low-memory	/etc/clickhouse-server/low-memory-config.xml	/etc/clickhouse-server/low-memory-config.xml
FB PR-4516 (with #5747), same variable	/etc/clickhouse-server/config.xml	/etc/clickhouse-server/low-memory-config.xml

Note the middle column: on main the config.xml symlink exists but always points at
default-config.xml and is unused, so resolving that fixed name instead of the logged one would
be wrong there. Resolving the logged path is what covers both.

Not blocked on #5747 — it passes before and after that PR lands (verified below).

Verification

Reproduced and fixed against the exact FB image perconalab/pmm-server-fb:PR-4516-99b98f1,
digest sha256:7577435e… — the same digest Launchable recorded for the failing CI run — and
against perconalab/pmm-server:3-dev-latest for the pre-#5747 shape. Real Docker, Chromium
151.0.7922.34 (the build CI uses), same npx playwright test --grep "@docker-configuration"
invocation:

Test version	FB PR-4516 image (has #5747)	3-dev-latest (no #5747)
main	3 failed — Expected substring: "default-config.xml", identical to CI	passes (this is why CI only went red on this FB)
this branch	3 passed	3 passed

The check still discriminates, rather than passing on anything. On the FB image with
PMM_CLICKHOUSE_CONFIG=low-memory, repointing the symlink back at the default config — the shape
a "PMM ignores the variable" regression would produce — resolves to default-config.xml, which
fails the assertion:

as shipped:                         /etc/clickhouse-server/low-memory-config.xml
env var ignored (link at default):  /etc/clickhouse-server/default-config.xml

npx tsc --noEmit, ESLint and Prettier are clean.

Not covered here, and only the next real FB/nightly run confirms: the rest of the
@docker-configuration suite under Launchable subsetting.

Other failure in that run (not fixed here)

E2E / Docker configuration tests / e2e tests: @docker-configuration appears twice in that run —
the check-run listing collapses both jobs under one name. The other one
(94467124622,
CodeceptJS) failed on PMM-T2020 - Verify external clickhouse as datasource on explore page, the
external-ClickHouse side of the same PMM-15309 change. That is already covered by open PR #1164,
so it is deliberately left alone here.

---

Generated by Claude Code