25.3 Stable Backport of #84317 - Report cgroup and system-wide metrics along with each other

[mkmkme]

Report cgroup and system-wide metrics along with each other

Changelog category (leave one):

• Improvement

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Cgroup-level and system-wide metrics are reported now altogether. Cgroup-level metrics have names CGroup<Metric> and OS-level metrics (collected from procfs) have names OS<Metric> (ClickHouse#84317 by @nickitat).

Documentation entry for user-facing changes

...

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Unit tests
• Performance tests
• Aarch64 tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• All Regression
• Disable CI Cache

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• CAS (content-addressed storage; Antalya only)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• OAuth (5m)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

[mkmkme]

@blau-ai

[blau-ai]

CI triage

Verdict: 0 PR-caused failures. All red checks are pre-existing flaky integration tests, an unrelated regression-suite assertion, a DCO/sign-off check, and the aggregate gate that just reflects them. Notably, this PR's own modified test — test_async_metrics_in_cgroup — passes in CI, which is the direct signal that the backported AsynchronousMetrics change behaves as intended.

The diff only touches src/Common/AsynchronousMetrics.{cpp,h} (new CGroup*Time[Normalized] async metrics) and tests/integration/test_async_metrics_in_cgroup/test.py. None of the failing tests exercise that code path.

Per-failure breakdown

1. Integration tests (asan) [5/8] and (release) [5/8] (= job runners IntegrationTestsAsan-4, IntegrationTestsRelease-4, IntegrationTestsAnalyzerAsan-4)

• Single failure in each: test_race_condition_for_replicated_merge_tree/test.py::test_partition_move_drop_race (fail: 1, passed: 427/429).
• Classification: flaky, not PR-related. It's a deliberate ReplicatedMergeTree partition move/drop race test with zero connection to async metrics. [5/8] is the 1-indexed report for the 0-indexed -4 shard, which is why those three job rows all point at the same test.
• Next step: safe to re-run.

2. IntegrationTestsAnalyzerAsan-0 (old analyzer, shard [1/8])

• Failures (from the job log): test_async_insert_adaptive_busy_timeout::test_change_queries_frequency, test_cluster_discovery::test_cluster_discovery_startup_and_stop, test_parallel_replicas_custom_key_load_balancing[...], test_parallel_replicas_over_distributed[...].
• Classification: flaky/infra, not PR-related. All timing- and replica-coordination-sensitive tests that are known to be unstable on loaded CI runners; none touch metrics. (The regular-analyzer asan [1/8] shard passed, consistent with flakiness.)
• Next step: re-run.

3. RegressionTestsRelease / Common (version) / version (a.k.a. Regression release version)

• Failing scenario: no_clickhouse_cloud_url_in_auth_error in version/tests/altinity_version.py. It asserts the wrong-password auth error does not contain https://clickhouse.cloud, but this build's AUTHENTICATION_FAILED message still includes:

  If you use ClickHouse Cloud, the password can be reset at https://clickhouse.cloud/ ...

• Classification: not PR-related (pre-existing build/branding gap). This is about the authentication error text, an entirely different subsystem from AsynchronousMetrics. The regression suite (release branch) checks for an Altinity patch that strips the Cloud URL from the auth message; that patch isn't present in this 25.3.14 build. It will fail identically on releases/25.3.14 regardless of this PR.
• Next step: belongs to the release/branding backlog, not this backport — compare against a base-branch (releases/25.3.14) regression run to confirm it's independent.

4. DCO

• Classification: metadata/process check, not code. The commits (the upstream merge of Report cgroup and system-wide metrics along with each other ClickHouse/ClickHouse#84317 and the Merge branch 'releases/25.3.14' commit) carry no Signed-off-by trailer. Common on backport branches.
• Next step: add sign-off, or have a maintainer apply the usual DCO waiver — no code change required.

5. FinishCheck

• Aggregate gate; red only because the checks above are red. No independent failure. Goes green once the flaky suites are re-run (and DCO/regression are waived).

Bottom line

Nothing here indicates a problem introduced by the backport. Re-running the two integration suites should clear items 1–2; items 3–4 are pre-existing/process issues to be waived or handled outside this PR. No code changes proposed.

_{— @blau-ai (automated CI triage). Evidence: praktika S3 reports for 188b1608 + job logs from run 34531380972.}