-
[OPIK-6905] [BE] feat: cluster + cold_s3 disk ClickHouse health checks (#7419)
frostbyte_neo 发布于
2026-07-13 07:29:39 +00:00 - [OPIK-6905] [BE] feat: cluster + cold_s3 disk ClickHouse health checks
Add two toggle-gated ClickHouse existence health checks for the Hyperscale
topology:- clickhouse-cluster: fails readiness when the Distributed cluster definition
('cluster') isn't visible from the node (system.clusters). - clickhouse-cold-storage-disk: fails readiness when the 'cold_s3' tiered
storage disk isn't reachable (system.disks).
Both are gated by new databaseAnalytics flags (clusterHealthCheckEnabled /
coldStorageDiskHealthCheckEnabled), default off, so single-shard / OSS Docker
deployments are unaffected. Because a count() query succeeds returning 0 when
the resource is absent, the shared base reads the count and reports unhealthy
when it is zero, reusing the existing timeout / max_execution_time /
cancel-on-failure machinery.- refactor(healthcheck): share probe execution + add log_comment
Address PR review: extract the timeout/cancel/interrupt flow into a shared
AbstractClickHouseHealthCheck.executeProbe(future, resultMapper) so the
existence probes supply only the query and result mapping instead of
duplicating the flow. Add a log_comment (health_check:) to the shared
newQuerySettings() so every probe query is traceable in system.query_log
(per .agents/skills/opik-backend/clickhouse.md); the read-only probe still
overrides newQuerySettings() to null and is unaffected. Strengthen the unit
matcher to assert log_comment on every enabled probe path.下载附件