Skip to content

GKGExecutionFailureRate

ClickHouse query execution is failing, above 1 failure per minute. Severity s3, fires after 5m.

Compiled queries are failing at execution against the graph ClickHouse. ClickHouse dominates query performance and shares its capacity between webserver reads and indexer writes, so the usual cause is load: a heavy indexing burst, a very heavy query, or the ClickHouse Cloud service itself being degraded. Timeouts (Code: 159 TIMEOUT_EXCEEDED, the 30s execution cap) land here too.

The firing expression, on gkg_query_pipeline_failed_total (emitted by the service, see the metrics catalog):

sum by (env, environment, stage) (rate(gkg_query_pipeline_failed_total{failure_reason="execution"}[5m])) * 60 > 1
  • Silence via Alertmanager filtered on the alertname.
  • The first queries after a quiet period can fail or crawl while ClickHouse Cloud wakes from idle; that self-resolves in minutes.
  • s3, no paging. Users see failed queries; indexing is unaffected unless ClickHouse itself is down.
sum by (env, environment, stage) (rate(gkg_query_pipeline_failed_total{failure_reason="execution"}[5m])) * 60

The webserver logs carry the ClickHouse error codes. Check the graph ClickHouse service health and current load in the ClickHouse Cloud console (Service Read Only role suffices), and whether indexer write volume rose at the same time.

SymptomCauseAction
Failures rise with indexer write volumeRead/write contention on the shared graph ClickHousePause the indexer pools (Pause and resume indexing)
Same query fails repeatedly with TIMEOUT_EXCEEDEDOne pathological query against the 30s execution capIdentify it in the logs; the fix is query optimization, not scaling
Broad failures, ClickHouse console shows degradationClickHouse Cloud incidentOpen a support case with ClickHouse Cloud
  • Indexing burst degrading reads: pause the indexer pools (indexer.enabled: false in the ArgoCD values); queries keep working while data goes stale.
  • One pathological query timing out repeatedly: identify it in the logs; the fix is application-side query optimization, not scaling.
  • ClickHouse Cloud degraded or waking from idle: idle_scaling makes the first queries after a quiet period slow; sustained failure is a support case with ClickHouse Cloud.
  • Graph ClickHouse (ClickHouse Cloud) dominates query performance and is shared with indexer writes.

Ask in #f_orbit_dev (context_systems). For a ClickHouse Cloud platform problem, escalate through the ClickHouse Cloud support channel.

  • Alert rule (hand-maintained; keep the expr above in sync when tuning it).