Reliability-rate flaky-test alerts were driving the production ClickHouse instance past its memory ceiling, which in turn killed unrelated analytics, build-count, and percentile queries with MEMORY_LIMIT_EXCEEDED and intermittently broke dashboards. Reliability rolling windows up to 750 now read from pre-aggregated bucket materialized views instead of doing a full 1000-entry sort over every active test case, and every ClickHouse read is now capped by a per-query memory limit (default 6 GiB). Result: reliability-rate alert evaluations drop from multi-GiB to sub-GiB, and dashboards/analytics stop failing intermittently.
Hive
Stop reliability-rate alerts and dashboards from triggering ClickHouse out-of-memory failures
Published
Jul 03, 2026 · 07:37 UTC
Repository
tuist/tuist