## Summary The following query runs a few hundred times per second distributed across nodes, and lacks a `partition_id` filter, causing full cross-partition scans on `p_ci_pipelines`: ```sql SELECT "p_ci_pipelines".* FROM "p_ci_pipelines" WHERE "p_ci_pipelines"."id" = $1 LIMIT $2 ``` - **Dashboard**: https://dashboards.gitlab.net/goto/efgz12snw1fcwf?orgId=1 - **Fingerprint**: `c29690408ab28010` - **Query ID**: `-6756117007090651256` A likely point of entry is `merge_request.head_pipeline`, which looks up a pipeline by ID without scoping to a specific partition. ## Proposed Solution Investigate where this query originates (likely through `merge_request.head_pipeline` or similar associations) and add `partition_id` pruning so the query is scoped to the correct partition rather than scanning all partitions. ## Impact - As the number of CI partitions grows, each unpruned scan becomes progressively more expensive - Running at a few hundred RPS across nodes means this has significant cumulative cost on both primary and replica nodes ## Parent Part of [#594500 - Immediate partition pruning for high-impact queries](https://gitlab.com/gitlab-org/gitlab/-/work_items/594500)