## Summary Partition `ci_job_artifact_states` by `partition_id`, the same way `p_ci_job_artifacts` is partitioned. The Transport group owns this epic unless Geo takes it over. Nobody is working on it yet. ## Why `ci_job_artifact_states` holds one Geo verification row per job artifact, so on any instance that runs Geo it grows with `p_ci_job_artifacts`. - **Scalability.** Large self-managed and GitLab Dedicated Geo sites add rows here at the same rate as job artifacts. On GitLab.com the table reached 50 GiB before it was [truncated on 2025-04-07](https://gitlab.com/gitlab-org/gitlab/-/issues/525414#note_2451032105). https://gitlab.com/gitlab-org/gitlab/-/work_items/521903+ is the size alert. - **Alignment with CI tables.** CI tables are list partitioned by `partition_id`, and CI requires that strategy for tables in its domain. `ci_job_artifact_states` already references `p_ci_job_artifacts` through `(partition_id, job_artifact_id)`, so the remaining steps finish what was started. This epic was first opened for the Geo-based organization move on GitLab.com, which would have backfilled this table for every selectively synced organization. That plan was dropped, but the reasons above still hold. ## Status Master already has the results of steps 1 to 5 of the [list partitioning guide](https://docs.gitlab.com/development/database/partitioning/list/): the table has `partition_id`, its primary key is `(job_artifact_id, partition_id)`, and its foreign key to `p_ci_job_artifacts` uses both columns. Step 6, creating the parent table and attaching the existing table as the initial partition, was attempted in https://gitlab.com/gitlab-org/gitlab/-/merge_requests/190290+ and closed unmerged on 2025-08-07. Steps 6 to 11 remain. - https://gitlab.com/gitlab-org/gitlab/-/work_items/525415+ tracks the remaining partitioning work. It is unassigned and in the Backlog milestone. - https://gitlab.com/gitlab-org/gitlab/-/work_items/525414+ truncated the table on GitLab.com. Done. ## Risks - The migrations run on every instance, including self-managed and Dedicated instances where this table can be large. Plan step 6 onward for large tables and around required stops. - The earlier primary key change caused a severity 2 regression on Geo primaries, https://gitlab.com/gitlab-org/gitlab/-/work_items/557340+, where `VerificationStateBackfillService` failed with "No unique index found for job_artifact_id". At each step, check the Geo verification state code (backfill, upserts) for assumptions about the table's keys. - The table has two triggers, `trigger_a465de38164e` and `trigger_mark_geo_ci_job_artifact_summary_dirty`, which must carry over to the partitioned parent. ## Exit criteria `ci_job_artifact_states` is a list partitioned table, the application uses it, and all child items are closed. <details><summary>Generated by Claude Opus 5.5 high in Claude Code</summary> Session a324da6e-c3e3-425a-bbb8-92214e040060 </details>