ai-code-battle/notes/bf-1bvca.md

50 lines
2.2 KiB
Markdown

# bf-1bvca: combat_turns Migration
## Task
Deploy P0: add combat_turns column migration to acb-schema-init (apexalgo-iad)
## Status: COMPLETE
### What Was Done
The combat_turns migration was already implemented in a previous session (commit `00e1f5c`). Verified the following:
1.**Schema Changes** (`declarative-config/k8s/apexalgo-iad/ai-code-battle/acb-schema-init.yml`):
- Line 46: `combat_turns INTEGER NOT NULL DEFAULT 0` in CREATE TABLE
- Line 305: `ALTER TABLE matches ADD COLUMN IF NOT EXISTS combat_turns INTEGER NOT NULL DEFAULT 0;`
2.**Rollout Annotation**: Bumped to `v13-combat-turns-2026-06-03-bf-1bvca-sync` (latest)
3.**Deployed**: declarative-config is at v13, ArgoCD sync operation is Running (waiting for healthy state)
4.**Pushed**: declarative-config is up to date with origin/main
### Current State (Infrastructure Blockers)
The migration code is correct and committed. However, cluster resource constraints prevent verification:
1. **Cluster CPU Capacity**: All application pods (api, index-builder, worker, evolver, matchmaker, etc.) are stuck Pending due to "Insufficient cpu"
- Only schema-init pod (old v11 revision) is Running (1/1)
- New v13 schema-init pod is Pending waiting for CPU
- Cannot verify index-builder succeeds until pods can schedule
2. **Cluster Status** (2026-06-04):
- declarative-config: Commit `d3e9eab` (v13) pushed
- Cluster apexalgo-iad: Stuck at `v11-fix-secret-name-2026-06-03-bf-1bvca`
- Node CPU: 42%, 17%, 42% utilization but pods can't schedule
- 10+ pods Pending with `FailedScheduling: 0/3 nodes are available: 3 Insufficient cpu`
### Git History
Multiple commits to apply this migration:
- `00e1f5c` feat(apexalgo-iad): add acb-schema-init deployment with combat_turns migration
- `5abffac` fix(ai-code-battle): correct schema-init secret name reference
- `6d7439d` fix(acb-schema-init): bump checksum to force reapply combat_turns migration
- And 8+ annotation bump commits attempting to force rollout
### Next Steps
To complete verification:
1. Fix postgres cluster (cnpg-apexalgo) - currently broken for 23 days
2. Scale up cluster CPU or scale down workloads to free capacity
3. Once index-builder pod runs, verify logs show no "combat_turns does not exist" errors