Commit aa0d31d
fix(provisioner): raise hot-pool POOL_*_SIZE 2/3/2 -> 10, add POOL_QUEUE_SIZE
Companion to provisioner master 4799395 (concurrent hot-pool refill — the F1
provisioning latency-cliff fix from LOAD-CHAOS-REPORT-2026-05-19).
The deployment pinned POOL_POSTGRES_SIZE/REDIS/MONGO to 2/3/2, which overrode
the provisioner code defaults. A concurrency-8 burst drained the pool in the
first few requests and every caller after fell through to slow (15-25s) live
provisioning. With the refill now concurrent (bounded), a 10-deep warm buffer
absorbs the burst on the fast pool-claim path. POOL_QUEUE_SIZE was absent
entirely — added at 10 to match.
Env values must stay >= the code default in
provisioner internal/config/config.go.
NOTE FOR OPERATOR: infra has no auto-apply. Apply manually:
kubectl apply -f k8s/provisioner/deployment.yaml
kubectl rollout status -n instant-infra deploy/instant-provisioner
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>1 parent 5e9cc7e commit aa0d31d
1 file changed
Lines changed: 13 additions & 3 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
122 | 122 | | |
123 | 123 | | |
124 | 124 | | |
| 125 | + | |
| 126 | + | |
| 127 | + | |
| 128 | + | |
| 129 | + | |
| 130 | + | |
| 131 | + | |
| 132 | + | |
125 | 133 | | |
126 | | - | |
| 134 | + | |
127 | 135 | | |
128 | | - | |
| 136 | + | |
129 | 137 | | |
130 | | - | |
| 138 | + | |
| 139 | + | |
| 140 | + | |
131 | 141 | | |
132 | 142 | | |
133 | 143 | | |
| |||
0 commit comments