Raise prod CPU limit to 2 and request to 500m to stop CFS throttling

[elffjs]

Why

Prod pods are being CPU-throttled in 5–35% of enforcement periods, continuously, on every pod (CPUThrottlingHigh alerts) — despite averaging only 0.2–0.33 cores against the 1-core limit.

The cause is a burst/limit mismatch, not capacity: Go 1.25's container-aware GOMAXPROCS has a documented floor of 2 (confirmed live: go_sched_gomaxprocs_threads 2), so the process can consume CPU at 2× the 1-core quota during bursts, exhausting a 100ms enforcement period's allowance in 50ms and then stalling for the remainder. Each stall lands on in-flight work — webhook HTTP sends and Kafka batches — adding tail latency that feeds both the consumer lag and the spurious webhook failure counts we've been fixing.

What

• CPU limit 1 → 2: matches GOMAXPROCS so bursts can no longer outrun the quota. Limits don't affect scheduling/bin-packing, so this reserves nothing on the node.
• CPU request 250m → 500m: reflects actual usage (~0.3 cores average) for honest scheduling and CFS share weight.

Verification

After deploy, the throttle ratio should drop to ~0:

sum by (pod) (rate(container_cpu_cfs_throttled_periods_total{namespace="prod",container="vehicle-triggers-api"}[5m]))
  / sum by (pod) (rate(container_cpu_cfs_periods_total{namespace="prod",container="vehicle-triggers-api"}[5m]))

🤖 Generated with Claude Code