We need to reduce compute start-up time. It frustrates our customers when they need to wait. This task is done when startup time p90 is 100ms

Task 1: 100ms p90 in us-east-2 in calm conditions

• Decrease inside pod p90 time to 90ms
• Compute pools
• https://github.com/neondatabase/cloud/issues/6243
• https://github.com/neondatabase/cloud/issues/6242
• Remove wal from basebackup #4912

Task 2: Good results during edge cases

• Compute pool size auto-tuning
• Optimize pool miss (no kube proxy)
• Slow console queries 1h after deploy due to cache warming
• https://github.com/neondatabase/cloud/issues/5620

Task 3: Performance parity in all regions

• Use the same metal nodes as us-east-2 (or better) everywhere
• control plane split

Task 4: Observability

• e2e metrics from proxy (including time to auth)
• alerts