Subject: Cold function isolates: ~50% of first outbound connections to external Postgres endpoint (AWS eu-west-1) hang past 10s; instant retries succeed
Project: mcwv-hub (prj_wbRccK5ik3J1cjvSGBEvIGqHnIZz), team ilostmymicrowaves-projects, Hobby plan. Function region: lhr1 (previously iad1 — same behaviour on both). First observed: 2026-09-12 ~17:44 UTC. Deployments affected: dpl_qyrEkpRXZmgVizGmdf5qLwF22xQn, dpl_GzyPLFaUiRBjJ8JsYEsHAsKCezTF.
What happens: When a function isolate cold-starts, its first outbound TCP/TLS connection to our Postgres pooler (aws-1-eu-west-1.pooler.supabase.com:6543, behind an AWS eu-west-1 ELB) frequently hangs and aborts after 10+ seconds. An immediate retry from the same isolate connects in under a second. ~30–50% of first connections are affected, in correlated bursts of 20–30s. When three attempts fail in a row, our API returns 503s to users.
Evidence this is on the egress path, not the target: From an external host, identical connections to the same endpoint succeed 27/27 (TCP+TLS to all 3 ELB IPs, ~450–600ms) and 5/5 (full Postgres auth, ~1.4s). Not fixed by changing function region (iad1→lhr1) or restarting the target. All target LB nodes healthy. Warm isolates and established connections are fast and stable — only NEW connections from COLD isolates fail.
Example request IDs (2026-09-12 UTC): zdr5w-1789237536432 (login → 503 after 3×10s timeouts, 18:25:36) · xq4sg-1789237705751 · fzftn-1789237702509 · fmr6h-1789237702442 · vcvg7-1789237687030 (internal API → 500, 18:28:07)
Ask: Please investigate the egress path from lhr1/iad1 isolates to AWS eu-west-1 around these timestamps — specifically whether SYNs or first packets from shared egress IPs are being dropped (NAT rebinding, rate limiting, degraded path). Happy to deploy a minimal repro function that times cold-start outbound connects.