Self-hosted Supabase: supabase-envoy 'rootlessport listen tcp 0.0.0.0:8000: bind: address already in use'; changing the port default in docker-compose.yml does nothing, .env sets API_GW_HTTP_PORT
Change the host port in .env, not in the compose file: set API_GW_HTTP_PORT=8200 (KONG_HTTP_PORT is only a fallback for it) plus SUPABASE_PUBLIC_URL=http://localhost:8200 and API_EXTERNAL_URL=http://localhost:8200/auth/v1, then recreate: docker compose up -d --force-recreate api-gw auth. Verify with docker port supabase-envoy (should show 8000/tcp -> 0.0.0.0:8200).
Current upstream docker/docker-compose.yml uses Envoy as the default API gateway (service api-gw, container supabase-envoy, image envoyproxy/envoy) instead of Kong, with ports: - ${API_GW_HTTP_PORT:-${KONG_HTTP_PORT:-8000}}:8000/tcp. The shipped .env.example defines both API_GW_HTTP_PORT=8000 and KONG_HTTP_PORT=8000, so editing the inner default to 8100 changes nothing: we did exactly that, curl to :8100 got 'Connection refused' and podman port supabase-envoy still said 0.0.0.0:8000. Variables put in a .env.local are not read by compose either; only .env (or --env-file) is.
When the gateway can't bind, the failure cascades: Container supabase-edge-functions not started: condition "service_healthy" of service "api-gw" not reached: interrupted by the failure of Container supabase-envoy. Find the squatter with docker ps --format '{{.Names}} {{.Ports}}' | grep ':8000->'. Inside the compose network services still talk to http://api-gw:8000, so only the host side changes; point supabase-js / VITE_SUPABASE_URL at the new host port.