· Product · 5 min read
Dependency-ordered deploys: starting Postgres before the API
By Ethan Cole, Co-Founder & CTO
Container start order is not readiness. A service graph only stops boot-time crash loops when upstream dependencies pass health checks before downstream services start.
Anyone who has brought up a fresh environment from scratch knows the noisy first minute: the API container starts, fails to reach Postgres, exits, and enters a backoff loop before the database finishes initializing.
Starting containers in topological order is not enough by itself. A process PID existing inside a container does not mean the socket is listening or the service is ready to serve queries.
Because Nimbus models your stack as a directed service graph, dependencies carry explicit readiness gates. If api depends on postgres, Nimbus waits for postgres to pass its readiness check before scheduling the api rollout.
The same graph governs partial deploys and fresh preview environments alike. New teammates no longer need tribal knowledge about which service has to be restarted twice after a cold boot.
Startup order should live alongside the architecture itself, enforced automatically on every deploy rather than documented in a wiki page nobody reads during an outage.