A single-VPS deploy usually means one of two things: a few seconds of downtime while the container restarts, or a hand-rolled blue/green setup that is more infrastructure than the app it’s protecting. Neither felt right for a personal stack, so I built Rollhook — a small webhook-triggered service that rolls a Docker Compose stack one replica at a time.

The problem with docker compose up -d

Restarting a service in place drops every in-flight request the moment the container exits, and the replacement isn’t ready to receive traffic until its health check passes. For a one-container service, that gap is unavoidable unless you run more than one replica and roll them individually.

The goal isn’t zero restarts — it’s zero requests hitting a container that isn’t ready yet.

How the rollout works

• A webhook (triggered from CI) hits Rollhook with the service name and new image tag
• Rollhook scales up one new replica alongside the old ones
• It polls the container’s health check endpoint until it reports healthy
• Traffic only shifts to the new replica once the check passes¹
• The old replica drains its remaining connections, then stops

// simplified rollout loop
async function rollService(service: string, replicas: string[]): Promise<void> {
  for (const replica of replicas) {
    await startReplica(replica);
    await waitUntilHealthy(replica);
    await drainAndStop(previousReplica(replica));
  }
}

The proxy in front (Traefik) only ever sees healthy backends in its pool, so from the outside the deploy is invisible — no dropped connections, no maintenance page.

Rollhook itself is intentionally small: a webhook receiver, a health-check poller, and a Compose wrapper. No orchestrator, no cluster — just enough automation to make deploys boring, which for a personal-infra project is exactly the point.