Zero-Downtime Deployment Script with Docker Compose: Migrations, Health Checks, Rollback

This post builds a ~100-line bash deployment automation script (zero-downtime in the blue/green sense described below) for a small Node + Postgres app on Docker Compose and nginx. It builds, backs up, migrates, health-checks, switches traffic and rolls back. I then shipped deliberately broken releases through it under a request loop. In my runs, a release that passes its healthcheck but returns 500 on / cost 13 and 15 failed requests with only a post-switch probe, and none with a pre-switch smoke test. If you later want this triggered from a pipeline, see the CI/CD pipelines with Jenkins and Docker guide.

Scope: "zero downtime" here means the old color is never stopped, so you can always swap back, not that requests never fail. Everything below ran on Docker Desktop on Windows 11 (Git Bash), where a bare docker compose exec takes about a second. I did not run it on Linux, so treat every timing and failed-request count as specific to this setup; drop export MSYS_NO_PATHCONV=1 on Linux.

Zero-downtime deployment with Docker Compose: the blue/green setup

The demo app is a Node 22 service (/, /version, /healthz) talking to Postgres 17. blue and green are the same image under two service names, and nginx sends traffic to one of them. Versions are pinned in the files below (node:22-alpine, postgres:17-alpine, nginx:1.27-alpine, pg 8.13.1); in the re-run for this post docker version printed 29.7.2 for client and server and docker compose version --short printed 5.3.1.

compose.yml. The [healthcheck](https://docs.docker.