shipline
One deploy pipeline for every service. Build, test, canary, promote, with a gate that compares the new version against the old one.

Shipline is the deploy pipeline behind every service at Northwind Labs, extracted into an open-source tool. It runs four stages, in order, for every merge to main:
- Build a container image, tagged with the commit.
- Test it: unit tests, then a smoke test against the built image.
- Canary it: route 5% of traffic to the new version, then 25% and 50%, checking error rate and p95 latency against the stable version between steps.
- Promote it to 100% and keep the previous version warm for a fast rollback.
Why it exists
Before Shipline, Northwind had three deploy paths and a median of two days from merge to production. Now there is one path and the median is 12 minutes. The story is in One pipeline, forty services.
Try it
npx shipline init --service my-service
That writes a shipline.yaml with the defaults and a GitHub Actions workflow that runs it:
service: my-service
canary:
steps: [5, 25, 50, 100]
hold: 10m
gate:
error_rate: +0.5pt
p95: +20%
freeze:
- fri 16:00 - mon 08:00
What it is not
Shipline does not manage clusters and has no service mesh. It needs a Kubernetes cluster, an NGINX ingress controller with canary annotations, and Prometheus. Canary deploys without a service mesh explains why.