Getting started with Swarm

Define a cluster

Settings → Swarm → Clusters → Manage clusters…, then New cluster….

A cluster needs a name and at least one candidate manager — a saved connection that can reach a Swarm manager node. Cleat monitors through the first candidate that answers and moves down the list when one stops answering.

Give it more than one. Monitoring through a single manager means your monitoring dies at the exact moment the cluster gets interesting, and Cleat will say so next to the field if you only add one.

The order is your failover preference. Cleat remembers which candidate answered last and resumes there on the next start, rather than walking the list from the top.

What the SSH user needs

The user Cleat connects as must be able to run docker on the manager. Either:

• add the user to the docker group, or
• allow passwordless sudo for docker and set the cluster's Docker command to sudo -n docker

The -n matters. Without it, sudo waits for a password that nothing will ever type, and the poll hangs instead of failing.

Cleat only runs read commands unless you turn on control actions — see Control actions.

Per-cluster settings

Most clusters should leave the poll intervals empty, which means "follow the global setting" in Settings → Swarm. Set them per cluster only when one cluster is large enough or far enough away to deserve its own cadence.

Allow control actions is per cluster and off by default. It is one of two switches between a misclick and a production service being scaled to zero.