Multi-cluster

One app for every cluster you run.

Most teams that take Argo Workflows seriously end up with more than one cluster. Prod and staging at the very least, usually a region or two, and something the data team stood up that you inherited. dagweave treats that as the normal case.

The problem

The cluster that never got connected.

When each cluster costs, somebody decides which ones are worth connecting. Staging never gets added, the EU region arrives six months late, and the cluster the data team runs stays outside the tool.

Which clusters get connected becomes a budget decision.

dagweave never counts clusters.

Enrollment

Enroll as many as you have

Each cluster gets its own connector and its own keypair, generated inside that cluster. One install covers the Argo Workflows instances running there.

One helm install per cluster. Enrolling the tenth cluster works like the first.

what a workspace can hold text

connectors            ->  as many as you enroll
served instances      ->  several per connector
each served instance  ->  one Argo Workflows instance, one namespace

Dispatch

Pick the target when you run it

The workflow and the cluster it runs on are separate choices. The same graph goes to staging on a branch push and to prod when someone approves it, without being copied or forked.

A cron in the EU region and a cron in us-east are one workflow with two triggers.

one workflow, two targets text

push to main   ->  connector: staging-eu   ns: pipelines
manual, gated  ->  connector: prod-us      ns: pipelines

Access

You choose who reaches which cluster

A cluster connection is its own permission, separate from whatever someone can already do to a workflow.

Give someone staging without prod. Revoke one, keep the other.

permission is layered, not inherited text

edit the workflow    ->  workflow permission
send it to staging   ->  staging-connector permission
send it to prod      ->  prod-connector permission

Fleet health

You find out when a connector drops

Every connector you enroll is watched, including the ones in regions nobody checks. dagweave tells you when one goes offline.

You get alerted when a connector drops, not just when a run fails.

Connect your cluster How a cluster gets enrolled, and who can dispatch to it afterwards. Security for cluster admins Zero inbound connections, and a keypair generated inside each cluster. Triggers and runs The triggers that carry a cluster target, and what comes back.