blissful-infra canary

canary drives the Argo Rollout that deploy starts. A rollout shifts traffic in steps and pauses between them; these commands let you promote early, hold, or abort back to stable.

blissful-infra canary status orders

Subcommands

Command	What it does
`canary status [service]`	Show rollout status: weight, step, health
`canary promote [service]`	Promote to the next step, or fully with `--full`
`canary abort [service]`	Abort the rollout and go back to stable
`canary pause [service]`	Pause at the current step
`canary resume [service]`	Resume a paused rollout
`canary test [service]`	Test canary rollback behaviour

Common options

Every subcommand takes:

Flag	What it does
`--tenant <tenant>`	Tenant (defaults to context)
`--project <project>`	Project (defaults to context, falls back to a registry scan)

canary promote additionally takes --full to skip the remaining steps and go straight to 100%.

The steps

A rollout walks four weights with a pause after each of the first three:

Step	Weight	Pause
1	10%	2m
2	25%	2m
3	50%	5m
4	100%	-

Left alone, it walks the whole way on its own. The pauses exist so you have a window to look at the service under partial traffic and decide.

Promoting

blissful-infra canary promote orders          # next step only
blissful-infra canary promote orders --full   # skip the rest, go to 100%

Aborting

blissful-infra canary abort orders

Sends traffic back to the stable version immediately. The canary ReplicaSet is scaled down. This does not revert the gitops commit. The manifests in Gitea still describe the new version, so ArgoCD may re-sync it. To undo the deploy properly, use rollback.

Why pauses instead of metrics

Argo Rollouts supports analysis steps that query Prometheus and promote or abort automatically. That needs an in-cluster Prometheus, which the current slice does not ship: the tenant’s Prometheus runs on the compose side. So the rollout template uses timed pauses and leaves the judgement call to you.

Wiring in-cluster Prometheus so analysis steps can come back is tracked as a follow-up in ADR-0020.

From the dashboard

The Environments tab renders the same rollout as a card with a live weight bar, a step counter and Promote / Promote Full / Abort buttons:

blissful-infra dashboard up

The card only appears for kubernetes-runtime projects. Compose projects have no Rollout.