zoorikctl

zoorikctl connects a Kubernetes cluster to Zoorik Coral, Zoorik's Kubernetes optimization. It prepares the cluster, installs the Zoorik agent, lets Zoorik remove spare machines if you agree, and takes everything out again on disconnect. It is open source, so you can read exactly what it does before you run it.

Install

macOS	brew tap getzoorik/tap && brew trust getzoorik/tap && brew install zoorikctl or curl -fsSL https://get.zoorik.com/macos | sh
Linux	curl -fsSL https://get.zoorik.com/linux | sh
Windows	scoop bucket add getzoorik https://github.com/getzoorik/scoop-bucket && scoop install zoorikctl or irm https://get.zoorik.com/windows.ps1 | iex

Binaries for macOS, Linux and Windows on amd64 and arm64 are on the Releases page.

Every installer checks the archive against the release's checksums.txt before it writes anything. When cosign is installed, it also checks that checksums.txt was signed by Zoorik's release workflow. To check a release by hand:

cosign verify-blob checksums.txt \
  --signature checksums.txt.sig \
  --certificate checksums.txt.pem \
  --certificate-identity-regexp '^https://github.com/getzoorik/(coral/\.github/workflows/zoorikctl-release|zoorikctl/\.github/workflows/release)\.yaml@refs/heads/main$' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

This command matches every release from v0.4.2 on. Up to v0.4.4, releases were built outside this repository; from the next release on, they are built here by release.yaml.

Use it

The Connect dialog in the Zoorik console prints the command to run, with a one-time connect token:

zoorikctl cluster connect --connect-token=<token> --api-region=us

Command	What it does
zoorikctl cluster prepare	Reads the cluster, says what adds its nodes, reports whether each node pool is ready for live migration, and sets up a node provisioner where the cluster has none.
zoorikctl cluster connect	Installs the Zoorik agent with the install the connect token was opened for.
zoorikctl cluster grant-node-deletion	Lets Zoorik remove machines it has emptied, and lower a node pool's minimum.
zoorikctl cluster disconnect	Undoes what Zoorik changed and removes the install.
zoorikctl version	Prints the version.

Every command acts on your kubeconfig's current context, or on --kube-context (or --context, as in kubectl), and prints that context first. --dry-run prints the full plan, with every command it would run, and changes nothing. Every change is printed before it is made, and nothing in your cloud account changes until you say yes (or pass --yes). Run zoorikctl --help for every flag and exit code.

What it changes

In the cluster

connect applies the install it downloads from Zoorik, as rendered:

• the namespace zoorik-agent, with the agent, a pod webhook that sizes pods as they are created, and the Evictor. The webhook fails open, so it never blocks a pod;
• the Recommendation and ActionRequest resource definitions, and the roles the agent needs, all labelled app.kubernetes.io/part-of=zoorik-agent;
• two admission policies, zoorik-agent-release and zoorik-agent-release-images, that bound how the agent updates itself. It never widens its own permissions;
• only when the connect token was opened with live migration: the DaemonSet zoorik-node in the namespace zoorik-node. It is privileged, and places one file on each node, a static CRIU at /usr/local/bin/criu. It installs no packages.

zoorikctl itself asks Zoorik only for the install, and sends nothing about your cluster. The agent it installs reports to Zoorik once it runs. On EKS, prepare also pulls the Karpenter Helm chart from public.ecr.aws.

In your cloud account

Cloud changes run through your own az or aws sign-in. zoorikctl never stores a cloud credential.

• Node provisioning (prepare, and the first question in connect). Where the cluster has no provisioner and no cluster autoscaler: on AKS it turns on Node Auto Provisioning; on EKS it installs Karpenter with its IAM roles, interruption queue and discovery tags, and, where nodes join only through the aws-auth ConfigMap, switches the authentication mode to API_AND_CONFIG_MAP, which cannot be switched back. A provisioner or autoscaler you already run is kept: prepare only adds a default NodePool where it has none of the cluster's own, and the node class zoorik-base that Zoorik's own pools copy.
• Removing spare machines (grant-node-deletion, and the second question in connect). On AKS: it turns on workload identity and the OIDC issuer if they are off, and creates a managed identity zoorik-agent-<cluster> with the custom role "Zoorik Node Removal v2" on this cluster only. On EKS: an IAM role zoorik-agent-<cluster> whose policy is limited to this cluster's node groups, bound with EKS Pod Identity.

Zoorik never creates a machine or raises a node pool's size. Before the agent lowers a minimum, it records yours in a tag, and disconnect puts it back, which may add machines up to it.

On disconnect

disconnect needs only your credentials and works whether or not the agent still runs. It removes the webhook, roles, policies and namespaces, removes Zoorik's taints and annotations from the nodes, hands back the autoscalers Zoorik took over, and deletes the resource definitions. It then offers to undo the cloud changes zoorikctl made for node removal, lowered minimums included. A disconnect that stops part-way picks up where it left off when you run it again.

Build from source

You need Go (the version in go.mod).

git clone https://github.com/getzoorik/zoorikctl.git
cd zoorikctl
go build -o bin/zoorikctl ./cmd/zoorikctl
./bin/zoorikctl version

This is the beginning of the README. Read the rest at the original. Read the full README ↗