ichor cluster rotate-ca
Replace the cluster's certificate authorities (Talos API, Kubernetes API, and optionally etcd and aggregator)
Synopsis
rotate-ca starts a CA rotation operation: the Talos API CA and the Kubernetes API CA are replaced in passes over every machine (accept the new CA, make it the issuer, drop the old one), each verified before the next. The kubelet restarts on each machine in the passes that change the Kubernetes CA.
--ca chooses the CAs (talos, kubernetes, etcd, kubernetes-aggregator; repeat the flag or separate with commas), --all takes the four. Without either, the Talos API and Kubernetes API CAs are replaced. The etcd CA is replaced by rebooting every control plane, one at a time, after each of the three passes (Talos neither restarts etcd on a configuration change nor lets an API call do it): each node is cordoned and drained, hands etcd leadership over, reboots, and the next one waits until it is back, Ready and etcd is healthy; with one control plane the Kubernetes API is unavailable during each reboot. That is three reboots of every control plane. Take an etcd snapshot first ('ichor etcd backup now NAME'). Deleting the operation before the new CA issues abandons the rotation (the old configuration is rendered again); 'ichor config apply' then drops what the nodes still accept. The aggregator CA needs the extension API servers (metrics-server, for one) to load the new CA: Ichor waits for kube-apiserver to publish it and restarts none of them. A rotation of etcd and the aggregator CA alone changes nothing your kubeconfig or talosconfig depends on, so it has no credential hand-off. The service-account signing key and the secretbox key are not rotated.
When the Talos API or Kubernetes API CA is replaced, your kubeconfig and talosconfig STOP WORKING from the second pass (Issuing) on: the nodes start serving certificates from the new CA, and a file that trusts only the old CA can no longer verify them. So after every machine accepts the new CAs, the operation WAITS, changing nothing, until you have new credentials. This command waits for that point while your current credentials still work, then writes
a kubeconfig and a talosconfig that trust the old and the new CAs and are signed by the new one (next to your current ones, with the suffix .ichor-new; your current files are not touched; mode 0600),
checks that the new kubeconfig authenticates, and only then lets the operation go on. The hand-off needs your own kubeconfig, so this command does it with or without --follow: it stays until the operation has gone on (a few seconds to a few minutes), then returns, and the rest of the rotation runs in the controller. The files' contents are never printed. If this command is interrupted while the operation waits, run 'ichor cluster credentials NAME --acknowledge' with your current kubeconfig. If you lost both, 'ichor cluster credentials NAME --bundle FILE' makes credentials from a current bundle file.
Before it starts, every machine must answer, run the configuration Ichor renders for it, and every Kubernetes Node must be Ready; otherwise the operation pauses without changing anything. While it runs, the machine controller does not reconcile the cluster's machines and other disruptive operations are refused. It resumes by itself after the controller restarts, and does not advance past a step it could not verify. Machines held with upgrade-hold are not skipped.
When it completes, the secrets bundle in the cluster has been replaced, and the copy saved on your machine is stale: run 'ichor cluster bundle-export NAME' and keep the file safe.
It shows which machines are reconfigured and in which order, and asks for confirmation unless --yes. --follow keeps reporting progress until the rotation completes, through the Kubernetes API with the new kubeconfig (the ichor-ui tunnel does not survive the rotation).
ichor cluster rotate-ca NAME [flags]Options
--all replace all four CAs
--ca strings the CAs to replace: talos, kubernetes, etcd, kubernetes-aggregator (default: talos and kubernetes)
--follow after the credential hand-off, follow the rotation until it completes or fails (without it the command still does the hand-off, then returns)
-h, --help help for rotate-caOptions inherited from parent commands
--bundle-passphrase-file string file holding the passphrase of the local secrets bundle; any readable path, e.g. /dev/fd/3; otherwise $ICHOR_BUNDLE_PASSPHRASE, then a prompt on a terminal. Never a value on the command line
--ca-file string PEM file of the CA that signed ichor-ui's serving certificate, when ichor-ui serves TLS (env ICHOR_CA_FILE)
--ichor-namespace string namespace Ichor is installed in (where ichor-ui runs and the secrets bundle lives) (default "ichor-system")
--kubeconfig string path to the kubeconfig for the cluster's Kubernetes API; defaults to $KUBECONFIG, then ~/.kube/config
--local do not use ichor-ui; compute from the Kubernetes API directly
-n, --namespace string namespace of the Ichor resources the command names (default "ichor-system")
-o, --output string output format: "yaml" or "json" for scripts; by default a command prints for people (commands without a human view print YAML)
--override incident only: override a safety check (reset a node, restore etcd, reboot or shut down the last worker, an upgrade the compatibility table refuses); needs the override RBAC verb on the resource; implies --local
--talosconfig string path to the talosconfig used to reach each node's Talos API; defaults to $TALOSCONFIG, then ~/.talos/config
--tls-server-name string name on ichor-ui's serving certificate, when it is not the Service's DNS name (<service>.<namespace>.svc)
--yes skip interactive confirmation before a dangerous operationSEE ALSO
- ichor cluster - List or inspect clusters