Destroy
A tofu destroy takes each shc_node out of the cluster before its machine
is deleted, and a node that still carries deployments refuses to go. That
refusal names the deployments in the way, and it is the right answer for a
node you are retiring from a cluster that lives on. alice is refused the
same way, as the last manager of prod. For a whole-cluster teardown you
state the intent once: on_destroy = "force" on every node and on alice.
It is read from state, so apply it before the destroy.
home_region = "nbg1"far_region = "hel1"home_count = 4server_type = "cx23"shc_version = "0.1.3"on_destroy = "force"$ tofu -chdir=learn/cluster plan...Plan: 0 to add, 6 to change, 0 to destroy....$ tofu -chdir=learn/cluster apply -auto-approve...Apply complete! Resources: 0 added, 6 changed, 0 destroyed....Now destroy. The order is the reverse of the applies: the record, the
connection and the deployments go first, then the nodes leave the cluster,
then Hetzner deletes the machines, the firewalls, the network and the key.
The depends_on = [module.shc] lines of pages 04 and 05 hold that order:
without them a node can leave while a deployment on it is still being
removed, and that uninstall has nowhere to run.
$ tofu -chdir=learn/cluster destroy -auto-approve...$ tofu -chdir=learn/cluster state listThe state is empty. Ask Hetzner directly: every resource the module made
carried the label cluster=learn, and none is left.
$ curl -s -H "Authorization: Bearer $HCLOUD_TOKEN" "https://api.hetzner.cloud/v1/servers?label_selector=cluster=learn"{ "servers": [], "meta": { "pagination": { "last_page": 1, "next_page": null, "page": 1, "per_page": 25, "previous_page": null, "total_entries": 0 } }}$ curl -s -H "Authorization: Bearer $HCLOUD_TOKEN" "https://api.hetzner.cloud/v1/networks?label_selector=cluster=learn"{ "meta": { "pagination": { "last_page": 1, "next_page": null, "page": 1, "per_page": 25, "previous_page": null, "total_entries": 0 } }, "networks": []}$ curl -s -H "Authorization: Bearer $HCLOUD_TOKEN" "https://api.hetzner.cloud/v1/firewalls?label_selector=cluster=learn"{ "firewalls": [], "meta": { "pagination": { "last_page": 1, "next_page": null, "page": 1, "per_page": 25, "previous_page": null, "total_entries": 0 } }}$ curl -s -H "Authorization: Bearer $HCLOUD_TOKEN" "https://api.hetzner.cloud/v1/ssh_keys?label_selector=cluster=learn"{ "ssh_keys": [], "meta": { "pagination": { "last_page": 1, "next_page": null, "page": 1, "per_page": 25, "previous_page": null, "total_entries": 0 } }}The DNS record went with the cluster. Neither the name nor the ownership TXT managed DNS wrote beside it resolves any more:
$ dig +short tf.$TF_VAR_dns_zone$ dig +short TXT _shc-owner.tf.$TF_VAR_dns_zoneThe licence token on work and the state file’s history are yours to keep or shred; the state file still holds the retired vault password and the SSH key.
The host keys in learn/cluster/ssh/known_hosts belonged to the deleted
machines, and the destroy does not remove that file. Hetzner can give a new
machine an address an old one had, and SSH then refuses it as a changed
host. Remove the file.
$ rm -f learn/cluster/ssh/known_hostsThe ladder is done. The same root module, with the count and the version
changed back, builds the cluster again from nothing. The pages after this one
each take another example root from terraform/examples and show what
differs from the Hetzner front door you wrote.
Next: The hetzner-shc example.