Remove the sshing in the platform about caravan-platform HOT 4 CLOSED

The SSH-ing does only one thing there, it replaces the consul.hcl.tmpl with one that contains a placeholder for the token:

   {{ with secret "consul/creds/consul-agent-role" }}tokens = { default = "{{ .Data.token }}" }{{ end }}

this can already be done in the consul-cluster. The consul-server.hcl.tmpl there is not being used until the vault-agent auto-authenticates. At which point it already has access to consul/role/... backend and can replace the token.
As soon as the token is there, the consul watcher should take care of restarting it.
This is the shady part:

"while [ \"$(curl -s 'http://127.0.0.1:8500/v1/status/peers' | jq '. | length')\" != \"3\"  ]; do echo \"Consul: no enough peers\"; sleep 3; done",

What the purpose of it and can we move to the consul watcher @efbar ? There is already a 60s sleep before the restart and we can add this waiter too before or after the restart with a sane TTL.

from caravan-platform.

That "wait_and_restart" part was added due to the fact that we had to be sure that a complete (number of nodes) cluster is formed before restart vault-agent and nomad. The culprit was that if you have a not complete cluster formed, with tokens and ca applied, you cannot add the last lonely node without acting through api or stuff like that.
Without that "wait and restart", at the end of the platform automation, you could have a not complete and unstable consul cluster going up and down.

from caravan-platform.

Which probably happens because the vault's cluster backend is not available so the vault-agent can't get a token so only one node has time to reload with token. I get it. We should try to move this waiting to the restarter then and we should be all set.

from caravan-platform.

Probably it could involve also the gcp auth too (and in this case it could be also google fault). Anyway you got the point: that "wait" is a sort buffer time for the right alignment of the consul cluster.

from caravan-platform.