Closed braska closed 3 months ago
Should be addressed with configurable health checks.
Also wait
wasn't deprecated, it was just broken immediately. https://github.com/siderolabs/terraform-provider-talos/pull/105 removed the logic but left the parameter. That's not really deprecation and it's actually kind of worse than just removing the parameter. Deprecation would leave it working with a clear alternative but mark that it's going to be removed.
@smira is there any information about configurable health checks? I can't find anything here or on https://github.com/siderolabs/talos
@smira is there any information about configurable health checks? I can't find anything here or on siderolabs/talos
There are no configurable health checks, my message was a planning note on how to implement a better way.
Since
wait
intalos_cluster_kubeconfig
was deprecated there is no way to await for Kubernetes API to be available.I have a use case when I deploy Talos cluster with
cluster.network.cni.name = "none"
. Sotalos_cluster_health
never resolves until I deploy CNI (it is actually even crushing on a first tf apply withrpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 10.40.0.10:50000: connect: connection refused"
, but that's a different issue). My plan was to use Helm Terraform provider to deploy CNI immediately after Kubernetes API becomes available. But there is no way to do it with latest version.As a workaround, I will use
null_resource
and custom script to check Kubernetes API availability, but what to you think about un-deprecatingwait
param intalos_cluster_kubeconfig
? Or maybe better to introduce changes totalos_cluster_health
and make it return detailed health report with bunch of true/false values related to health of individual components?