Terraform forces full VMCluster CR replace because of VM* CRD schemas

[artmakh]

Hello!

We are using VMOperator + VM CR recourses, and manage them via terraform.

We modify Cluster CR inside terraform configuration and apply them via terraform's kubernetes provider.
In some cases this results in full replace of CR inside kubernetes cluster, which results in downtimes.

For example, if you modify anything inside vmstorage volumeClaimTemplate it will force replace of full CR, instead of modifying it.
This happens because of x-kubernetes-preserve-unknown-fields: true flag, which exists in a lot of parts of victoriametrics CRDs.

This flag basically forces terraform to replace resource, because there is no provided schema for terraform inside this part of CRD. Here is better explanation from one of Terraform's developers .

Is it possible to add schema information for at least some, presumably often modified, parts of Victoriametrics CRDs?

[f41gh7]

Hello, thanks for reporting. It's really unexpected to me, that terraform handles such fields with replace.

x-kubernetes-preserve-unknown-fields was added, because CRD definition grows extremely large.
And it's not possible to use all CRDs as bundle. There is 1 MB limit for kubernetes configmaps.

Probably, there is some work arounds for terraform kubernetes provider case?

[f41gh7]

hashicorp/terraform-provider-kubernetes#1928

[hatemosphere]

Probably, there is some work arounds for terraform kubernetes provider case?

nah, there are no workarounds, as far as we know