Support for GKE clusters with no default node pools

[ciroclover]

TL;DR

Currently, the private-cluster module forces the creation of a default node pool even when node_pools = [] is specified. This prevents us to manage the node pools externally with dedicated node-pool modules.

Problem statement

Currently, the private-cluster module forces the creation of a default node pool even when node_pools = [] is specified. This prevents us to manage the node pools externally with dedicated node-pool modules.

The issue happens because:

1. The node_pools variables defaults to [{name = "default-node-pool"}] [ref]
2. The node_pool block in the cluster resource is not conditional [ref]
3. When node_pools = [] is set, the module still tries to access var.node_pool[0] causing index errors [ref]

Detailed design:

1. Update variables.tf:

variable "node_pools" {
  default = []      # changed from {name = "default-node-pool"}
 ...

2. Create a conditional to the node_pool block:

dynamic "node_pool" {
  for_each = length(var.node_pools) > 0 ? [1] : []
  content {
    name                  = "default-pool"
    initial_node_count    = var.initial_node_count
  # ... rest of the config
  }
}

3. Handle an edge case in the main.tf [ref] :

...
  cluster_name_parts_from_nodepool           = length(values(google_container_node_pool.pools)) > 0 ? split("/", element(values(google_container_node_pool.pools)[*].id, 0)) : []
  cluster_name_computed                      = length(local.cluster_name_parts_from_nodepool) > 0 ? element(local.cluster_name_parts_from_nodepool, length(local.cluster_name_parts_from_nodepool) - 3)

...

I have tested the code above, and I was able to create a cluster with no default node pool.

Would the maintainers be open to a pull request implementing this feature? I'm happy to submit a clean PR with the changes, and update the current documentation.

Additional information

No response

Edit

The solution above is not backward compatible.

The correct way to implement is to not change the variable default, and use the conditional:

for_each = length(var.node_pools) == 0 ? [] : [1]

In the pool-node block.

[m0ps]

I was about to start working on a PR to implement the same feature, as we plan to migrate completely to custom compute classes and avoid manually managed node pools. I've noticed that it's already implemented in gke-standard-cluster submodule. So it just needed to port this logic to other submodules.

[ciroclover]

Raised the PR feat: Support for GKE private clusters without default node pool #2408.

I edited this issue because the first solution I provided is not backward compatible.
If the PR gets merged, we just need to set var.node_pools and a clean cluster with no node-pools will be provisioned.