* Add ephemeral_storage_local_ssd_config support to modules/gke-nodepool
Adds ephemeral_storage_local_ssd_count to node_config variable and the
corresponding dynamic ephemeral_storage_local_ssd_config block in the
node pool resource, enabling use of local SSDs as ephemeral storage.
* feat(gke-nodepool): add flex_start support to node_config
Add `flex_start` as an optional bool to the `node_config` variable type
and wire it through to the `google_container_node_pool` resource's
node_config block. This enables DWS (Dynamic Workload Scheduler)
flex-start mode for node pools, used for on-demand capacity access
without requiring ProvisioningRequest objects (e.g. spot TPU pools).
* feat(gke-nodepool): add flex_start support to node_config
Add `flex_start` as an optional bool to the `node_config` variable type
and wire it through to the `google_container_node_pool` resource's
node_config block. This enables DWS (Dynamic Workload Scheduler)
flex-start mode for node pools, which allows the Cluster Autoscaler to
request capacity on-demand without requiring ProvisioningRequest objects
(unlike queued_provisioning). Typical use case is spot TPU node pools.
* feat(gke-nodepool): add advanced_machine_features support to node_config
Add `advanced_machine_features` as an optional object to the `node_config`
variable type and wire it through to the `google_container_node_pool`
resource via a dynamic block. This allows callers to configure
`threads_per_core` (e.g. set to 1 to disable hyperthreading) and
`enable_nested_virtualization` for node pools that require fine-grained
CPU threading control or nested hypervisor support.
GKE auto-sets `advanced_machine_features` (threads_per_core=1) on
ct6e/TPU machine types; exposing this field also lets consumers add it to
ignore_changes in their own lifecycle blocks to avoid forced replacements.
* feat(gke-nodepool): add containerd_config support to node_config
Add `containerd_config` as an optional object to the `node_config` variable
and wire it through to the `google_container_node_pool` resource via a
dynamic block. This allows callers to configure private registry mirrors or
custom containerd registry hosts per node pool — useful for air-gapped
environments and internal registry proxies.
The `registry_hosts` list maps each upstream server to one or more mirror
hosts, with optional `capabilities`, `override_path`, and `dial_timeout`
fields (all defaulting to sensible values).
* refactor(gke-nodepool): use maps for containerd_config registry_hosts and hosts
Convert registry_hosts and hosts from lists to maps so that the registry
server and host URLs serve as stable keys, avoiding index-shifting issues
with for_each. Add default values for capabilities, override_path, and
dial_timeout. Update README example and test inventory accordingly.
* Remove default values from containerd_config hosts fields
Leave capabilities, override_path, and dial_timeout without defaults
so the provider/API picks them rather than the module imposing values.
* Refine containerd_config variable interface
- Simplify header to optional(map(list(string)))
- Flatten ca, client cert/key to strings with descriptive names
- Derive private_registry_access_config enabled from ca domain config list
- Simplify writable_cgroups to optional(bool)
- Flatten gcp_secret_manager_certificate_config to string
- Remove redundant defaults where try() handles null in main.tf
- Fix long lines in main.tf to stay within 79-char limit
- Update copyright year to 2026 in inventory files
* fix(gke-nodepool): run terraform fmt to fix attribute alignment in containerd_config
* docs(gke-nodepool): regenerate README with updated variable line numbers
* fix(gke-nodepool): use coalesce instead of try for null header map in for_each
* tests(gke-nodepool): update containerd-config inventory to match actual plan output
---------
Co-authored-by: Julio Castillo <jccb@google.com>
