Add opslevel_integration_kubernetes resource

[saditya370]

Add opslevel_integration_kubernetes resource

Lets users manage a Kubernetes integration's ETL config as code (GitLab issue 14148):

resource "opslevel_integration_kubernetes" "this" {
  name                 = "Kubernetes Integration"
  extract_definition   = file("extract.yaml")   # optional
  transform_definition = file("transform.yaml") # optional
}

Bumps opslevel-go to v2026.7.21 for the new client methods.

Notes for reviewers

• The API parses and re-serializes the YAML definitions, so the returned string never matches
  the configured one byte-for-byte, and the server fills in default definitions when none are
  given. Both fields are therefore Optional + Computed, and state keeps the configured value
  (falling back to the API value on import or when unset). Trade-off: terraform plan won't
  catch ETL edits made through the UI. That's documented on the attributes — a semantic-YAML
  plan modifier could restore drift detection later if we want it.
• The extractDefinition / transformDefinition arguments sit behind the k8s_etl_flow
  feature flag (default-on for new accounts); older accounts without it get a 402.

Testing

• tests/remote/integration_kubernetes.tftest.hcl against a dev instance: create with both
  definitions, update both in place, destroy — all green.
• go build, unit tests, golangci-lint, gofumpt, terraform fmt all clean; docs generated
  with tfplugindocs.

[andrewstillv15]

Can we update the opslevel-go submodule too?

[saditya370]

Addressed all three comments in d09e7ad:

• extract/transform now live under a single optional etl_definition object with both members required, so the provider always sends the pair together and can't hit the clear-one-clears-both behavior.
• Definitions are compared by unmarshalled value on refresh — formatting normalization no longer churns state, but edits made through the UI now show up as drift.

Re-ran the acceptance test against a dev instance: create with both definitions, in-place update, destroy — all green.

[saditya370]

Submodule bumped to v2026.7.21 as well.

[andrewstillv15]

Suggested change

	// IntegrationKubernetesEtlDefinitionModel describes the extract/transform definition pair.
	// The API manages the two definitions as a unit (clearing one clears both), so they are
	// modeled as a single object.

[andrewstillv15]

Suggested change

	// yamlEquivalent reports whether two YAML documents carry the same data. The API parses and
	// re-serializes stored definitions, so equivalent documents can differ in formatting.

[andrewstillv15]

I think this should go in helpers.go

[andrewstillv15]

Suggested change

	"github.com/hashicorp/terraform-plugin-framework/resource/schema/planmodifier"
	"github.com/hashicorp/terraform-plugin-framework/resource/schema/objectdefault"
	"github.com/hashicorp/terraform-plugin-framework/resource/schema/planmodifier"

Needed for the Default added below.

[andrewstillv15]

Suggested change

	"etl_definition": schema.SingleNestedAttribute{
	Description: "The ETL definitions used to import data from the integration. If not set, OpsLevel's default definitions are used. The API manages the two definitions as a unit, so both must be set together.",
	Optional: true,
	Computed: true,
	Attributes: map[string]schema.Attribute{
	"extract_definition": schema.StringAttribute{
	Description: "The YAML definition for extracting data from inbound payloads.",
	Required: true,
	},
	"transform_definition": schema.StringAttribute{
	Description: "The YAML definition for transforming extracted data to OpsLevel resources.",
	Required: true,
	},
	},
	},
	"etl_definition": schema.SingleNestedAttribute{
	Description: "The ETL definitions used to import data from the integration. If not set (or removed), OpsLevel's default definitions are used. The API manages the two definitions as a unit, so both must be set together.",
	Optional: true,
	Computed: true,
	Attributes: map[string]schema.Attribute{
	"extract_definition": schema.StringAttribute{
	Description: "The YAML definition for extracting data from inbound payloads.",
	Required: true,
	},
	"transform_definition": schema.StringAttribute{
	Description: "The YAML definition for transforming extracted data to OpsLevel resources.",
	Required: true,
	},
	},
	Default: objectdefault.StaticValue(types.ObjectValueMust(
	integrationKubernetesEtlDefinitionAttrs(),
	map[string]attr.Value{
	"extract_definition": types.StringValue(""),
	"transform_definition": types.StringValue(""),
	},
	)),
	},

Without this, an omitted etl_definition resolves to Unknown on every plan (no default, no plan modifier), so it never converges.

[andrewstillv15]

Suggested change

	// newKubernetesIntegrationInput only sets the definition fields when the etl_definition object
	// has a known value. Unknown values (unset in config) must be omitted so the API keeps its
	// defaults - sending an explicit null would clear both definitions on the server.
	func newKubernetesIntegrationInput(ctx context.Context, planModel IntegrationKubernetesResourceModel, diags *diag.Diagnostics) opslevel.KubernetesIntegrationInput {
	input := opslevel.KubernetesIntegrationInput{
	Name: nullable(planModel.Name.ValueStringPointer()),
	}
	if !planModel.EtlDefinition.IsNull() && !planModel.EtlDefinition.IsUnknown() {
	var etlModel IntegrationKubernetesEtlDefinitionModel
	diags.Append(planModel.EtlDefinition.As(ctx, &etlModel, basetypes.ObjectAsOptions{UnhandledNullAsEmpty: true, UnhandledUnknownAsEmpty: true})...)
	input.ExtractDefinition = refOf(opslevel.YAML(etlModel.ExtractDefinition.ValueString()))
	input.TransformDefinition = refOf(opslevel.YAML(etlModel.TransformDefinition.ValueString()))
	}
	func newKubernetesIntegrationInput(ctx context.Context, planModel IntegrationKubernetesResourceModel, diags *diag.Diagnostics) opslevel.KubernetesIntegrationInput {
	input := opslevel.KubernetesIntegrationInput{
	Name: nullable(planModel.Name.ValueStringPointer()),
	}
	if !planModel.EtlDefinition.IsNull() && !planModel.EtlDefinition.IsUnknown() {
	var etlModel IntegrationKubernetesEtlDefinitionModel
	diags.Append(planModel.EtlDefinition.As(ctx, &etlModel, basetypes.ObjectAsOptions{UnhandledNullAsEmpty: true, UnhandledUnknownAsEmpty: true})...)
	if etlModel.ExtractDefinition.ValueString() != "" || etlModel.TransformDefinition.ValueString() != "" {
	input.ExtractDefinition = refOf(opslevel.YAML(etlModel.ExtractDefinition.ValueString()))
	input.TransformDefinition = refOf(opslevel.YAML(etlModel.TransformDefinition.ValueString()))
	}
	}

Sending extractDefinition: "" / transformDefinition: "" explicitly is rejected with a GraphQL coercion error (Could not coerce value "" to YAML), so always sending both fields can't work - omitting both together is the only way to trigger the server's reset-to-default behavior.

[saditya370]

all six suggestions are in.
The import, the Default on etl_definition, and the omit-both-when-empty guard went in as you wrote them. yamlEquivalent now lives in helpers.go, and I dropped both doc comments.
One thing I added on top: the model now keeps the planned {"", ""} sentinel on refresh instead of adopting whatever defaults the server sends back. Without that the unset apply blows up with "Provider produced inconsistent result after apply", since the plan already promised empty strings. Happy to move that logic elsewhere if you had somewhere better in mind.

I also added an unset run to the acceptance test - set the definitions, null them out, set them again. Confirms the id stays put and both fields land on "". 3/3 green against a dev instance.