Add validation and translation for served_models and served_entities in serving endpoints (#3880)

## Changes && Why
`served_models` was deprecated and the fields in it were moved to
`served_endpoints`. The GET API response for a model serving endpoint
returns with both `served_models` and `served_entities` even if one is
set on the client side (not AIP complaint).

In order to simplify the direct deployment logic, this PR translates the
`served_models` config to `served_entities`, so that we can keep the
core diff logic in direct deployment simple.

Note: This is not a breaking change. TF also had the same logic with a
conflicts_with annotation for `served_models` and `served_entities`


## Tests
New acceptance tests. Manually confirmed that a deployment succeeds with
no issues.

Author: shreyas-goenka

NEXT_CHANGELOG.md

@@ -9,5 +9,6 @@
 ### Dependency updates
 
 ### Bundles
+* Add validation that served_models and served_entities are not used at the same time. Add client side translation logic. ([#3880](https://github.com/databricks/cli/pull/3880))
 
 ### API Changes

acceptance/bundle/validate/model_serving_both_fields_error/databricks.yml

@@ -0,0 +1,14 @@
+bundle:
+  name: "model_serving_error"
+
+resources:
+  model_serving_endpoints:
+    endpoint1:
+      name: "test-endpoint"
+      config:
+        served_models:
+          - model_name: "my-model"
+            model_version: "1"
+        served_entities:
+          - entity_name: "my-entity"
+            entity_version: "2"

acceptance/bundle/validate/model_serving_both_fields_error/out.test.toml

@@ -0,0 +1,14 @@
+Error: Cannot use both served_models and served_entities
+  in databricks.yml:7:7
+
+Model serving endpoint cannot specify both served_models and served_entities at the same time.
+
+Name: model_serving_error
+Target: default
+Workspace:
+  User: [USERNAME]
+  Path: /Workspace/Users/[USERNAME]/.bundle/model_serving_error/default
+
+Found 1 error
+
+Exit code: 1

acceptance/bundle/validate/model_serving_both_fields_error/output.txt

@@ -0,0 +1 @@
+$CLI bundle validate

acceptance/bundle/validate/model_serving_both_fields_error/script

@@ -0,0 +1,12 @@
+bundle:
+  name: "model_serving_conversion"
+
+resources:
+  model_serving_endpoints:
+    endpoint1:
+      name: "test-endpoint"
+      config:
+        served_models:
+          - model_name: "my-model"
+            model_version: "1"
+            scale_to_zero_enabled: true

acceptance/bundle/validate/model_serving_conversion/databricks.yml

@@ -0,0 +1,14 @@
+Warning: Using served_models is deprecated
+  in databricks.yml:10:11
+
+The served_models field is deprecated. Please use served_entities instead.
+
+{
+  "served_entities": [
+    {
+      "entity_name": "my-model",
+      "entity_version": "1",
+      "scale_to_zero_enabled": true
+    }
+  ]
+}

acceptance/bundle/validate/model_serving_conversion/out.test.toml

@@ -0,0 +1 @@
+$CLI bundle validate -o json | jq .resources.model_serving_endpoints.endpoint1.config

acceptance/bundle/validate/model_serving_conversion/output.txt

@@ -0,0 +1,89 @@
+package resourcemutator
+
+import (
+	"context"
+
+	"github.com/databricks/cli/bundle"
+	"github.com/databricks/cli/libs/diag"
+	"github.com/databricks/cli/libs/dyn"
+	"github.com/databricks/cli/libs/utils"
+	"github.com/databricks/databricks-sdk-go/service/serving"
+)
+
+type modelServingEndpointFixups struct{}
+
+func ModelServingEndpointFixups() bundle.Mutator {
+	return &modelServingEndpointFixups{}
+}
+
+func (m *modelServingEndpointFixups) Name() string {
+	return "ModelServingEndpointFixups"
+}
+
+func servedModelToServedEntity(model serving.ServedModelInput) serving.ServedEntityInput {
+	return serving.ServedEntityInput{
+		// served_models does not support ExternalModel, so we set ExternalModel to nil
+		ExternalModel: nil,
+
+		EntityName:                model.ModelName,
+		EntityVersion:             model.ModelVersion,
+		EnvironmentVars:           model.EnvironmentVars,
+		InstanceProfileArn:        model.InstanceProfileArn,
+		MaxProvisionedThroughput:  model.MaxProvisionedThroughput,
+		MinProvisionedThroughput:  model.MinProvisionedThroughput,
+		Name:                      model.Name,
+		ProvisionedModelUnits:     model.ProvisionedModelUnits,
+		ScaleToZeroEnabled:        model.ScaleToZeroEnabled,
+		WorkloadSize:              model.WorkloadSize,
+		WorkloadType:              serving.ServingModelWorkloadType(model.WorkloadType),
+		MaxProvisionedConcurrency: model.MaxProvisionedConcurrency,
+		MinProvisionedConcurrency: model.MinProvisionedConcurrency,
+		ForceSendFields:           utils.FilterFields[serving.ServedEntityInput](model.ForceSendFields),
+	}
+}
+
+func (m *modelServingEndpointFixups) Apply(ctx context.Context, b *bundle.Bundle) diag.Diagnostics {
+	var diags diag.Diagnostics
+
+	for key, endpoint := range b.Config.Resources.ModelServingEndpoints {
+		if endpoint == nil || endpoint.Config == nil {
+			continue
+		}
+
+		// Validate that both ServedModels and ServedEntities are not used at the same time.
+		if len(endpoint.Config.ServedModels) > 0 && len(endpoint.Config.ServedEntities) > 0 {
+			diags = diags.Append(diag.Diagnostic{
+				Severity: diag.Error,
+				Summary:  "Cannot use both served_models and served_entities",
+				Detail:   "Model serving endpoint cannot specify both served_models and served_entities at the same time.",
+				Locations: []dyn.Location{
+					b.Config.GetLocation("resources.model_serving_endpoints." + key),
+				},
+			})
+			continue
+		}
+
+		// Convert ServedModels to ServedEntities if specified. ServedModels is a deprecated field, and the service recommends using ServedEntities instead.
+		// We perform this translation here so that the deployment plan only has to detect served_entities and can ignore served_models.
+		if len(endpoint.Config.ServedModels) > 0 {
+			// Add warning recommending served_entities
+			diags = diags.Append(diag.Diagnostic{
+				Severity: diag.Warning,
+				Summary:  "Using served_models is deprecated",
+				Detail:   "The served_models field is deprecated. Please use served_entities instead.",
+				Locations: []dyn.Location{
+					b.Config.GetLocation("resources.model_serving_endpoints." + key + ".config.served_models"),
+				},
+			})
+
+			endpoint.Config.ServedEntities = make([]serving.ServedEntityInput, len(endpoint.Config.ServedModels))
+			for i, model := range endpoint.Config.ServedModels {
+				endpoint.Config.ServedEntities[i] = servedModelToServedEntity(model)
+			}
+			// Clear ServedModels after conversion
+			endpoint.Config.ServedModels = nil
+		}
+	}
+
+	return diags
+}