fix(BA-5983): accept partial ModelDefinition input in deployment API

changes/11531.fix.md

@@ -0,0 +1 @@
+Make ModelConfig / ModelDefinition / ModelServiceConfig / ModelHealthCheck GraphQL input fields optional so addModelRevision can inherit values from the runtime variant, model-definition.yaml, or revision preset.

docs/manager/graphql-reference/supergraph.graphql

@@ -9235,10 +9235,10 @@ input ModelConfigInput
   @join__type(graph: STRAWBERRY)
 {
   """Name of the model."""
-  name: String!
+  name: String = null
 
   """Path to the model file."""
-  modelPath: String!
+  modelPath: String = null
 
   """Configuration for the model service."""
   service: ModelServiceConfigInput = null
@@ -9264,7 +9264,7 @@ input ModelDefinitionInput
   @join__type(graph: STRAWBERRY)
 {
   """List of models in the model definition."""
-  models: [ModelConfigInput!]!
+  models: [ModelConfigInput!] = null
 }
 
 """
@@ -9431,22 +9431,22 @@ input ModelHealthCheckInput
   @join__type(graph: STRAWBERRY)
 {
   """Interval in seconds between health checks."""
-  interval: Float! = 10
+  interval: Float = null
 
   """Path to check for health status."""
-  path: String!
+  path: String = null
 
   """Maximum number of retries for health check."""
-  maxRetries: Int! = 10
+  maxRetries: Int = null
 
   """Maximum time in seconds to wait for a health check response."""
-  maxWaitTime: Float! = 15
+  maxWaitTime: Float = null
 
   """Expected HTTP status code for a healthy response."""
-  expectedStatusCode: Int! = 200
+  expectedStatusCode: Int = null
 
   """Initial delay in seconds before the first health check."""
-  initialDelay: Float! = 60
+  initialDelay: Float = null
 }
 
 """Added in 26.4.2. Metadata describing a model entry."""
@@ -9803,16 +9803,16 @@ input ModelServiceConfigInput
   """
   List of pre-start actions to execute before starting the model service.
   """
-  preStartActions: [PreStartActionInput!]!
+  preStartActions: [PreStartActionInput!] = null
 
   """Command to start the model service."""
   startCommand: [String!] = null
 
   """Shell configured for the model service."""
-  shell: String! = "/bin/bash"
+  shell: String = null
 
-  """Port number for the model service. Must be greater than 1."""
-  port: Int!
+  """Port number for the model service."""
+  port: Int = null
 
   """Health check configuration for the model service."""
   healthCheck: ModelHealthCheckInput = null

docs/manager/graphql-reference/v2-schema.graphql

@@ -6045,10 +6045,10 @@ Added in 26.4.0. Configuration for a single model within a model definition.
 """
 input ModelConfigInput {
   """Name of the model."""
-  name: String!
+  name: String = null
 
   """Path to the model file."""
-  modelPath: String!
+  modelPath: String = null
 
   """Configuration for the model service."""
   service: ModelServiceConfigInput = null
@@ -6070,7 +6070,7 @@ Added in 26.4.0. Model definition containing a list of model configurations.
 """
 input ModelDefinitionInput {
   """List of models in the model definition."""
-  models: [ModelConfigInput!]!
+  models: [ModelConfigInput!] = null
 }
 
 """
@@ -6218,22 +6218,22 @@ type ModelHealthCheck {
 """Added in 26.4.0. Health check configuration for a model service."""
 input ModelHealthCheckInput {
   """Interval in seconds between health checks."""
-  interval: Float! = 10
+  interval: Float = null
 
   """Path to check for health status."""
-  path: String!
+  path: String = null
 
   """Maximum number of retries for health check."""
-  maxRetries: Int! = 10
+  maxRetries: Int = null
 
   """Maximum time in seconds to wait for a health check response."""
-  maxWaitTime: Float! = 15
+  maxWaitTime: Float = null
 
   """Expected HTTP status code for a healthy response."""
-  expectedStatusCode: Int! = 200
+  expectedStatusCode: Int = null
 
   """Initial delay in seconds before the first health check."""
-  initialDelay: Float! = 60
+  initialDelay: Float = null
 }
 
 """Added in 26.4.2. Metadata describing a model entry."""
@@ -6554,16 +6554,16 @@ input ModelServiceConfigInput {
   """
   List of pre-start actions to execute before starting the model service.
   """
-  preStartActions: [PreStartActionInput!]!
+  preStartActions: [PreStartActionInput!] = null
 
   """Command to start the model service."""
   startCommand: [String!] = null
 
   """Shell configured for the model service."""
-  shell: String! = "/bin/bash"
+  shell: String = null
 
-  """Port number for the model service. Must be greater than 1."""
-  port: Int!
+  """Port number for the model service."""
+  port: Int = null
 
   """Health check configuration for the model service."""
   healthCheck: ModelHealthCheckInput = null

src/ai/backend/common/dto/manager/v2/deployment/request.py

@@ -13,7 +13,10 @@
 from pydantic import Field, field_validator
 
 from ai.backend.common.api_handlers import SENTINEL, BaseRequestModel, Sentinel
-from ai.backend.common.config import ModelDefinitionDraft
+from ai.backend.common.config import (
+    ModelDefinitionDraft,
+    PreStartAction,
+)
 from ai.backend.common.data.model_deployment.types import (
     DeploymentStrategy,
     RouteHealthStatus,
@@ -80,10 +83,15 @@
     "EnvironmentVariablesInput",
     "ExtraVFolderMountInput",
     "ImageInput",
+    "ModelConfigInput",
+    "ModelDefinitionInput",
     "ModelDeploymentMetadataInput",
     "ModelDeploymentNetworkAccessInput",
+    "ModelHealthCheckInput",
+    "ModelMetadataInput",
     "ModelMountConfigInput",
     "ModelRuntimeConfigInput",
+    "ModelServiceConfigInput",
     "ReplicaFilter",
     "ReplicaOrder",
     "ReplicaStatusFilter",
@@ -116,6 +124,67 @@
 )
 
 
+class ModelHealthCheckInput(BaseRequestModel):
+    interval: float | None = None
+    path: str | None = None
+    max_retries: int | None = None
+    max_wait_time: float | None = None
+    expected_status_code: int | None = None
+    initial_delay: float | None = None
+
+
+class ModelMetadataInput(BaseRequestModel):
+    author: str | None = None
+    title: str | None = None
+    version: str | None = None
+    created: str | None = None
+    last_modified: str | None = None
+    description: str | None = None
+    task: str | None = None
+    category: str | None = None
+    architecture: str | None = None
+    framework: list[str] | None = None
+    label: list[str] | None = None
+    license: str | None = None
+    min_resource: dict[str, Any] | None = None
+
+
+class ModelServiceConfigInput(BaseRequestModel):
+    pre_start_actions: list[PreStartAction] | None = None
+    start_command: list[str] | None = None
+    shell: str | None = None
+    port: int | None = None
+    health_check: ModelHealthCheckInput | None = None
+
+
+class ModelConfigInput(BaseRequestModel):
+    name: str | None = None
+    model_path: str | None = None
+    service: ModelServiceConfigInput | None = None
+    metadata: ModelMetadataInput | None = None
+
+
+class ModelDefinitionInput(BaseRequestModel):
+    """All-optional v2 input mirror of :class:`ModelDefinitionDraft`.
+
+    Fields a request omits are filled by lower-priority sources in the
+    revision merge chain (runtime variant baseline, revision preset,
+    vfolder ``model-definition.yaml``, ``model_mount_destination``
+    default). Required-field enforcement happens later in
+    ``ModelDefinitionDraft.to_resolved`` after the merge.
+    """
+
+    models: list[ModelConfigInput] | None = None
+
+
+def to_model_definition_draft(
+    input: ModelDefinitionInput | None,
+) -> ModelDefinitionDraft | None:
+    if input is None:
+        return None
+    return ModelDefinitionDraft.model_validate(input.model_dump())
+
+
 class ClusterConfigInput(BaseRequestModel):
     """Cluster configuration input for a revision."""
 
@@ -233,7 +302,7 @@ class CreateRevisionInputDTO(BaseRequestModel):
     image: ImageInput = Field(description="Container image")
     model_runtime_config: ModelRuntimeConfigInput = Field(description="Runtime configuration")
     model_mount_config: ModelMountConfigInput = Field(description="Model mount configuration")
-    model_definition: ModelDefinitionDraft | None = Field(
+    model_definition: ModelDefinitionInput | None = Field(
         default=None,
         description="Model definition to override the default values generated by the server",
     )
@@ -269,7 +338,7 @@ class AddRevisionGQLInputDTO(BaseRequestModel):
     image: ImageInput = Field(description="Container image")
     model_runtime_config: ModelRuntimeConfigInput = Field(description="Runtime configuration")
     model_mount_config: ModelMountConfigInput = Field(description="Model mount configuration")
-    model_definition: ModelDefinitionDraft | None = Field(
+    model_definition: ModelDefinitionInput | None = Field(
         default=None,
         description="Model definition to override the default values generated by the server",
     )
@@ -396,7 +465,7 @@ class RevisionInput(BaseRequestModel):
         default="/models", description="Mount destination for model vfolder"
     )
     model_definition_path: str = Field(description="Path to model definition file")
-    model_definition: ModelDefinitionDraft | None = Field(
+    model_definition: ModelDefinitionInput | None = Field(
         default=None,
         description="Model definition to override the default values generated by the server",
     )

src/ai/backend/manager/api/adapters/deployment/adapter.py

@@ -62,6 +62,7 @@
     SyncReplicaInput,
     UpdateDeploymentInput,
     UpsertDeploymentPolicyInput,
+    to_model_definition_draft,
 )
 from ai.backend.common.dto.manager.v2.deployment.response import (
     AccessTokenNode,
@@ -505,7 +506,7 @@ async def create(
                     else None,
                 ),
                 mounts=mounts_creator,
-                model_definition=initial_revision.model_definition,
+                model_definition=to_model_definition_draft(initial_revision.model_definition),
                 revision_preset_id=initial_revision.revision_preset_id,
                 execution=ExecutionSpec(
                     runtime_variant_id=initial_revision.model_runtime_config.runtime_variant_id,
@@ -1110,7 +1111,7 @@ async def add_revision(
                 else None,
                 inference_runtime_config=input.model_runtime_config.inference_runtime_config,
             ),
-            model_definition=input.model_definition,
+            model_definition=to_model_definition_draft(input.model_definition),
             revision_preset_id=input.revision_preset_id,
         )
         action_result = await self._processors.deployment.add_model_revision.wait_for_complete(