feat: Implment basic api layers for model deployment gql types

Author: seedspirit

docs/manager/graphql-reference/supergraph.graphql

@@ -2991,11 +2991,11 @@ type ModelDeployment implements Node
   metadata: ModelDeploymentMetadata!
   networkAccess: ModelDeploymentNetworkAccess!
   revision: ModelRevision
-  revisionHistory: ModelRevisionConnection!
-  scalingRule: ScalingRule!
-  replicaState: ReplicaState!
   defaultDeploymentStrategy: DeploymentStrategy!
   createdUser: UserNode!
+  scalingRule: ScalingRule!
+  replicaState: ReplicaState!
+  revisionHistory: ModelRevisionConnection!
 }
 
 """Added in 25.13.0"""
@@ -3028,10 +3028,10 @@ type ModelDeploymentMetadata
   name: String!
   status: DeploymentStatus!
   tags: [String!]!
-  project: GroupNode!
-  domain: DomainNode!
   createdAt: DateTime!
   updatedAt: DateTime!
+  project: GroupNode!
+  domain: DomainNode!
 }
 
 """Added in 25.13.0"""
@@ -3066,9 +3066,9 @@ input ModelDeploymentNetworkAccessInput
 type ModelMountConfig
   @join__type(graph: STRAWBERRY)
 {
-  vfolder: VirtualFolderNode!
   mountDestination: String!
   definitionPath: String!
+  vfolder: VirtualFolderNode!
 }
 
 """Added in 25.13.0"""
@@ -3087,7 +3087,6 @@ type ModelReplica implements Node
 {
   """The Globally Unique ID of this object"""
   id: ID!
-  revision: ModelRevision!
 
   """
   This represents whether the replica has been checked and its health state.
@@ -3120,6 +3119,7 @@ type ModelReplica implements Node
   The session ID associated with the replica. This can be null right after replica creation.
   """
   session: ComputeSessionNode!
+  revision: ModelRevision!
 }
 
 """Added in 25.13.0"""
@@ -3158,8 +3158,8 @@ type ModelRevision implements Node
   modelRuntimeConfig: ModelRuntimeConfig!
   modelMountConfig: ModelMountConfig!
   extraMounts: ExtraVFolderMountConnection!
-  image: ImageNode!
   createdAt: DateTime!
+  image: ImageNode!
 }
 
 """Added in 25.13.0"""
@@ -4084,7 +4084,9 @@ type Mutation
   """Added in 25.13.0"""
   addModelRevision(input: AddModelRevisionInput!): AddModelRevisionPayload! @join__field(graph: STRAWBERRY)
 
-  """Added in 25.13.0"""
+  """
+  Added in 25.13.0. Create model revision which is not attached to any deployment.
+  """
   createModelRevision(input: CreateModelRevisionInput!): CreateModelRevisionPayload! @join__field(graph: STRAWBERRY)
 
   """Added in 25.14.0"""
@@ -4838,7 +4840,7 @@ type Query
   revisions(filter: ModelRevisionFilter = null, orderBy: [ModelRevisionOrderBy!] = null, before: String = null, after: String = null, first: Int = null, last: Int = null, limit: Int = null, offset: Int = null): ModelRevisionConnection! @join__field(graph: STRAWBERRY)
 
   """Added in 25.13.0"""
-  revision(id: ID!): ModelRevision @join__field(graph: STRAWBERRY)
+  revision(id: ID!): ModelRevision! @join__field(graph: STRAWBERRY)
 
   """Added in 25.13.0"""
   replicas(filter: ReplicaFilter = null, orderBy: [ReplicaOrderBy!] = null, before: String = null, after: String = null, first: Int = null, last: Int = null, limit: Int = null, offset: Int = null): ModelReplicaConnection! @join__field(graph: STRAWBERRY)
@@ -5044,8 +5046,6 @@ type ReservoirRegistryEdge
 type ResourceConfig
   @join__type(graph: STRAWBERRY)
 {
-  resourceGroup: ScalingGroupNode!
-
   """
   Resource Slots are a JSON string that describes the resources allocated for the deployment. Example: "resourceSlots": "{\"cpu\": \"1\", \"mem\": \"1073741824\", \"cuda.device\": \"0\"}"
   """
@@ -5055,6 +5055,7 @@ type ResourceConfig
   Resource Options are a JSON string that describes additional options for the resources. This is especially used for shared memory configurations. Example: "resourceOpts": "{\"shmem\": \"64m\"}"
   """
   resourceOpts: JSONString
+  resourceGroup: ScalingGroupNode!
 }
 
 """Added in 25.13.0"""

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

@@ -670,7 +670,7 @@ The `JSON` scalar type represents JSON values as specified by [ECMA-404](https:/
 """
 scalar JSON @specifiedBy(url: "https://ecma-international.org/wp-content/uploads/ECMA-404_2nd_edition_december_2017.pdf")
 
-"""A custom scalar for JSON strings using orjson"""
+"""Added in 25.13.0"""
 scalar JSONString
 
 """
@@ -696,11 +696,11 @@ type ModelDeployment implements Node {
   metadata: ModelDeploymentMetadata!
   networkAccess: ModelDeploymentNetworkAccess!
   revision: ModelRevision
-  revisionHistory: ModelRevisionConnection!
-  scalingRule: ScalingRule!
-  replicaState: ReplicaState!
   defaultDeploymentStrategy: DeploymentStrategy!
   createdUser: UserNode!
+  scalingRule: ScalingRule!
+  replicaState: ReplicaState!
+  revisionHistory: ModelRevisionConnection!
 }
 
 """Added in 25.13.0"""
@@ -727,10 +727,10 @@ type ModelDeploymentMetadata {
   name: String!
   status: DeploymentStatus!
   tags: [String!]!
-  project: GroupNode!
-  domain: DomainNode!
   createdAt: DateTime!
   updatedAt: DateTime!
+  project: GroupNode!
+  domain: DomainNode!
 }
 
 """Added in 25.13.0"""
@@ -757,9 +757,9 @@ input ModelDeploymentNetworkAccessInput {
 
 """Added in 25.13.0"""
 type ModelMountConfig {
-  vfolder: VirtualFolderNode!
   mountDestination: String!
   definitionPath: String!
+  vfolder: VirtualFolderNode!
 }
 
 """Added in 25.13.0"""
@@ -773,7 +773,6 @@ input ModelMountConfigInput {
 type ModelReplica implements Node {
   """The Globally Unique ID of this object"""
   id: ID!
-  revision: ModelRevision!
 
   """
   This represents whether the replica has been checked and its health state.
@@ -806,6 +805,7 @@ type ModelReplica implements Node {
   The session ID associated with the replica. This can be null right after replica creation.
   """
   session: ComputeSessionNode!
+  revision: ModelRevision!
 }
 
 """Added in 25.13.0"""
@@ -837,8 +837,8 @@ type ModelRevision implements Node {
   modelRuntimeConfig: ModelRuntimeConfig!
   modelMountConfig: ModelMountConfig!
   extraMounts: ExtraVFolderMountConnection!
-  image: ImageNode!
   createdAt: DateTime!
+  image: ImageNode!
 }
 
 """Added in 25.13.0"""
@@ -946,7 +946,9 @@ type Mutation {
   """Added in 25.13.0"""
   addModelRevision(input: AddModelRevisionInput!): AddModelRevisionPayload!
 
-  """Added in 25.13.0"""
+  """
+  Added in 25.13.0. Create model revision which is not attached to any deployment.
+  """
   createModelRevision(input: CreateModelRevisionInput!): CreateModelRevisionPayload!
 
   """Added in 25.14.0"""
@@ -1118,7 +1120,7 @@ type Query {
   revisions(filter: ModelRevisionFilter = null, orderBy: [ModelRevisionOrderBy!] = null, before: String = null, after: String = null, first: Int = null, last: Int = null, limit: Int = null, offset: Int = null): ModelRevisionConnection!
 
   """Added in 25.13.0"""
-  revision(id: ID!): ModelRevision
+  revision(id: ID!): ModelRevision!
 
   """Added in 25.13.0"""
   replicas(filter: ReplicaFilter = null, orderBy: [ReplicaOrderBy!] = null, before: String = null, after: String = null, first: Int = null, last: Int = null, limit: Int = null, offset: Int = null): ModelReplicaConnection!
@@ -1254,8 +1256,6 @@ type ReservoirRegistryEdge {
 
 """Added in 25.13.0"""
 type ResourceConfig {
-  resourceGroup: ScalingGroupNode!
-
   """
   Resource Slots are a JSON string that describes the resources allocated for the deployment. Example: "resourceSlots": "{\"cpu\": \"1\", \"mem\": \"1073741824\", \"cuda.device\": \"0\"}"
   """
@@ -1265,6 +1265,7 @@ type ResourceConfig {
   Resource Options are a JSON string that describes additional options for the resources. This is especially used for shared memory configurations. Example: "resourceOpts": "{\"shmem\": \"64m\"}"
   """
   resourceOpts: JSONString
+  resourceGroup: ScalingGroupNode!
 }
 
 """Added in 25.13.0"""

src/ai/backend/common/exception.py

@@ -165,6 +165,7 @@ class ErrorDomain(enum.StrEnum):
     PERMISSION = "permission"
     METRIC = "metric"
     STORAGE_PROXY = "storage-proxy"
+    MODEL_DEPLOYMENT = "model-deployment"
 
 
 class ErrorOperation(enum.StrEnum):
@@ -643,3 +644,16 @@ def error_code(cls) -> ErrorCode:
             operation=ErrorOperation.READ,
             error_detail=ErrorDetail.NOT_FOUND,
         )
+
+
+class ModelDeploymentUnavailableError(BackendAIError, web.HTTPServiceUnavailable):
+    error_type = "https://api.backend.ai/probs/model-deployment-unavailable"
+    error_title = "Model Deployment Unavailable"
+
+    @classmethod
+    def error_code(cls) -> ErrorCode:
+        return ErrorCode(
+            domain=ErrorDomain.MODEL_DEPLOYMENT,
+            operation=ErrorOperation.EXECUTE,
+            error_detail=ErrorDetail.UNAVAILABLE,
+        )

src/ai/backend/manager/api/admin.py

@@ -23,7 +23,9 @@
 from ai.backend.common.dto.manager.response import GraphQLResponse
 from ai.backend.logging import BraceStyleAdapter
 from ai.backend.manager.api.gql.types import StrawberryGQLContext
-from ai.backend.manager.dto.context import ProcessorsCtx
+from ai.backend.manager.dto.context import (
+    ProcessorsCtx,
+)
 
 from ..api.gql.schema import schema as strawberry_schema
 from ..errors.api import GraphQLError as BackendGQLError

src/ai/backend/manager/api/gql/base.py

@@ -1,15 +1,18 @@
 from __future__ import annotations
 
 import uuid
+from collections.abc import Mapping
 from enum import StrEnum
-from typing import TYPE_CHECKING, Any, Optional, Type
+from typing import TYPE_CHECKING, Any, Optional, Type, cast
 
 import orjson
 import strawberry
 from graphql import StringValueNode
 from graphql_relay.utils import base64, unbase64
 from strawberry.types import get_object_definition, has_object_definition
 
+from ai.backend.common.types import ResourceSlot
+
 if TYPE_CHECKING:
     from ai.backend.manager.types import (
         PaginationOptions,
@@ -133,33 +136,28 @@ class Ordering(StrEnum):
     DESC_NULLS_LAST = "DESC_NULLS_LAST"
 
 
-def serialize_json(value: Any) -> str:
-    if isinstance(value, (dict, list)):
-        return orjson.dumps(value).decode("utf-8")
-    elif isinstance(value, str):
-        return value
-    else:
-        return orjson.dumps(value).decode("utf-8")
-
-
-def parse_json(value: str | bytes) -> Any:
-    if isinstance(value, str):
-        return orjson.loads(value)
-    elif isinstance(value, bytes):
-        return orjson.loads(value)
-    else:
+@strawberry.scalar(description="Added in 25.13.0")
+class JSONString:
+    @staticmethod
+    def parse_value(value: str | bytes) -> Mapping[str, Any]:
+        if isinstance(value, str):
+            return orjson.loads(value)
+        if isinstance(value, bytes):
+            return orjson.loads(value)
         return value
 
+    @staticmethod
+    def serialize(value: Any) -> JSONString:
+        if isinstance(value, (dict, list)):
+            return cast(JSONString, orjson.dumps(value).decode("utf-8"))
+        elif isinstance(value, str):
+            return cast(JSONString, value)
+        else:
+            return cast(JSONString, orjson.dumps(value).decode("utf-8"))
 
-@strawberry.scalar(
-    name="JSONString",
-    description="A custom scalar for JSON strings using orjson",
-    serialize=serialize_json,
-    parse_value=parse_json,
-    parse_literal=lambda v: parse_json(v.value) if hasattr(v, "value") else v,
-)
-class JSONString:
-    pass
+    @staticmethod
+    def from_resource_slot(resource_slot: ResourceSlot) -> JSONString:
+        return JSONString.serialize(resource_slot.to_json())
 
 
 def to_global_id(type_: Type[Any], local_id: uuid.UUID | str) -> str: