feat(BA-5373): Add blue-green deployment infrastructure and promote API

changes/10426.feature.md

@@ -0,0 +1 @@
+Add blue-green deployment infrastructure and promote API

docs/manager/graphql-reference/supergraph.graphql

@@ -10207,6 +10207,11 @@ type Mutation
   """
   activateDeploymentRevision(input: ActivateRevisionInput!): ActivateRevisionPayload! @join__field(graph: STRAWBERRY)
 
+  """
+  Added in UNRELEASED. Manually promote a blue-green deployment awaiting promotion. When `auto_promote=false`, the deployment stays in AWAITING_PROMOTION once all Green routes are healthy until an operator triggers this mutation, which atomically switches traffic (Green→ACTIVE, Blue→INACTIVE→TERMINATING).
+  """
+  promoteDeployment(input: PromoteDeploymentInput!): PromoteDeploymentPayload! @join__field(graph: STRAWBERRY)
+
   """Added in 25.19.0. Update the traffic status of a route"""
   updateRouteTrafficStatus(input: UpdateRouteTrafficStatusInput!): UpdateRouteTrafficStatusPayload! @join__field(graph: STRAWBERRY)
 
@@ -12140,6 +12145,23 @@ input ProjectWeightInputItem
   weight: Decimal = null
 }
 
+"""
+Added in UNRELEASED. Input for manually promoting a blue-green deployment.
+"""
+input PromoteDeploymentInput
+  @join__type(graph: STRAWBERRY)
+{
+  deploymentId: ID!
+}
+
+"""Added in UNRELEASED. Result of manually promoting a deployment."""
+type PromoteDeploymentPayload
+  @join__type(graph: STRAWBERRY)
+{
+  """The promoted deployment"""
+  deployment: ModelDeployment!
+}
+
 """
 Completely delete domain from DB.
 

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

@@ -6209,6 +6209,11 @@ type Mutation {
   """
   activateDeploymentRevision(input: ActivateRevisionInput!): ActivateRevisionPayload!
 
+  """
+  Added in UNRELEASED. Manually promote a blue-green deployment awaiting promotion. When `auto_promote=false`, the deployment stays in AWAITING_PROMOTION once all Green routes are healthy until an operator triggers this mutation, which atomically switches traffic (Green→ACTIVE, Blue→INACTIVE→TERMINATING).
+  """
+  promoteDeployment(input: PromoteDeploymentInput!): PromoteDeploymentPayload!
+
   """Added in 25.19.0. Update the traffic status of a route"""
   updateRouteTrafficStatus(input: UpdateRouteTrafficStatusInput!): UpdateRouteTrafficStatusPayload!
 
@@ -7849,6 +7854,19 @@ input ProjectWeightInputItem {
   weight: Decimal = null
 }
 
+"""
+Added in UNRELEASED. Input for manually promoting a blue-green deployment.
+"""
+input PromoteDeploymentInput {
+  deploymentId: ID!
+}
+
+"""Added in UNRELEASED. Result of manually promoting a deployment."""
+type PromoteDeploymentPayload {
+  """The promoted deployment"""
+  deployment: ModelDeployment!
+}
+
 """Added in 26.4.2. Payload for domain permanent deletion mutation."""
 type PurgeDomainPayloadGQL {
   """Whether the purge was successful."""

src/ai/backend/client/cli/v2/deployment/commands.py

@@ -226,6 +226,28 @@ async def _run() -> None:
     asyncio.run(_run())
 
 
+@deployment.command()
+@click.argument("deployment_id", type=str)
+def promote(deployment_id: str) -> None:
+    """Manually promote a blue-green deployment awaiting promotion."""
+
+    from ai.backend.common.dto.manager.v2.deployment.request import (
+        PromoteDeploymentInput,
+    )
+
+    body = PromoteDeploymentInput(deployment_id=UUID(deployment_id))
+
+    async def _run() -> None:
+        registry = await create_v2_registry(load_v2_config())
+        try:
+            result = await registry.deployment.promote(body)
+            print_result(result)
+        finally:
+            await registry.close()
+
+    asyncio.run(_run())
+
+
 @deployment.command()
 @click.argument("deployment_id", type=str)
 def delete(deployment_id: str) -> None:

src/ai/backend/client/v2/domains_v2/deployment.py

@@ -21,6 +21,7 @@
     CreateDeploymentInput,
     DeleteAccessTokenInput,
     DeleteDeploymentInput,
+    PromoteDeploymentInput,
     SearchAccessTokensInput,
     SearchAutoScalingRulesInput,
     SearchDeploymentPoliciesInput,
@@ -48,6 +49,7 @@
     GetAccessTokenPayload,
     GetAutoScalingRulePayload,
     GetDeploymentPolicyPayload,
+    PromoteDeploymentPayload,
     ReplicaNode,
     RevisionNode,
     SearchAccessTokensPayload,
@@ -233,6 +235,17 @@ async def activate_revision(
             response_model=ActivateRevisionPayload,
         )
 
+    async def promote(
+        self,
+        body: PromoteDeploymentInput,
+    ) -> PromoteDeploymentPayload:
+        """Manually promote a blue-green deployment awaiting promotion."""
+        return await self._client.typed_request(
+            "POST",
+            _PATH + f"/{body.deployment_id}/promote",
+            response_model=PromoteDeploymentPayload,
+        )
+
     # ------------------------------------------------------------------
     # Replica operations
     # ------------------------------------------------------------------

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

@@ -850,6 +850,12 @@ class ActivateRevisionInput(BaseRequestModel):
     revision_id: UUID = Field(description="Revision ID to activate")
 
 
+class PromoteDeploymentInput(BaseRequestModel):
+    """Input for manually promoting a blue-green deployment."""
+
+    deployment_id: UUID = Field(description="Deployment ID to promote")
+
+
 class UpdateRouteTrafficStatusInput(BaseRequestModel):
     """Input for updating a route's traffic status."""
 

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

@@ -397,6 +397,12 @@ class ActivateRevisionPayload(BaseResponseModel):
     )
 
 
+class PromoteDeploymentPayload(BaseResponseModel):
+    """Payload for promote deployment mutation result."""
+
+    deployment: DeploymentNode = Field(description="The promoted deployment")
+
+
 class UpdateRouteTrafficStatusPayload(BaseResponseModel):
     """Payload for update route traffic status mutation result."""
 

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

@@ -38,6 +38,7 @@
     DeleteAccessTokenInput,
     DeleteDeploymentInput,
     DeploymentOrder,
+    PromoteDeploymentInput,
     ReplicaOrder,
     RevisionOrder,
     RouteOrder,
@@ -70,6 +71,7 @@
     GetAccessTokenPayload,
     GetAutoScalingRulePayload,
     GetDeploymentPolicyPayload,
+    PromoteDeploymentPayload,
     ReplicaNode,
     RevisionNode,
     RouteNode,
@@ -276,6 +278,7 @@
 )
 from ai.backend.manager.services.deployment.actions.revision_operations import (
     ActivateRevisionAction,
+    PromoteDeploymentAction,
 )
 from ai.backend.manager.services.deployment.actions.route.search_routes import SearchRoutesAction
 from ai.backend.manager.services.deployment.actions.route.update_route_traffic_status import (
@@ -814,6 +817,15 @@ async def activate_revision(self, input: ActivateRevisionInput) -> ActivateRevis
             deployment_policy=self._policy_data_to_dto(action_result.deployment_policy),
         )
 
+    async def promote_deployment(self, input: PromoteDeploymentInput) -> PromoteDeploymentPayload:
+        """Manually promote a blue-green deployment."""
+        action_result = await self._processors.deployment.promote_deployment.wait_for_complete(
+            PromoteDeploymentAction(deployment_id=input.deployment_id)
+        )
+        return PromoteDeploymentPayload(
+            deployment=self._deployment_data_to_dto(action_result.deployment),
+        )
+
     async def delete(self, input: DeleteDeploymentInput) -> DeleteDeploymentPayload:
         """Delete a deployment."""
         await self._processors.deployment.destroy_deployment.wait_for_complete(

src/ai/backend/manager/api/gql/deployment/__init__.py

@@ -38,6 +38,7 @@
     inference_runtime_configs,
     my_deployments,
     project_deployments,
+    promote_deployment,
     # Replica
     replica,
     replica_status_changed,
@@ -126,6 +127,8 @@
     ModelRuntimeConfigInput,
     MountPermission,
     ProjectDeploymentScopeGQL,
+    PromoteDeploymentInputGQL,
+    PromoteDeploymentPayloadGQL,
     ReadinessStatus,
     ReplicaFilter,
     ReplicaOrderBy,
@@ -236,6 +239,8 @@
     # Revision Types
     "ActivateRevisionInputGQL",
     "ActivateRevisionPayloadGQL",
+    "PromoteDeploymentInputGQL",
+    "PromoteDeploymentPayloadGQL",
     "AddRevisionInput",
     "AddRevisionPayload",
     "ClusterConfig",
@@ -295,6 +300,7 @@
     # Resolvers - Revision
     "activate_deployment_revision",
     "add_model_revision",
+    "promote_deployment",
     "inference_runtime_config",
     "inference_runtime_configs",
     "revision",

src/ai/backend/manager/api/gql/deployment/resolver/__init__.py

@@ -37,6 +37,7 @@
     add_model_revision,
     inference_runtime_config,
     inference_runtime_configs,
+    promote_deployment,
     revision,
     revisions,
 )
@@ -84,6 +85,7 @@
     "inference_runtime_configs",
     "add_model_revision",
     "activate_deployment_revision",
+    "promote_deployment",
     # Route
     "routes",
     "route",

src/ai/backend/manager/api/gql/deployment/resolver/revision.py

@@ -15,6 +15,7 @@
 from ai.backend.common.dto.manager.v2.deployment.request import (
     AdminSearchRevisionsInput,
 )
+from ai.backend.common.meta.meta import NEXT_RELEASE_VERSION
 from ai.backend.manager.api.gql.base import encode_cursor, resolve_global_id
 from ai.backend.manager.api.gql.decorators import (
     BackendAIGQLMeta,
@@ -34,6 +35,8 @@
     ModelRevisionEdge,
     ModelRevisionFilter,
     ModelRevisionOrderBy,
+    PromoteDeploymentInputGQL,
+    PromoteDeploymentPayloadGQL,
 )
 from ai.backend.manager.api.gql.types import StrawberryGQLContext
 from ai.backend.manager.data.deployment.inference_runtime_config import (
@@ -181,3 +184,25 @@ async def activate_deployment_revision(
         activated_revision_id=ID(str(payload.activated_revision_id)),
         deployment_policy=DeploymentPolicyGQL.from_pydantic(payload.deployment_policy),
     )
+
+
+@gql_mutation(
+    BackendAIGQLMeta(
+        added_version=NEXT_RELEASE_VERSION,
+        description=(
+            "Manually promote a blue-green deployment awaiting promotion. "
+            "When `auto_promote=false`, the deployment stays in AWAITING_PROMOTION "
+            "once all Green routes are healthy until an operator triggers this mutation, "
+            "which atomically switches traffic (Green→ACTIVE, Blue→INACTIVE→TERMINATING)."
+        ),
+    )
+)  # type: ignore[misc]
+async def promote_deployment(
+    input: PromoteDeploymentInputGQL,
+    info: Info[StrawberryGQLContext, None],
+) -> PromoteDeploymentPayloadGQL:
+    """Promote a deployment that is in AWAITING_PROMOTION state."""
+    payload = await info.context.adapters.deployment.promote_deployment(input.to_pydantic())
+    return PromoteDeploymentPayloadGQL(
+        deployment=ModelDeployment.from_pydantic(payload.deployment),
+    )

src/ai/backend/manager/api/gql/deployment/types/__init__.py

@@ -105,6 +105,8 @@
     ModelRuntimeConfig,
     ModelRuntimeConfigInput,
     MountPermission,
+    PromoteDeploymentInputGQL,
+    PromoteDeploymentPayloadGQL,
     ResourceConfig,
     ResourceConfigInput,
     ResourceGroupInput,
@@ -214,6 +216,8 @@
     # Revision
     "ActivateRevisionInputGQL",
     "ActivateRevisionPayloadGQL",
+    "PromoteDeploymentInputGQL",
+    "PromoteDeploymentPayloadGQL",
     "AddRevisionInput",
     "AddRevisionPayload",
     "ClusterConfig",

src/ai/backend/manager/api/gql/deployment/types/revision.py

@@ -62,6 +62,9 @@
 from ai.backend.common.dto.manager.v2.deployment.request import (
     ModelRuntimeConfigInput as ModelRuntimeConfigInputDTO,
 )
+from ai.backend.common.dto.manager.v2.deployment.request import (
+    PromoteDeploymentInput as PromoteDeploymentInputDTO,
+)
 from ai.backend.common.dto.manager.v2.deployment.request import (
     ResourceConfigInput as ResourceConfigInputDTO,
 )
@@ -83,6 +86,9 @@
 from ai.backend.common.dto.manager.v2.deployment.response import (
     AddRevisionPayload as AddRevisionPayloadDTO,
 )
+from ai.backend.common.dto.manager.v2.deployment.response import (
+    PromoteDeploymentPayload as PromoteDeploymentPayloadDTO,
+)
 from ai.backend.common.dto.manager.v2.deployment.response import (
     RevisionNode as RevisionNodeDTO,
 )
@@ -101,6 +107,7 @@
     PreStartActionInfoDTO,
     ResourceConfigInfoDTO,
 )
+from ai.backend.common.meta.meta import NEXT_RELEASE_VERSION
 from ai.backend.common.types import MountPermission as CommonMountPermission
 from ai.backend.manager.api.gql.base import (
     OrderDirection,
@@ -576,6 +583,29 @@ class ActivateRevisionPayloadGQL:
     deployment_policy: Annotated[DeploymentPolicyGQL, strawberry.lazy(".policy")]
 
 
+@gql_pydantic_input(
+    BackendAIGQLMeta(
+        description="Input for manually promoting a blue-green deployment.",
+        added_version=NEXT_RELEASE_VERSION,
+    ),
+    name="PromoteDeploymentInput",
+)
+class PromoteDeploymentInputGQL(PydanticInputMixin[PromoteDeploymentInputDTO]):
+    deployment_id: ID
+
+
+@gql_pydantic_type(
+    BackendAIGQLMeta(
+        added_version=NEXT_RELEASE_VERSION,
+        description="Result of manually promoting a deployment.",
+    ),
+    model=PromoteDeploymentPayloadDTO,
+    name="PromoteDeploymentPayload",
+)
+class PromoteDeploymentPayloadGQL:
+    deployment: Annotated[ModelDeployment, strawberry.lazy(".deployment")]
+
+
 # Input Types
 @gql_pydantic_input(
     BackendAIGQLMeta(description="", added_version="25.19.0"),

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

@@ -81,6 +81,7 @@
     inference_runtime_configs,
     my_deployments,
     project_deployments,
+    promote_deployment,
     # Replica
     replica,
     replica_status_changed,
@@ -704,6 +705,7 @@ class Mutation:
     reject_artifact_revision = reject_artifact_revision
     create_access_token = create_access_token
     activate_deployment_revision = activate_deployment_revision
+    promote_deployment = promote_deployment
     update_route_traffic_status = update_route_traffic_status
     # Fair Share - Admin APIs
     admin_upsert_domain_fair_share_weight = admin_upsert_domain_fair_share_weight