Publish curated ITWorx-ModelForge source from 990401737f96c8d46a9f7b0d84994516cadcb6a0

backend/src/modelforge_api/domain/serving.py

@@ -8,7 +8,7 @@ import uuid
 from datetime import datetime
 from typing import Any, Literal
 
-from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
+from pydantic import AwareDatetime, BaseModel, ConfigDict, Field, field_validator, model_validator
 
 CAPABILITY_VERSION_KEY = re.compile(r"^[a-z][a-z0-9_]*(?:\.[a-z][a-z0-9_]*)+@[1-9][0-9]*$")
 ServingOperation = Literal["load", "invoke", "health", "drain", "unload"]
@@ -119,6 +119,7 @@ class ResidencyResponse(ServingModel):
 
 class CapabilityDeploymentResponse(ServingModel):
     id: uuid.UUID
+    route_generation: int = Field(ge=0)
     capability: str
     contract_version: int
     deployment_candidate_id: uuid.UUID
@@ -159,7 +160,8 @@ class ServiceClientCreate(ServingModel):
     project_key: str | None = Field(default=None, pattern=r"^[a-z][a-z0-9-]*$")
     integration_environment: Literal["production", "shadow", "evaluation", "lab"] | None = None
     purpose: str | None = Field(default=None, min_length=3, max_length=500)
-    credential_expires_at: datetime | None = None
+    credential_expires_at: AwareDatetime | None = None
+    project_binding_id: uuid.UUID | None = None
 
     @field_validator("allowed_capabilities")
     @classmethod
@@ -176,9 +178,13 @@ class ServiceClientCreate(ServingModel):
         if any(value is not None for value in project_fields) and not all(
             value is not None for value in project_fields
         ):
-            raise ValueError("project_key, integration_environment and purpose are required together")
+            raise ValueError(
+                "project_key, integration_environment and purpose are required together"
+            )
         if self.project_key is not None and len(self.allowed_capabilities) != 1:
             raise ValueError("a project service identity must bind exactly one capability")
+        if self.project_binding_id is not None and self.project_key is None:
+            raise ValueError("an explicit binding requires a project identity")
         return self
 
 
@@ -195,6 +201,7 @@ class ServiceClientResponse(ServingModel):
     integration_environment: str | None
     purpose: str | None
     credential_prefix: str | None
+    credential_id: uuid.UUID | None = None
     credential_created_at: datetime | None
     credential_expires_at: datetime | None
     credential_revoked_at: datetime | None
@@ -206,20 +213,35 @@ class ServiceClientCreated(ServiceClientResponse):
     credential: str
 
 
+class ServiceCredentialRotate(ServingModel):
+    expected_credential_id: uuid.UUID
+    overlap_seconds: int = Field(default=0, strict=True, ge=0, le=86400)
+    expires_at: AwareDatetime
+
+
+class ServiceCredentialRevoke(ServingModel):
+    expected_credential_id: uuid.UUID
+    confirmation: str = Field(min_length=1, max_length=255)
+
+
+class ServiceCredentialResponse(ServingModel):
+    id: uuid.UUID
+    secret_prefix: str
+    created_at: datetime
+    expires_at: datetime | None
+    superseded_at: datetime | None
+    revoked_at: datetime | None
+    last_used_at: datetime | None
+
+
 class ProjectFitEvidenceCreate(ServingModel):
     project_key: str = Field(pattern=r"^[a-z][a-z0-9-]*$")
     capability: str = Field(pattern=r"^[a-z][a-z0-9.]*@[1-9][0-9]*$")
     environment: Literal["production", "shadow", "evaluation", "lab"]
-    recommendation: Literal[
-        "PROMOTION_ELIGIBLE", "KEEP_LAB", "BLOCKED", "REQUIRES_MORE_EVIDENCE"
-    ]
-    evidence_class: Literal[
-        "OWNER_PHOTO", "PUBLIC_PHYSICAL_CAPTURE", "CATALOG_REFERENCE"
-    ]
+    recommendation: Literal["PROMOTION_ELIGIBLE", "KEEP_LAB", "BLOCKED", "REQUIRES_MORE_EVIDENCE"]
+    evidence_class: Literal["OWNER_PHOTO", "PUBLIC_PHYSICAL_CAPTURE", "CATALOG_REFERENCE"]
     engineering_integration: Literal["PASS", "INCOMPLETE", "BLOCKED"]
-    production_validation: Literal[
-        "DEFERRED_EXTERNAL_VALIDATION", "REQUIRED", "SATISFIED"
-    ]
+    production_validation: Literal["DEFERRED_EXTERNAL_VALIDATION", "REQUIRED", "SATISFIED"]
     production_action: Literal["NONE"] = "NONE"
     case_count: int = Field(ge=0, le=1_000_000)
     metric_values: dict[str, float | int | str | bool | None]
@@ -693,6 +715,7 @@ class AgentServingJobLease(ServingModel):
     top_n: int | None = Field(default=None, ge=1, le=10)
     modality_payload: dict[str, Any] | None = None
     request_id: uuid.UUID | None = None
+    lab_run_id: uuid.UUID | None = None
 
     @model_validator(mode="after")
     def validate_capability_payload(self) -> AgentServingJobLease:
@@ -736,6 +759,20 @@ class AgentServingJobFailure(ServingModel):
     details: dict[str, Any] = Field(default_factory=dict)
 
 
+class AgentServingJobCancellationCheck(ServingModel):
+    lease_token: str = Field(min_length=32, max_length=512)
+
+
+class AgentServingJobCancellationStatus(ServingModel):
+    cancel_requested: bool
+
+
+class AgentServingJobCancellationAck(ServingModel):
+    lease_token: str = Field(min_length=32, max_length=512)
+    worker_instance_id: str = Field(min_length=1, max_length=255)
+    state: Literal["cold", "warm", "busy", "loading", "draining", "unloading", "failed"]
+
+
 class AgentResidentState(ServingModel):
     deployment_id: uuid.UUID
     artifact_set_id: uuid.UUID