refactor(BA-4330): Scope PR to OTel infrastructure only

Remove GQL resolver span changes (schema.py) and unrelated AgentRow
ORM fix (cli/__main__.py) to keep this PR focused on OTel trace
propagation infrastructure. GQL spans will be a separate PR.

Update the TODO comment on the manual middleware injection workaround
to accurately describe the coupling with aiohttp's Application lifecycle.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: hhoikoo

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

@@ -15,8 +15,6 @@
 from graphene.types.inputobjecttype import set_input_object_type_default_value
 from graphql import GraphQLError, OperationType, Undefined
 from graphql.type import GraphQLField
-from opentelemetry import trace
-from opentelemetry.trace import StatusCode
 from sqlalchemy.orm import joinedload, selectinload
 
 from ai.backend.common.clients.valkey_client.valkey_image.client import ValkeyImageClient
@@ -3360,23 +3358,11 @@ def resolve(
         operation_name = (
             info.operation.name.value if info.operation.name is not None else "anonymous"
         )
-        tracer = trace.get_tracer(__name__)
-        span = tracer.start_span(
-            f"gql.{operation_name}.{field_name}",
-            attributes={
-                "graphql.operation_name": operation_name,
-                "graphql.field_name": field_name,
-                "graphql.parent_type": parent_type,
-            },
-        )
 
         start = time.perf_counter()
         try:
             res = next(root, info, **args)
         except BackendAIError as e:
-            span.record_exception(e)
-            span.set_status(StatusCode.ERROR, str(e))
-            span.end()
             graph_ctx.metric_observer.observe_request(
                 operation_type=operation_type,
                 field_name=field_name,
@@ -3387,10 +3373,7 @@ def resolve(
                 duration=time.perf_counter() - start,
             )
             raise
-        except BaseException as e:
-            span.record_exception(e)
-            span.set_status(StatusCode.ERROR)
-            span.end()
+        except BaseException:
             graph_ctx.metric_observer.observe_request(
                 operation_type=operation_type,
                 field_name=field_name,
@@ -3406,52 +3389,43 @@ def resolve(
 
             async def _timed_coro() -> Any:
                 coro_start = time.perf_counter()
-                with trace.use_span(span, end_on_exit=False):
-                    try:
-                        result = await res
-                    except BackendAIError as e:
-                        span.record_exception(e)
-                        span.set_status(StatusCode.ERROR, str(e))
-                        graph_ctx.metric_observer.observe_request(
-                            operation_type=operation_type,
-                            field_name=field_name,
-                            parent_type=parent_type,
-                            operation_name=operation_name,
-                            error_code=e.error_code(),
-                            success=False,
-                            duration=time.perf_counter() - coro_start,
-                        )
-                        raise
-                    except BaseException as e:
-                        span.record_exception(e)
-                        span.set_status(StatusCode.ERROR)
-                        graph_ctx.metric_observer.observe_request(
-                            operation_type=operation_type,
-                            field_name=field_name,
-                            parent_type=parent_type,
-                            operation_name=operation_name,
-                            error_code=ErrorCode.default(),
-                            success=False,
-                            duration=time.perf_counter() - coro_start,
-                        )
-                        raise
-                    else:
-                        graph_ctx.metric_observer.observe_request(
-                            operation_type=operation_type,
-                            field_name=field_name,
-                            parent_type=parent_type,
-                            operation_name=operation_name,
-                            error_code=None,
-                            success=True,
-                            duration=time.perf_counter() - coro_start,
-                        )
-                        return result
-                    finally:
-                        span.end()
+                try:
+                    result = await res
+                except BackendAIError as e:
+                    graph_ctx.metric_observer.observe_request(
+                        operation_type=operation_type,
+                        field_name=field_name,
+                        parent_type=parent_type,
+                        operation_name=operation_name,
+                        error_code=e.error_code(),
+                        success=False,
+                        duration=time.perf_counter() - coro_start,
+                    )
+                    raise
+                except BaseException:
+                    graph_ctx.metric_observer.observe_request(
+                        operation_type=operation_type,
+                        field_name=field_name,
+                        parent_type=parent_type,
+                        operation_name=operation_name,
+                        error_code=ErrorCode.default(),
+                        success=False,
+                        duration=time.perf_counter() - coro_start,
+                    )
+                    raise
+                graph_ctx.metric_observer.observe_request(
+                    operation_type=operation_type,
+                    field_name=field_name,
+                    parent_type=parent_type,
+                    operation_name=operation_name,
+                    error_code=None,
+                    success=True,
+                    duration=time.perf_counter() - coro_start,
+                )
+                return result
 
             return _timed_coro()
 
-        span.end()
         graph_ctx.metric_observer.observe_request(
             operation_type=operation_type,
             field_name=field_name,

src/ai/backend/manager/cli/__main__.py

@@ -265,17 +265,11 @@ def clear_history(cli_ctx: CLIContext, retention: str, vacuum_full: bool) -> Non
     from more_itertools import chunked
 
     from ai.backend.common.validators import TimeDuration
-    from ai.backend.manager.models.agent import AgentRow
     from ai.backend.manager.models.error_logs import error_logs
     from ai.backend.manager.models.kernel import kernels
     from ai.backend.manager.models.session import SessionRow
     from ai.backend.manager.models.utils import connect_database, vacuum_db
 
-    # AgentRow must be imported to register it with SQLAlchemy's ORM mapper.
-    # KernelRow (via SessionRow) has a relationship to AgentRow using a string reference,
-    # which requires AgentRow to be in the mapper registry.
-    _ = AgentRow
-
     from .context import redis_ctx
 
     log = _get_logger()

src/ai/backend/manager/server.py

@@ -1777,9 +1777,14 @@ async def webapp_ctx(root_app: web.Application) -> AsyncGenerator[None]:
         jwt_config = root_ctx.config_provider.config.jwt.to_jwt_config()
         root_ctx.jwt_validator = JWTValidator(jwt_config)
 
-        # Initialize OTel aiohttp instrumentation before the app is frozen.
-        # instrument_aiohttp_server() patches the Application class, but since
-        # root_app is already instantiated, we must manually inject the middleware.
+        # TODO(BA-4330): Remove this manual middleware injection once the manager's
+        # setup procedure is decoupled from the aiohttp Application lifecycle.
+        # instrument_aiohttp_server() patches the Application class via setattr, so it
+        # only affects instances created AFTER the call. Since root_app is already
+        # instantiated by build_root_app(), we must manually inject the middleware.
+        # A proper fix would restructure startup so that OTel instrumentation runs
+        # before Application creation (e.g., by moving OTel config to BootstrapConfig
+        # or by decoupling config loading from aiohttp cleanup contexts).
         if root_ctx.config_provider.config.otel.enabled:
             instrument_aiohttp_server()
             instrument_aiohttp_client()