feat: remove core.telemetry as a dependency of llama_stack.apis (#4064)

# What does this PR do?

Remove circular dependency by moving tracing from API protocol
definitions
 to router implementation layer.

This gets us closer to having a self contained API package with no other
cross-cutting dependencies to other parts of the llama stack codebase.
To the best of our ability, the llama_stack.api should only be type and
protocol definitions.

  Changes:
- Create apis/common/tracing.py with marker decorator (zero core
dependencies)
- Add the _new_ `@telemetry_traceable` marker decorator to 11 protocol
classes
- Apply actual tracing in core/resolver.py in `instantiate_provider`
based on protocol marker
- Move MetricResponseMixin from core to apis (it's an API response type)
  - APIs package is now self-contained with zero core dependencies

The tracing functionality remains identical - actual trace_protocol from
core
is applied to router implementations at runtime when both telemetry is
enabled
  and the protocol has the `__marked_for_tracing__` marker.

  ## Test Plan

  Manual integration test confirms identical behavior to main branch:

  ```bash
  llama stack list-deps --format uv starter | sh
  export OLLAMA_URL=http://localhost:11434
  llama stack run starter

  curl -X POST http://localhost:8321/v1/chat/completions \
    -H "Content-Type: application/json" \
    -d '{"model": "ollama/gpt-oss:20b",
         "messages": [{"role": "user", "content": "Say hello"}],
         "max_tokens": 10}'
         
```

  Verified identical between main and this branch:
  - trace_id present in response
  - metrics array with prompt_tokens, completion_tokens, total_tokens
  - Server logs show trace_protocol applied to all routers

  Existing telemetry integration tests (tests/integration/telemetry/) validate
  trace context propagation and span attributes.


relates to #3895

---------

Signed-off-by: Charlie Doern <cdoern@redhat.com>

src/llama_stack/apis/common/responses.py

@@ -34,3 +34,44 @@ class PaginatedResponse(BaseModel):
     data: list[dict[str, Any]]
     has_more: bool
     url: str | None = None
+
+
+# This is a short term solution to allow inference API to return metrics
+# The ideal way to do this is to have a way for all response types to include metrics
+# and all metric events logged to the telemetry API to be included with the response
+# To do this, we will need to augment all response types with a metrics field.
+# We have hit a blocker from stainless SDK that prevents us from doing this.
+# The blocker is that if we were to augment the response types that have a data field
+# in them like so
+# class ListModelsResponse(BaseModel):
+# metrics: Optional[List[MetricEvent]] = None
+# data: List[Models]
+# ...
+# The client SDK will need to access the data by using a .data field, which is not
+# ergonomic. Stainless SDK does support unwrapping the response type, but it
+# requires that the response type to only have a single field.
+
+# We will need a way in the client SDK to signal that the metrics are needed
+# and if they are needed, the client SDK has to return the full response type
+# without unwrapping it.
+
+
+@json_schema_type
+class MetricInResponse(BaseModel):
+    """A metric value included in API responses.
+    :param metric: The name of the metric
+    :param value: The numeric value of the metric
+    :param unit: (Optional) The unit of measurement for the metric value
+    """
+
+    metric: str
+    value: int | float
+    unit: str | None = None
+
+
+class MetricResponseMixin(BaseModel):
+    """Mixin class for API responses that can include metrics.
+    :param metrics: (Optional) List of metrics associated with the API response
+    """
+
+    metrics: list[MetricInResponse] | None = None

src/llama_stack/apis/common/tracing.py

@@ -0,0 +1,22 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+
+def telemetry_traceable(cls):
+    """
+    Mark a protocol for automatic tracing when telemetry is enabled.
+
+    This is a metadata-only decorator with no dependencies on core.
+    Actual tracing is applied by core routers at runtime if telemetry is enabled.
+
+    Usage:
+        @runtime_checkable
+        @telemetry_traceable
+        class MyProtocol(Protocol):
+            ...
+    """
+    cls.__marked_for_tracing__ = True
+    return cls

src/llama_stack/apis/conversations/conversations.py

@@ -20,8 +20,8 @@ from llama_stack.apis.agents.openai_responses import (
     OpenAIResponseOutputMessageMCPListTools,
     OpenAIResponseOutputMessageWebSearchToolCall,
 )
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, register_schema, webmethod
 
 Metadata = dict[str, str]
@@ -157,7 +157,7 @@ class ConversationItemDeletedResource(BaseModel):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class Conversations(Protocol):
     """Conversations
 

src/llama_stack/apis/files/files.py

@@ -11,8 +11,8 @@ from fastapi import File, Form, Response, UploadFile
 from pydantic import BaseModel, Field
 
 from llama_stack.apis.common.responses import Order
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, webmethod
 
 
@@ -102,7 +102,7 @@ class OpenAIFileDeleteResponse(BaseModel):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class Files(Protocol):
     """Files
 

src/llama_stack/apis/inference/inference.py

@@ -19,11 +19,10 @@ from pydantic import BaseModel, Field, field_validator
 from typing_extensions import TypedDict
 
 from llama_stack.apis.common.content_types import ContentDelta, InterleavedContent
-from llama_stack.apis.common.responses import Order
+from llama_stack.apis.common.responses import MetricResponseMixin, Order
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.models import Model
 from llama_stack.apis.version import LLAMA_STACK_API_V1, LLAMA_STACK_API_V1ALPHA
-from llama_stack.core.telemetry.telemetry import MetricResponseMixin
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.models.llama.datatypes import (
     BuiltinTool,
     StopReason,
@@ -1160,7 +1159,7 @@ class OpenAIEmbeddingsRequestWithExtraBody(BaseModel, extra="allow"):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class InferenceProvider(Protocol):
     """
     This protocol defines the interface that should be implemented by all inference providers.

src/llama_stack/apis/models/models.py

@@ -9,9 +9,9 @@ from typing import Any, Literal, Protocol, runtime_checkable
 
 from pydantic import BaseModel, ConfigDict, Field, field_validator
 
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.resource import Resource, ResourceType
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, webmethod
 
 
@@ -105,7 +105,7 @@ class OpenAIListModelsResponse(BaseModel):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class Models(Protocol):
     async def list_models(self) -> ListModelsResponse:
         """List all models.

src/llama_stack/apis/prompts/prompts.py

@@ -10,8 +10,8 @@ from typing import Protocol, runtime_checkable
 
 from pydantic import BaseModel, Field, field_validator, model_validator
 
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, webmethod
 
 
@@ -92,7 +92,7 @@ class ListPromptsResponse(BaseModel):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class Prompts(Protocol):
     """Prompts
 

src/llama_stack/apis/safety/safety.py

@@ -9,10 +9,10 @@ from typing import Any, Protocol, runtime_checkable
 
 from pydantic import BaseModel, Field
 
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.inference import OpenAIMessageParam
 from llama_stack.apis.shields import Shield
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, webmethod
 
 
@@ -94,7 +94,7 @@ class ShieldStore(Protocol):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class Safety(Protocol):
     """Safety
 

src/llama_stack/apis/shields/shields.py

@@ -8,9 +8,9 @@ from typing import Any, Literal, Protocol, runtime_checkable
 
 from pydantic import BaseModel
 
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.resource import Resource, ResourceType
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, webmethod
 
 
@@ -48,7 +48,7 @@ class ListShieldsResponse(BaseModel):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class Shields(Protocol):
     @webmethod(route="/shields", method="GET", level=LLAMA_STACK_API_V1)
     async def list_shields(self) -> ListShieldsResponse:

src/llama_stack/apis/tools/tools.py

@@ -11,9 +11,9 @@ from pydantic import BaseModel
 from typing_extensions import runtime_checkable
 
 from llama_stack.apis.common.content_types import URL, InterleavedContent
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.resource import Resource, ResourceType
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, webmethod
 
 
@@ -107,7 +107,7 @@ class ListToolDefsResponse(BaseModel):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class ToolGroups(Protocol):
     @webmethod(route="/toolgroups", method="POST", level=LLAMA_STACK_API_V1)
     async def register_tool_group(
@@ -189,7 +189,7 @@ class SpecialToolGroup(Enum):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class ToolRuntime(Protocol):
     tool_store: ToolStore | None = None
 

src/llama_stack/apis/vector_io/vector_io.py

@@ -13,10 +13,10 @@ from typing import Annotated, Any, Literal, Protocol, runtime_checkable
 from fastapi import Body
 from pydantic import BaseModel, Field
 
+from llama_stack.apis.common.tracing import telemetry_traceable
 from llama_stack.apis.inference import InterleavedContent
 from llama_stack.apis.vector_stores import VectorStore
 from llama_stack.apis.version import LLAMA_STACK_API_V1
-from llama_stack.core.telemetry.trace_protocol import trace_protocol
 from llama_stack.schema_utils import json_schema_type, webmethod
 from llama_stack.strong_typing.schema import register_schema
 
@@ -502,7 +502,7 @@ class VectorStoreTable(Protocol):
 
 
 @runtime_checkable
-@trace_protocol
+@telemetry_traceable
 class VectorIO(Protocol):
     vector_store_table: VectorStoreTable | None = None