Merge branch 'main' into patch-metadata

llama_stack/providers/inline/inference/meta_reference/inference.py

@@ -28,7 +28,7 @@ from llama_stack.apis.inference import (
     CompletionRequest,
     CompletionResponse,
     CompletionResponseStreamChunk,
-    Inference,
+    InferenceProvider,
     InterleavedContent,
     LogProbConfig,
     Message,
@@ -86,7 +86,7 @@ class MetaReferenceInferenceImpl(
     OpenAICompletionToLlamaStackMixin,
     OpenAIChatCompletionToLlamaStackMixin,
     SentenceTransformerEmbeddingMixin,
-    Inference,
+    InferenceProvider,
     ModelsProtocolPrivate,
 ):
     def __init__(self, config: MetaReferenceInferenceConfig) -> None:

llama_stack/providers/inline/inference/sentence_transformers/sentence_transformers.py

@@ -9,7 +9,7 @@ from collections.abc import AsyncGenerator
 
 from llama_stack.apis.inference import (
     CompletionResponse,
-    Inference,
+    InferenceProvider,
     InterleavedContent,
     LogProbConfig,
     Message,
@@ -38,7 +38,7 @@ class SentenceTransformersInferenceImpl(
     OpenAIChatCompletionToLlamaStackMixin,
     OpenAICompletionToLlamaStackMixin,
     SentenceTransformerEmbeddingMixin,
-    Inference,
+    InferenceProvider,
     ModelsProtocolPrivate,
 ):
     def __init__(self, config: SentenceTransformersInferenceConfig) -> None:

llama_stack/providers/inline/telemetry/meta_reference/telemetry.py

@@ -16,6 +16,7 @@ from opentelemetry.sdk.resources import Resource
 from opentelemetry.sdk.trace import TracerProvider
 from opentelemetry.sdk.trace.export import BatchSpanProcessor
 from opentelemetry.semconv.resource import ResourceAttributes
+from opentelemetry.trace.propagation.tracecontext import TraceContextTextMapPropagator
 
 from llama_stack.apis.telemetry import (
     Event,
@@ -44,6 +45,7 @@ from llama_stack.providers.inline.telemetry.meta_reference.sqlite_span_processor
 )
 from llama_stack.providers.utils.telemetry.dataset_mixin import TelemetryDatasetMixin
 from llama_stack.providers.utils.telemetry.sqlite_trace_store import SQLiteTraceStore
+from llama_stack.providers.utils.telemetry.tracing import ROOT_SPAN_MARKERS
 
 from .config import TelemetryConfig, TelemetrySink
 
@@ -206,6 +208,15 @@ class TelemetryAdapter(TelemetryDatasetMixin, Telemetry):
                 event.attributes = {}
             event.attributes["__ttl__"] = ttl_seconds
 
+            # Extract these W3C trace context attributes so they are not written to
+            # underlying storage, as we just need them to propagate the trace context.
+            traceparent = event.attributes.pop("traceparent", None)
+            tracestate = event.attributes.pop("tracestate", None)
+            if traceparent:
+                # If we have a traceparent header value, we're not the root span.
+                for root_attribute in ROOT_SPAN_MARKERS:
+                    event.attributes.pop(root_attribute, None)
+
             if isinstance(event.payload, SpanStartPayload):
                 # Check if span already exists to prevent duplicates
                 if span_id in _GLOBAL_STORAGE["active_spans"]:
@@ -216,8 +227,12 @@ class TelemetryAdapter(TelemetryDatasetMixin, Telemetry):
                     parent_span_id = int(event.payload.parent_span_id, 16)
                     parent_span = _GLOBAL_STORAGE["active_spans"].get(parent_span_id)
                     context = trace.set_span_in_context(parent_span)
-                else:
-                    event.attributes["__root_span__"] = "true"
+                elif traceparent:
+                    carrier = {
+                        "traceparent": traceparent,
+                        "tracestate": tracestate,
+                    }
+                    context = TraceContextTextMapPropagator().extract(carrier=carrier)
 
                 span = tracer.start_span(
                     name=event.payload.name,

llama_stack/providers/remote/inference/cerebras_openai_compat/__init__.py

@@ -4,12 +4,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from llama_stack.apis.inference import Inference
+from llama_stack.apis.inference import InferenceProvider
 
 from .config import CerebrasCompatConfig
 
 
-async def get_adapter_impl(config: CerebrasCompatConfig, _deps) -> Inference:
+async def get_adapter_impl(config: CerebrasCompatConfig, _deps) -> InferenceProvider:
     # import dynamically so the import is used only when it is needed
     from .cerebras import CerebrasCompatInferenceAdapter
 

llama_stack/providers/remote/inference/fireworks_openai_compat/__init__.py

@@ -4,12 +4,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from llama_stack.apis.inference import Inference
+from llama_stack.apis.inference import InferenceProvider
 
 from .config import FireworksCompatConfig
 
 
-async def get_adapter_impl(config: FireworksCompatConfig, _deps) -> Inference:
+async def get_adapter_impl(config: FireworksCompatConfig, _deps) -> InferenceProvider:
     # import dynamically so the import is used only when it is needed
     from .fireworks import FireworksCompatInferenceAdapter
 

llama_stack/providers/remote/inference/groq_openai_compat/__init__.py

@@ -4,12 +4,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from llama_stack.apis.inference import Inference
+from llama_stack.apis.inference import InferenceProvider
 
 from .config import GroqCompatConfig
 
 
-async def get_adapter_impl(config: GroqCompatConfig, _deps) -> Inference:
+async def get_adapter_impl(config: GroqCompatConfig, _deps) -> InferenceProvider:
     # import dynamically so the import is used only when it is needed
     from .groq import GroqCompatInferenceAdapter
 

llama_stack/providers/remote/inference/llama_openai_compat/__init__.py

@@ -4,12 +4,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from llama_stack.apis.inference import Inference
+from llama_stack.apis.inference import InferenceProvider
 
 from .config import LlamaCompatConfig
 
 
-async def get_adapter_impl(config: LlamaCompatConfig, _deps) -> Inference:
+async def get_adapter_impl(config: LlamaCompatConfig, _deps) -> InferenceProvider:
     # import dynamically so the import is used only when it is needed
     from .llama import LlamaCompatInferenceAdapter
 

llama_stack/providers/remote/inference/ollama/ollama.py

@@ -28,7 +28,7 @@ from llama_stack.apis.inference import (
     EmbeddingsResponse,
     EmbeddingTaskType,
     GrammarResponseFormat,
-    Inference,
+    InferenceProvider,
     JsonSchemaResponseFormat,
     LogProbConfig,
     Message,
@@ -82,7 +82,7 @@ logger = get_logger(name=__name__, category="inference")
 
 
 class OllamaInferenceAdapter(
-    Inference,
+    InferenceProvider,
     ModelsProtocolPrivate,
 ):
     def __init__(self, url: str) -> None:

llama_stack/providers/remote/inference/sambanova_openai_compat/__init__.py

@@ -4,12 +4,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from llama_stack.apis.inference import Inference
+from llama_stack.apis.inference import InferenceProvider
 
 from .config import SambaNovaCompatConfig
 
 
-async def get_adapter_impl(config: SambaNovaCompatConfig, _deps) -> Inference:
+async def get_adapter_impl(config: SambaNovaCompatConfig, _deps) -> InferenceProvider:
     # import dynamically so the import is used only when it is needed
     from .sambanova import SambaNovaCompatInferenceAdapter
 

llama_stack/providers/remote/inference/together_openai_compat/__init__.py

@@ -4,12 +4,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from llama_stack.apis.inference import Inference
+from llama_stack.apis.inference import InferenceProvider
 
 from .config import TogetherCompatConfig
 
 
-async def get_adapter_impl(config: TogetherCompatConfig, _deps) -> Inference:
+async def get_adapter_impl(config: TogetherCompatConfig, _deps) -> InferenceProvider:
     # import dynamically so the import is used only when it is needed
     from .together import TogetherCompatInferenceAdapter
 

llama_stack/providers/utils/inference/litellm_openai_mixin.py

@@ -19,7 +19,7 @@ from llama_stack.apis.inference import (
     ChatCompletionResponseStreamChunk,
     EmbeddingsResponse,
     EmbeddingTaskType,
-    Inference,
+    InferenceProvider,
     JsonSchemaResponseFormat,
     LogProbConfig,
     Message,
@@ -59,7 +59,7 @@ logger = get_logger(name=__name__, category="inference")
 
 class LiteLLMOpenAIMixin(
     ModelRegistryHelper,
-    Inference,
+    InferenceProvider,
     NeedsRequestProviderData,
 ):
     # TODO: avoid exposing the litellm specific model names to the user.

llama_stack/providers/utils/telemetry/tracing.py

@@ -34,6 +34,8 @@ logger = get_logger(__name__, category="core")
 INVALID_SPAN_ID = 0x0000000000000000
 INVALID_TRACE_ID = 0x00000000000000000000000000000000
 
+ROOT_SPAN_MARKERS = ["__root__", "__root_span__"]
+
 
 def trace_id_to_str(trace_id: int) -> str:
     """Convenience trace ID formatting method
@@ -178,7 +180,8 @@ async def start_trace(name: str, attributes: dict[str, Any] = None) -> TraceCont
 
     trace_id = generate_trace_id()
     context = TraceContext(BACKGROUND_LOGGER, trace_id)
-    context.push_span(name, {"__root__": True, **(attributes or {})})
+    attributes = {marker: True for marker in ROOT_SPAN_MARKERS} | (attributes or {})
+    context.push_span(name, attributes)
 
     CURRENT_TRACE_CONTEXT.set(context)
     return context