Scope A365 baggage processing to GenAI spans

A365_DOCUMENTATION.md

@@ -231,7 +231,11 @@ ObservabilityHostingManager.configure(
 
 ## Baggage
 
-Baggage sets per-request context (tenant, agent, user) that flows to all spans. **Without `tenant_id` and `agent_id`, the exporter silently drops spans.**
+Baggage sets per-request context (tenant, agent, user) that flows to recognized GenAI spans only. **Without `tenant_id` and `agent_id`, the exporter silently drops spans.**
+
+A span is recognized as GenAI at span start by evaluating these signals in order: a supported `gen_ai.operation.name` attribute; if that attribute is present but unrecognized (`chain`, `embeddings`, `text_completion`, `generate_content`, `create_agent`, ...) it is authoritative, so baggage and span-name inference are skipped and only the instrumentation scope can still classify the span; otherwise a span name matching a supported operation (`invoke_agent ...`, `chat ...`, ...) or a known pre-rename name (`chat.completions ...`), then a recognized `gen_ai.operation.name` baggage entry, then a supported GenAI instrumentation scope (`Agent365Sdk`, `semantic_kernel.*`, `agent_framework`, `microsoft.opentelemetry._genai.*`, `opentelemetry.instrumentation.openai_v2`, `opentelemetry.instrumentation.openai_agents`). The scope signal matters because LangChain, Semantic Kernel, Agent Framework, and OpenAI Agents all set `gen_ai.operation.name` *after* the span starts — a LangChain chat span begins life named `ChatOpenAI`, and a Semantic Kernel one as `chat.completions gpt-4o`.
+
+Spans classified only by instrumentation scope are GenAI with an unknown operation: they receive the common baggage attributes, but never the `invoke_agent`-only ones (caller agent details, `server.address`, `server.port`).
 
 ### BaggageBuilder
 

CHANGELOG.md

@@ -9,6 +9,9 @@
 - Remove optional dependency for langchain-core and document guidance for installation
   ([#269](https://github.com/microsoft/opentelemetry-distro-python/pull/269))
 
+### Bugs Fixed
+- Restrict A365 identity and baggage enrichment to recognized GenAI spans while preserving supported span-start signals. ([#265](https://github.com/microsoft/opentelemetry-distro-python/pull/265))
+
 # 1.3.9 (2026-09-09)
 ### Features Added
 - Update OpenTelemetry dependencies to latest versions, bump `langchain-core` minimum version to address S360, and support the new `httpx2` entry point exposed by `opentelemetry-instrumentation-httpx`.

src/microsoft/opentelemetry/a365/README.md

@@ -47,7 +47,7 @@ Span export pipeline — processors and exporters for Agent365 and Spectra backe
 | `agent365_exporter_options.py` | `Agent365ExporterOptions` — configuration for the Agent365 exporter (cluster category, token resolver, endpoint flags, batch settings). |
 | `enriched_span.py` | `EnrichedReadableSpan` — wrapper allowing extra attributes on immutable `ReadableSpan` objects. |
 | `enriching_span_processor.py` | Span enrichment support with registration for platform instrumentors (LangChain, Semantic Kernel, OpenAI Agents). `_EnrichingBatchSpanProcessor` applies enrichers before batching. |
-| `span_processor.py` | `A365SpanProcessor` — propagates documented OpenTelemetry baggage entries onto spans, keeps invoke_agent-specific handling, and copies only opted-in custom baggage keys onto recognized GenAI spans. |
+| `span_processor.py` | `A365SpanProcessor` — propagates documented and opted-in custom baggage entries onto recognized GenAI spans only (operation attribute, span name, operation baggage, or supported GenAI instrumentation scope), with special handling for invoke_agent spans. |
 | `spectra_exporter_options.py` | `SpectraExporterOptions` — configuration for OTLP export to a Spectra Collector sidecar (gRPC or HTTP, tuned for Kubernetes). |
 | `utils.py` | Exporter utilities: hex encoding for trace/span IDs, span size truncation, span partitioning, environment variable handling, payload building helpers. |
 

src/microsoft/opentelemetry/a365/core/exporters/span_processor.py

@@ -4,39 +4,9 @@
 # license information.
 # --------------------------------------------------------------------------
 
-"""Span processor for propagating OpenTelemetry baggage entries onto spans.
-
-For every new span:
-  * Retrieve the current (or parent) context
-  * Obtain all baggage entries
-  * For each documented key with a truthy value not already present as a span
-    attribute, add it via span.set_attribute
-  * Never overwrites existing attributes
-
-Custom baggage is propagated only to recognized GenAI spans. A span is
-recognized as GenAI by evaluating these signals in order at ``on_start``:
-
-  1. An explicit ``gen_ai.operation.name`` attribute holding a recognized
-     operation: GenAI with a known operation.
-  2. An explicit but *unrecognized* ``gen_ai.operation.name`` attribute: the
-     attribute is authoritative, so the baggage and span-name inference of
-     signals 3 and 4 is skipped. The span is still GenAI when a supported
-     instrumentation emitted it (signal 5), with an unknown operation.
-  3. A recognized ``gen_ai.operation.name`` baggage entry.
-  4. A span name that is (or starts with) a recognized operation name, or a
-     name a supported instrumentation is known to use before it renames the
-     span (Semantic Kernel ``chat.completions <model>``).
-  5. The instrumentation scope (source) name of a supported GenAI
-     instrumentation: GenAI with an unknown operation.
-
-Signals 4 and 5 exist because most GenAI instrumentations apply
-``gen_ai.operation.name`` *after* the span starts: LangChain chat spans start
-as ``ChatOpenAI`` and the OpenAI Agents processor starts workflow spans as
-``Agent workflow``. Signal 5 also keeps spans whose operation this processor
-does not model (``chain``, ``embeddings``, ``text_completion``,
-``generate_content``, ``create_agent``) from being dropped. Only signals 1, 3
-and 4 identify *which* operation a span represents, which is what gates the
-invoke_agent-only attributes.
+"""Propagate A365 identity and baggage to recognized GenAI spans.
+
+Existing span attributes are never overwritten.
 """
 
 from __future__ import annotations
@@ -84,7 +54,8 @@
 
 # mypy: disable-error-code="no-untyped-def"
 
-# Generic / common tracing attributes propagated from baggage to all spans
+
+# Baggage attributes for all recognized GenAI spans.
 COMMON_ATTRIBUTES = [
     TENANT_ID_KEY,
     CUSTOM_PARENT_SPAN_ID_KEY,
@@ -111,7 +82,7 @@
     SERVICE_NAME_KEY,
 ]
 
-# Invoke Agent-specific attributes (only propagated to invoke_agent spans)
+# Additional baggage attributes for invoke_agent spans.
 INVOKE_AGENT_ATTRIBUTES = [
     GEN_AI_CALLER_AGENT_ID_KEY,
     GEN_AI_CALLER_AGENT_NAME_KEY,
@@ -145,7 +116,7 @@ def _custom_baggage_keys(baggage_map) -> list[str]:
 class A365SpanProcessor(BaseSpanProcessor):
     """Span processor that stamps agent identity and propagates baggage to span attributes.
 
-    Static identity (tenant_id, agent_id) is set from configuration on every span.
+    Static identity (tenant_id, agent_id) is set from configuration on qualifying GenAI spans.
     Additional baggage entries are propagated selectively for documented keys.
     Never overwrites existing attributes.
     """
@@ -162,12 +133,23 @@ def __init__(
     def on_start(self, span, parent_context=None):  # type: ignore[override]
         ctx = parent_context or context.get_current()
 
-        # Stamp static identity from configuration (never overwrite existing)
         try:
             existing = getattr(span, "attributes", {}) or {}
         except Exception:
             existing = {}
 
+        if ctx is None:
+            baggage_map = {}
+        else:
+            try:
+                baggage_map = baggage.get_all(ctx) or {}
+            except Exception:
+                baggage_map = {}
+
+        classification = _classify_gen_ai_span(span, existing, baggage_map)
+        if not classification.is_gen_ai_span:
+            return super().on_start(span, parent_context)
+
         if self._tenant_id and TENANT_ID_KEY not in existing:
             try:
                 span.set_attribute(TENANT_ID_KEY, self._tenant_id)
@@ -179,22 +161,11 @@ def on_start(self, span, parent_context=None):  # type: ignore[override]
             except Exception:
                 pass
 
-        # Refresh existing after stamping identity
         try:
             existing = getattr(span, "attributes", {}) or {}
         except Exception:
             existing = {}
 
-        if ctx is None:
-            return super().on_start(span, parent_context)
-
-        try:
-            baggage_map = baggage.get_all(ctx) or {}
-        except Exception:
-            baggage_map = {}
-
-        classification = _classify_gen_ai_span(span, existing, baggage_map)
-
         target_keys = list(COMMON_ATTRIBUTES)
         if classification.operation_name == INVOKE_AGENT_OPERATION_NAME:
             for k in INVOKE_AGENT_ATTRIBUTES:

src/microsoft/opentelemetry/a365/core/exporters/utils.py

@@ -60,9 +60,7 @@
 # Maximum allowed span size in bytes (250KB)
 MAX_SPAN_SIZE_BYTES = 250 * 1024
 
-# Operation names that identify a span as eligible for export to the Agent 365
-# observability ingest service. Only spans whose gen_ai.operation.name matches
-# one of these values are included; all other spans are filtered out.
+# Export eligibility is intentionally narrower than GenAI span recognition.
 GEN_AI_OPERATION_NAMES: frozenset[str] = frozenset(
     {
         INVOKE_AGENT_OPERATION_NAME,