fix: handle multiple embedding events for llama-index (#1166)

python/instrumentation/openinference-instrumentation-llama-index/src/openinference/instrumentation/llama_index/_handler.py

@@ -4,6 +4,7 @@
 import json
 import logging
 import weakref
+from collections import defaultdict
 from dataclasses import dataclass
 from enum import Enum, auto
 from functools import singledispatch, singledispatchmethod
@@ -15,6 +16,7 @@
     TYPE_CHECKING,
     Any,
     AsyncGenerator,
+    DefaultDict,
     Dict,
     Generator,
     Iterable,
@@ -183,6 +185,7 @@ def __init__(
         self._attributes = {}
         self._end_time = None
         self._last_updated_at = time()
+        self._list_attr_len: DefaultDict[str, int] = defaultdict(int)
 
     def __setitem__(self, key: str, value: AttributeValue) -> None:
         self._attributes[key] = value
@@ -362,9 +365,12 @@ def _(self, event: EmbeddingStartEvent) -> None:
 
     @_process_event.register
     def _(self, event: EmbeddingEndEvent) -> None:
-        for i, (text, vector) in enumerate(zip(event.chunks, event.embeddings)):
+        i = self._list_attr_len[EMBEDDING_EMBEDDINGS]
+        for text, vector in zip(event.chunks, event.embeddings):
             self[f"{EMBEDDING_EMBEDDINGS}.{i}.{EMBEDDING_TEXT}"] = text
             self[f"{EMBEDDING_EMBEDDINGS}.{i}.{EMBEDDING_VECTOR}"] = vector
+            i += 1
+        self._list_attr_len[EMBEDDING_EMBEDDINGS] = i
 
     @_process_event.register
     def _(self, event: StreamChatStartEvent) -> None:

python/instrumentation/openinference-instrumentation-llama-index/tests/openinference/instrumentation/llama_index/conftest.py

@@ -0,0 +1,18 @@
+import pytest
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import SimpleSpanProcessor
+from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
+
+
+@pytest.fixture
+def in_memory_span_exporter() -> InMemorySpanExporter:
+    return InMemorySpanExporter()
+
+
+@pytest.fixture
+def tracer_provider(
+    in_memory_span_exporter: InMemorySpanExporter,
+) -> TracerProvider:
+    tracer_provider = TracerProvider()
+    tracer_provider.add_span_processor(SimpleSpanProcessor(in_memory_span_exporter))
+    return tracer_provider

python/instrumentation/openinference-instrumentation-llama-index/tests/openinference/instrumentation/llama_index/test_multiple_embedding_events.py

@@ -0,0 +1,49 @@
+from itertools import product
+from typing import Iterator
+
+import pytest
+from llama_index.core.instrumentation import get_dispatcher
+from llama_index.core.instrumentation.events.embedding import EmbeddingEndEvent
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
+
+from openinference.instrumentation.llama_index import LlamaIndexInstrumentor
+from openinference.semconv.trace import EmbeddingAttributes, SpanAttributes
+
+dispatcher = get_dispatcher(__name__)
+
+
+@dispatcher.span  # type: ignore[misc,unused-ignore]
+def foo(m: int, n: int) -> None:
+    for i in range(m):
+        chunks = [f"{i}-{j}" for j in range(n)]
+        embeddings = [list(map(float, [i, j])) for j in range(n)]
+        dispatcher.event(EmbeddingEndEvent(chunks=chunks, embeddings=embeddings))
+
+
+async def test_multiple_embedding_events(
+    in_memory_span_exporter: InMemorySpanExporter,
+) -> None:
+    m, n = 3, 2
+    foo(m, n)
+    span = in_memory_span_exporter.get_finished_spans()[0]
+    assert span.attributes
+    for k, (i, j) in enumerate(product(range(m), range(n))):
+        text, vector = f"{i}-{j}", tuple(map(float, [i, j]))
+        assert span.attributes[f"{EMBEDDING_EMBEDDINGS}.{k}.{EMBEDDING_TEXT}"] == text
+        assert span.attributes[f"{EMBEDDING_EMBEDDINGS}.{k}.{EMBEDDING_VECTOR}"] == vector
+
+
+@pytest.fixture(autouse=True)
+def instrument(
+    tracer_provider: TracerProvider,
+    in_memory_span_exporter: InMemorySpanExporter,
+) -> Iterator[None]:
+    LlamaIndexInstrumentor().instrument(tracer_provider=tracer_provider)
+    yield
+    LlamaIndexInstrumentor().uninstrument()
+
+
+EMBEDDING_EMBEDDINGS = SpanAttributes.EMBEDDING_EMBEDDINGS
+EMBEDDING_TEXT = EmbeddingAttributes.EMBEDDING_TEXT
+EMBEDDING_VECTOR = EmbeddingAttributes.EMBEDDING_VECTOR