Merge branch 'main' of github.com:Scale3-Labs/langtrace-python-sdk into release

Author: karthikscale3

src/examples/milvus_example/main.py

@@ -0,0 +1,106 @@
+from pymilvus import MilvusClient, model
+from typing import List
+from langtrace_python_sdk import langtrace, with_langtrace_root_span
+from dotenv import load_dotenv
+
+load_dotenv()
+langtrace.init()
+
+client = MilvusClient("milvus_demo.db")
+
+COLLECTION_NAME = "demo_collection"
+embedding_fn = model.DefaultEmbeddingFunction()
+
+
+def create_collection(collection_name: str = COLLECTION_NAME):
+    if client.has_collection(collection_name=collection_name):
+        client.drop_collection(collection_name=collection_name)
+
+    client.create_collection(
+        collection_name=collection_name,
+        dimension=768,  # The vectors we will use in this demo has 768 dimensions
+    )
+
+
+def create_embedding(docs: List[str] = [], subject: str = "history"):
+    """
+    Create embeddings for the given documents.
+    """
+
+    vectors = embedding_fn.encode_documents(docs)
+    # Each entity has id, vector representation, raw text, and a subject label that we use
+    # to demo metadata filtering later.
+    data = [
+        {"id": i, "vector": vectors[i], "text": docs[i], "subject": subject}
+        for i in range(len(vectors))
+    ]
+    # print("Data has", len(data), "entities, each with fields: ", data[0].keys())
+    # print("Vector dim:", len(data[0]["vector"]))
+    return data
+
+
+def insert_data(collection_name: str = COLLECTION_NAME, data: List[dict] = []):
+    client.insert(
+        collection_name=collection_name,
+        data=data,
+    )
+
+
+def vector_search(collection_name: str = COLLECTION_NAME, queries: List[str] = []):
+    query_vectors = embedding_fn.encode_queries(queries)
+    # If you don't have the embedding function you can use a fake vector to finish the demo:
+    # query_vectors = [ [ random.uniform(-1, 1) for _ in range(768) ] ]
+
+    res = client.search(
+        collection_name="demo_collection",  # target collection
+        data=query_vectors,  # query vectors
+        limit=2,  # number of returned entities
+        output_fields=["text", "subject"],  # specifies fields to be returned
+        timeout=10,
+        partition_names=["history"],
+        anns_field="vector",
+        search_params={"nprobe": 10},
+    )
+
+
+def query(collection_name: str = COLLECTION_NAME, query: str = ""):
+    res = client.query(
+        collection_name=collection_name,
+        filter=query,
+        # output_fields=["text", "subject"],
+    )
+
+    # print(res)
+
+
+@with_langtrace_root_span("milvus_example")
+def main():
+    create_collection()
+    # insert Alan Turing's history
+    turing_data = create_embedding(
+        docs=[
+            "Artificial intelligence was founded as an academic discipline in 1956.",
+            "Alan Turing was the first person to conduct substantial research in AI.",
+            "Born in Maida Vale, London, Turing was raised in southern England.",
+        ]
+    )
+    insert_data(data=turing_data)
+
+    # insert AI Drug Discovery
+    drug_data = create_embedding(
+        docs=[
+            "Machine learning has been used for drug design.",
+            "Computational synthesis with AI algorithms predicts molecular properties.",
+            "DDR1 is involved in cancers and fibrosis.",
+        ],
+        subject="biology",
+    )
+    insert_data(data=drug_data)
+
+    vector_search(queries=["Who is Alan Turing?"])
+    query(query="subject == 'history'")
+    query(query="subject == 'biology'")
+
+
+if __name__ == "__main__":
+    main()

src/langtrace_python_sdk/constants/instrumentation/common.py

@@ -37,6 +37,7 @@
     "MONGODB": "MongoDB",
     "AWS_BEDROCK": "AWS Bedrock",
     "CEREBRAS": "Cerebras",
+    "MILVUS": "Milvus",
 }
 
 LANGTRACE_ADDITIONAL_SPAN_ATTRIBUTES_KEY = "langtrace_additional_attributes"

src/langtrace_python_sdk/constants/instrumentation/milvus.py

@@ -0,0 +1,38 @@
+APIS = {
+    "INSERT": {
+        "MODULE": "pymilvus",
+        "METHOD": "MilvusClient.insert",
+        "OPERATION": "insert",
+        "SPAN_NAME": "Milvus Insert",
+    },
+    "QUERY": {
+        "MODULE": "pymilvus",
+        "METHOD": "MilvusClient.query",
+        "OPERATION": "query",
+        "SPAN_NAME": "Milvus Query",
+    },
+    "SEARCH": {
+        "MODULE": "pymilvus",
+        "METHOD": "MilvusClient.search",
+        "OPERATION": "search",
+        "SPAN_NAME": "Milvus Search",
+    },
+    "DELETE": {
+        "MODULE": "pymilvus",
+        "METHOD": "MilvusClient.delete",
+        "OPERATION": "delete",
+        "SPAN_NAME": "Milvus Delete",
+    },
+    "CREATE_COLLECTION": {
+        "MODULE": "pymilvus",
+        "METHOD": "MilvusClient.create_collection",
+        "OPERATION": "create_collection",
+        "SPAN_NAME": "Milvus Create Collection",
+    },
+    "UPSERT": {
+        "MODULE": "pymilvus",
+        "METHOD": "MilvusClient.upsert",
+        "OPERATION": "upsert",
+        "SPAN_NAME": "Milvus Upsert",
+    },
+}

src/langtrace_python_sdk/instrumentation/__init__.py

@@ -23,6 +23,7 @@
 from .litellm import LiteLLMInstrumentation
 from .pymongo import PyMongoInstrumentation
 from .cerebras import CerebrasInstrumentation
+from .milvus import MilvusInstrumentation
 
 __all__ = [
     "AnthropicInstrumentation",
@@ -50,4 +51,5 @@
     "PyMongoInstrumentation",
     "AWSBedrockInstrumentation",
     "CerebrasInstrumentation",
+    "MilvusInstrumentation",
 ]

src/langtrace_python_sdk/instrumentation/langchain_community/patch.py

@@ -71,16 +71,26 @@ def traced_method(wrapped, instance, args, kwargs):
                 result = wrapped(*args, **kwargs)
                 if trace_output:
                     span.set_attribute("langchain.outputs", to_json_string(result))
-
-                    prompt_tokens = instance.get_num_tokens(args[0])
-                    completion_tokens = instance.get_num_tokens(result)
-                    if hasattr(result, 'usage'):
+                    prompt_tokens = (
+                        instance.get_num_tokens(args[0])
+                        if hasattr(instance, "get_num_tokens")
+                        else None
+                    )
+                    completion_tokens = (
+                        instance.get_num_tokens(result)
+                        if hasattr(instance, "get_num_tokens")
+                        else None
+                    )
+                    if hasattr(result, "usage"):
                         prompt_tokens = result.usage.prompt_tokens
                         completion_tokens = result.usage.completion_tokens
 
-                    span.set_attribute(SpanAttributes.LLM_USAGE_COMPLETION_TOKENS, prompt_tokens)
-                    span.set_attribute(SpanAttributes.LLM_USAGE_PROMPT_TOKENS, completion_tokens)
-
+                    span.set_attribute(
+                        SpanAttributes.LLM_USAGE_COMPLETION_TOKENS, prompt_tokens
+                    )
+                    span.set_attribute(
+                        SpanAttributes.LLM_USAGE_PROMPT_TOKENS, completion_tokens
+                    )
 
                 span.set_status(StatusCode.OK)
                 return result
@@ -102,9 +112,17 @@ def clean_empty(d):
     if not isinstance(d, (dict, list, tuple)):
         return d
     if isinstance(d, tuple):
-        return tuple(val for val in (clean_empty(val) for val in d) if val != () and val is not None)
+        return tuple(
+            val
+            for val in (clean_empty(val) for val in d)
+            if val != () and val is not None
+        )
     if isinstance(d, list):
-        return [val for val in (clean_empty(val) for val in d) if val != [] and val is not None]
+        return [
+            val
+            for val in (clean_empty(val) for val in d)
+            if val != [] and val is not None
+        ]
     result = {}
     for k, val in d.items():
         if isinstance(val, dict):
@@ -120,7 +138,7 @@ def clean_empty(d):
                 result[k] = val.strip()
         elif isinstance(val, object):
             # some langchain objects have a text attribute
-            val = getattr(val, 'text', None)
+            val = getattr(val, "text", None)
             if val is not None and val.strip() != "":
                 result[k] = val.strip()
     return result

src/langtrace_python_sdk/instrumentation/milvus/__init__.py

@@ -0,0 +1,3 @@
+from .instrumentation import MilvusInstrumentation
+
+__all__ = ["MilvusInstrumentation"]

src/langtrace_python_sdk/instrumentation/milvus/instrumentation.py

@@ -0,0 +1,29 @@
+from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
+from opentelemetry.trace import get_tracer
+
+from typing import Collection
+from importlib_metadata import version as v
+from wrapt import wrap_function_wrapper as _W
+
+from langtrace_python_sdk.constants.instrumentation.milvus import APIS
+from .patch import generic_patch
+
+
+class MilvusInstrumentation(BaseInstrumentor):
+
+    def instrumentation_dependencies(self) -> Collection[str]:
+        return ["pymilvus >= 2.4.1"]
+
+    def _instrument(self, **kwargs):
+        tracer_provider = kwargs.get("tracer_provider")
+        tracer = get_tracer(__name__, "", tracer_provider)
+        version = v("pymilvus")
+        for api in APIS.values():
+            _W(
+                module=api["MODULE"],
+                name=api["METHOD"],
+                wrapper=generic_patch(api, version, tracer),
+            )
+
+    def _uninstrument(self, **kwargs):
+        pass

src/langtrace_python_sdk/instrumentation/milvus/patch.py

@@ -0,0 +1,125 @@
+from langtrace_python_sdk.utils.silently_fail import silently_fail
+from opentelemetry.trace import Tracer
+from opentelemetry.trace import SpanKind
+from langtrace_python_sdk.utils import handle_span_error, set_span_attribute
+from langtrace_python_sdk.utils.llm import (
+    get_extra_attributes,
+    set_span_attributes,
+)
+import json
+
+
+def generic_patch(api, version: str, tracer: Tracer):
+    def traced_method(wrapped, instance, args, kwargs):
+        span_name = api["SPAN_NAME"]
+        operation = api["OPERATION"]
+        with tracer.start_as_current_span(span_name, kind=SpanKind.CLIENT) as span:
+            try:
+                span_attributes = {
+                    "db.system": "milvus",
+                    "db.operation": operation,
+                    "db.name": kwargs.get("collection_name", None),
+                    **get_extra_attributes(),
+                }
+
+                if operation == "create_collection":
+                    set_create_collection_attributes(span_attributes, kwargs)
+
+                elif operation == "insert" or operation == "upsert":
+                    set_insert_or_upsert_attributes(span_attributes, kwargs)
+
+                elif operation == "search":
+                    set_search_attributes(span_attributes, kwargs)
+
+                elif operation == "query":
+                    set_query_attributes(span_attributes, kwargs)
+
+                set_span_attributes(span, span_attributes)
+                result = wrapped(*args, **kwargs)
+
+                if operation == "query":
+                    set_query_response_attributes(span, result)
+
+                if operation == "search":
+                    set_search_response_attributes(span, result)
+                return result
+            except Exception as err:
+                handle_span_error(span, err)
+                raise
+
+    return traced_method
+
+
+@silently_fail
+def set_create_collection_attributes(span_attributes, kwargs):
+    span_attributes["db.dimension"] = kwargs.get("dimension", None)
+
+
+@silently_fail
+def set_insert_or_upsert_attributes(span_attributes, kwargs):
+    data = kwargs.get("data")
+    timeout = kwargs.get("timeout")
+    partition_name = kwargs.get("partition_name")
+
+    span_attributes["db.num_entities"] = len(data) if data else None
+    span_attributes["db.timeout"] = timeout
+    span_attributes["db.partition_name"] = partition_name
+
+
+@silently_fail
+def set_search_attributes(span_attributes, kwargs):
+    data = kwargs.get("data")
+    filter = kwargs.get("filter")
+    limit = kwargs.get("limit")
+    output_fields = kwargs.get("output_fields")
+    search_params = kwargs.get("search_params")
+    timeout = kwargs.get("timeout")
+    partition_names = kwargs.get("partition_names")
+    anns_field = kwargs.get("anns_field")
+    span_attributes["db.num_queries"] = len(data) if data else None
+    span_attributes["db.filter"] = filter
+    span_attributes["db.limit"] = limit
+    span_attributes["db.output_fields"] = json.dumps(output_fields)
+    span_attributes["db.search_params"] = json.dumps(search_params)
+    span_attributes["db.partition_names"] = json.dumps(partition_names)
+    span_attributes["db.anns_field"] = anns_field
+    span_attributes["db.timeout"] = timeout
+
+
+@silently_fail
+def set_query_attributes(span_attributes, kwargs):
+    filter = kwargs.get("filter")
+    output_fields = kwargs.get("output_fields")
+    timeout = kwargs.get("timeout")
+    partition_names = kwargs.get("partition_names")
+    ids = kwargs.get("ids")
+
+    span_attributes["db.filter"] = filter
+    span_attributes["db.output_fields"] = output_fields
+    span_attributes["db.timeout"] = timeout
+    span_attributes["db.partition_names"] = partition_names
+    span_attributes["db.ids"] = ids
+
+
+@silently_fail
+def set_query_response_attributes(span, result):
+    set_span_attribute(span, name="db.num_matches", value=len(result))
+    for match in result:
+        span.add_event(
+            "db.query.match",
+            attributes=match,
+        )
+
+
+@silently_fail
+def set_search_response_attributes(span, result):
+    for res in result:
+        for match in res:
+            span.add_event(
+                "db.search.match",
+                attributes={
+                    "id": match["id"],
+                    "distance": str(match["distance"]),
+                    "entity": json.dumps(match["entity"]),
+                },
+            )

src/langtrace_python_sdk/instrumentation/pymongo/patch.py

@@ -38,7 +38,6 @@ def traced_method(wrapped, instance, args, kwargs):
 
             try:
                 result = wrapped(*args, **kwargs)
-                print(result)
                 for doc in result:
                     if span.is_recording():
                         span.add_event(