feat: adds decorator implementation (#425)

* feat: adds decorator implementation

* fix tests

* adds async and generation support

* adds trycatch around wrapper operations

* adds langfuse singleton

* adds get_current_observation_id"

* removes unused imports

* refactor: exception handling

* docs: make pep compliant

* chore: v2.20.0a0

* fix: missing output if result is None

* chore: renames methods to private

* rename update methods

* renames to observe and langfuse_context

* adds scoring to decorators

* adds public option to trace params

* chore: v2.20.0a1

* chore: v2.20.0a2

* chore: v2.20.0a3

* add LangfuseDecorator to init to add it to docs

* fix: make llama-index optional in decorators implementation

* chore: v2.20.0a4

* fix: handle function args that are circular refs

* feat: adds capture_io param

* fix: remove wrong comments

* chore: v2.20.0a5

* fix: handle tuples in serialization

* chore: add pytest vscode support

* remove self argument of instance methods from logging

* split capture_io

* adds support for returned generators

* chore: update precommit hook to use ruff lint

* adds public option to llamaindex trace params

* refactor: async gen handling

* adds decorator support for nested openai integration

* add reset method to langfuse singleton

* remove print statements

* remove log_warning argument

* chore: v2.20.2a3

* fixes resetting of observation params on multiple update calls

* fix: allow observation scoring on trace level

* fixes score test

* feat: create wrapper trace if top level is generation

---------

Co-authored-by: Max Deichmann <[email protected]>

.pre-commit-config.yaml

@@ -1,5 +1,13 @@
 repos:
-  - repo: https://github.com/psf/black
-    rev: 23.3.0
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.3.2
     hooks:
-      - id: black
+      # Run the linter and fix 
+      - id: ruff
+        types_or: [ python, pyi, jupyter ]
+        args: [ --fix,  --config=ci.ruff.toml ]
+
+      # Run the formatter.
+      - id: ruff-format
+        types_or: [ python, pyi, jupyter ]
+

.vscode/settings.json

@@ -1,7 +1,7 @@
 {
   "[python]": {
     "editor.formatOnSave": true,
-    "editor.defaultFormatter": "charliermarsh.ruff",
+    "editor.defaultFormatter": "charliermarsh.ruff"
     // "editor.codeActionsOnSave": {
     //   "source.fixAll": "explicit",
     //   "source.organizeImports": "explicit"
@@ -13,5 +13,8 @@
   },
   "prettier.documentSelectors": [
     "**/*.{cjs,mjs,ts,tsx,astro,md,mdx,json,yaml,yml}"
-  ]
-}
+  ],
+  "python.testing.pytestArgs": ["tests"],
+  "python.testing.unittestEnabled": false,
+  "python.testing.pytestEnabled": true
+}

langfuse/client.py

@@ -5,7 +5,7 @@
 import uuid
 import httpx
 from enum import Enum
-from typing import Any, Literal, Optional
+from typing import Any, Optional
 
 from langfuse.api.resources.ingestion.types.create_event_body import CreateEventBody
 from langfuse.api.resources.ingestion.types.create_generation_body import (
@@ -47,6 +47,7 @@
 from langfuse.model import Dataset, MapValue, Observation, TraceWithFullDetails
 from langfuse.request import LangfuseClient
 from langfuse.task_manager import TaskManager
+from langfuse.types import SpanLevel
 from langfuse.utils import _convert_usage_input, _create_prompt_context, _get_timestamp
 
 from .version import __version__ as version
@@ -97,7 +98,7 @@ def __init__(
         flush_interval: float = 0.5,
         max_retries: int = 3,
         timeout: int = 10,  # seconds
-        sdk_integration: str = "default",
+        sdk_integration: Optional[str] = "default",
         httpx_client: Optional[httpx.Client] = None,
     ):
         """Initialize the Langfuse client.
@@ -783,7 +784,7 @@ def span(
         start_time: typing.Optional[dt.datetime] = None,
         end_time: typing.Optional[dt.datetime] = None,
         metadata: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         input: typing.Optional[typing.Any] = None,
         output: typing.Optional[typing.Any] = None,
@@ -889,7 +890,7 @@ def event(
         metadata: typing.Optional[typing.Any] = None,
         input: typing.Optional[typing.Any] = None,
         output: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         **kwargs,
@@ -986,7 +987,7 @@ def generation(
         end_time: typing.Optional[dt.datetime] = None,
         completion_start_time: typing.Optional[dt.datetime] = None,
         metadata: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         model: typing.Optional[str] = None,
@@ -1246,7 +1247,7 @@ def generation(
         start_time: typing.Optional[dt.datetime] = None,
         end_time: typing.Optional[dt.datetime] = None,
         metadata: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         completion_start_time: typing.Optional[dt.datetime] = None,
@@ -1359,7 +1360,7 @@ def span(
         metadata: typing.Optional[typing.Any] = None,
         input: typing.Optional[typing.Any] = None,
         output: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         **kwargs,
@@ -1524,7 +1525,7 @@ def event(
         metadata: typing.Optional[typing.Any] = None,
         input: typing.Optional[typing.Any] = None,
         output: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         **kwargs,
@@ -1638,7 +1639,7 @@ def update(
         end_time: typing.Optional[dt.datetime] = None,
         completion_start_time: typing.Optional[dt.datetime] = None,
         metadata: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         model: typing.Optional[str] = None,
@@ -1739,7 +1740,7 @@ def end(
         end_time: typing.Optional[dt.datetime] = None,
         completion_start_time: typing.Optional[dt.datetime] = None,
         metadata: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         model: typing.Optional[str] = None,
@@ -1841,7 +1842,7 @@ def update(
         metadata: typing.Optional[typing.Any] = None,
         input: typing.Optional[typing.Any] = None,
         output: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         **kwargs,
@@ -1924,7 +1925,7 @@ def end(
         metadata: typing.Optional[typing.Any] = None,
         input: typing.Optional[typing.Any] = None,
         output: typing.Optional[typing.Any] = None,
-        level: typing.Optional[Literal["DEBUG", "DEFAULT", "WARNING", "ERROR"]] = None,
+        level: typing.Optional[SpanLevel] = None,
         status_message: typing.Optional[str] = None,
         version: typing.Optional[str] = None,
         **kwargs,
@@ -1998,11 +1999,6 @@ def get_langchain_handler(self):
 
         return CallbackHandler(stateful_client=self)
 
-    def get_llama_index_handler(self):
-        from langfuse.llama_index import LlamaIndexCallbackHandler
-
-        return LlamaIndexCallbackHandler(stateful_client=self)
-
 
 class StatefulTraceClient(StatefulClient):
     """Class for handling stateful operations of traces in the Langfuse system. Inherits from StatefulClient.

langfuse/decorators/__init__.py

@@ -0,0 +1,3 @@
+from .langfuse_decorator import langfuse_context, observe, LangfuseDecorator
+
+__all__ = ["langfuse_context", "observe", "LangfuseDecorator"]