[BUZZOK-26844] Remove drum server from everywhere in genai_agents image, replace with drum inline predictor

public_dropin_environments/python311_genai_agents/env_info.json

@@ -4,7 +4,7 @@
   "description": "This template environment can be used to create GenAI-powered agents using CrewAI, LangGraph, or Llama-Index. Similar to other drop-in environments, you can either include a .pth artifact or any other code needed to deserialize your model, and optionally a custom.py file. You can also use this environment in codespaces.",
   "programmingLanguage": "python",
   "label": "",
-  "environmentVersionId": "6876b52bd0b01d0513661808",
+  "environmentVersionId": "6879340432d95c11903cb7bb",
   "environmentVersionDescription": "",
   "isPublic": true,
   "isDownloadable": true,
@@ -15,8 +15,8 @@
   "contextUrl": "https://github.com/datarobot/datarobot-user-models/tree/master/public_dropin_environments/python311_genai_agents",
   "imageRepository": "env-python-genai-agents",
   "tags": [
-    "v11.2.0-6876b52bd0b01d0513661808",
-    "6876b52bd0b01d0513661808",
+    "v11.2.0-6879340432d95c11903cb7bb",
+    "6879340432d95c11903cb7bb",
     "v11.2.0-latest"
   ]
 }

public_dropin_environments/python311_genai_agents/requirements.in

@@ -26,9 +26,10 @@ crewai-tools>=0.48.0
 datarobot-drum>=1.16.19
 datarobot-moderations>=11.2.0
 datarobot-mlops>=11.1.0
-datarobot>=3.7.0
+datarobot[auth]>=3.8.1
 dotenv~=0.9.9
 langchain-community~=0.3.23
+langchain-mcp-adapters~=0.1.9
 langchain~=0.3.23
 langgraph~=0.4.10
 langgraph-prebuilt~=0.2.3
@@ -39,18 +40,19 @@ llama-index-llms-langchain~=0.6.1
 llama-index-llms-litellm~=0.4.1
 llama-index-llms-openai~=0.3.38
 llama-index~=0.12.33
+mcp~=1.11.0
 multidict~=6.5.0
 onnxruntime~=1.22.0
 openai~=1.76.2
-opentelemetry-api~=1.33.0
-opentelemetry-instrumentation-aiohttp-client~=0.54b0
-opentelemetry-instrumentation-crewai~=0.40.5
-opentelemetry-instrumentation-httpx~=0.54b0
-opentelemetry-instrumentation-langchain~=0.40.5
-opentelemetry-instrumentation-llamaindex~=0.40.5
-opentelemetry-instrumentation-openai~=0.40.5
-opentelemetry-instrumentation-requests~=0.54b0
-opentelemetry-sdk~=1.33.0
+opentelemetry-api>=1.33.0,<2.0.0
+opentelemetry-instrumentation-aiohttp-client>=0.54b0
+opentelemetry-instrumentation-crewai>=0.40.5
+opentelemetry-instrumentation-httpx>=0.54b0
+opentelemetry-instrumentation-langchain>=0.40.5
+opentelemetry-instrumentation-llamaindex>=0.40.5
+opentelemetry-instrumentation-openai>=0.40.5
+opentelemetry-instrumentation-requests>=0.54b0
+opentelemetry-sdk>=1.33.0
 python-dotenv~=1.1.0
 ragas @ git+https://github.com/explodinggradients/ragas@5d59549ad5ef511f621502c563bc55ac5aeb9188#subdirectory=ragas
 requests~=2.32.4

public_dropin_environments/python311_genai_agents/requirements.txt

@@ -57,7 +57,7 @@ crewai==0.140.0
 crewai-tools==0.48.0
 cryptography==44.0.3
 dataclasses-json==0.6.7
-datarobot==3.7.1
+datarobot[auth]==3.8.2
 datarobot-drum==1.16.19
 datarobot-mlops==11.1.0
 datarobot-moderations==11.2.0
@@ -169,6 +169,7 @@ langchain-cohere==0.3.5
 langchain-community==0.3.23
 langchain-core==0.3.59
 langchain-experimental==0.3.4
+langchain-mcp-adapters==0.1.9
 langchain-nvidia-ai-endpoints==0.3.10
 langchain-openai==0.2.14
 langchain-text-splitters==0.3.8
@@ -208,6 +209,7 @@ markdown-it-py==3.0.0
 markupsafe==3.0.2
 marshmallow==3.26.1
 matplotlib-inline==0.1.7
+mcp==1.11.0
 mdurl==0.1.2
 mem0ai==0.1.98
 memory-profiler==0.61.0
@@ -334,6 +336,7 @@ pytest-xdist==3.7.0
 python-dateutil==2.9.0.post0
 python-dotenv==1.1.0
 python-json-logger==3.3.0
+python-multipart==0.0.20
 pytube==15.0.0
 pytz==2024.2
 pyvis==0.3.2
@@ -368,6 +371,7 @@ six==1.17.0
 sniffio==1.3.1
 soupsieve==2.7
 sqlalchemy[asyncio]==2.0.40
+sse-starlette==2.4.1
 stack-data==0.6.3
 starlette==0.46.2
 strenum==0.4.15
@@ -393,7 +397,7 @@ twine==5.1.1
 typer==0.15.3
 types-python-dateutil==2.9.0.20241206
 types-requests==2.32.0.20250328
-typing-extensions==4.13.2
+typing-extensions==4.14.1
 typing-inspect==0.9.0
 typing-inspection==0.4.0
 tzdata==2025.2

public_dropin_environments/python311_genai_agents/run_agent.py

@@ -22,18 +22,14 @@
 from typing import Any, TextIO, cast
 from urllib.parse import urlparse, urlunparse
 
-import requests
 from datarobot_drum.drum.enum import TargetType
 from datarobot_drum.drum.root_predictors.drum_inline_utils import drum_inline_predictor
-from datarobot_drum.drum.root_predictors.drum_server_utils import DrumServerRun
-from openai import OpenAI
 from openai.types.chat import ChatCompletion
 from openai.types.chat.completion_create_params import (
     CompletionCreateParamsBase,
 )
 from opentelemetry import trace
 from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
-from opentelemetry.propagate import inject
 from opentelemetry.sdk.trace import TracerProvider
 from opentelemetry.sdk.trace.export import SimpleSpanProcessor
 from opentelemetry.trace import Span, use_span
@@ -89,11 +85,6 @@ def argparse_args() -> argparse.Namespace:
         default=None,
         help="Custom attributes for tracing. Should be a JSON dictionary.",
     )
-    parser.add_argument(
-        "--use_serverless",
-        action="store_true",
-        help="Use DRUM serverless predictor.",
-    )
     args = parser.parse_args()
     return args
 
@@ -203,55 +194,6 @@ def setup_otel(args: Any) -> Span:
     return span
 
 
-def execute_drum(
-    chat_completion: CompletionCreateParamsBase,
-    default_headers: dict[str, str],
-    custom_model_dir: Path,
-) -> ChatCompletion:
-    root.info("Executing agent as [chat] endpoint. DRUM Executor.")
-    root.info("Starting DRUM server.")
-    with DrumServerRun(
-        target_type=TargetType.AGENTIC_WORKFLOW.value,
-        labels=None,
-        custom_model_dir=custom_model_dir,
-        with_error_server=True,
-        production=False,
-        verbose=True,
-        logging_level="info",
-        target_name="response",
-        wait_for_server_timeout=360,
-        port=get_open_port(),
-        stream_output=True,
-        max_workers=2,  # this will force drum tracing to not use batchprocessor
-    ) as drum_runner:
-        root.info("Verifying DRUM server")
-        response = requests.get(drum_runner.url_server_address)
-        if not response.ok:
-            root.error("Server failed to start")
-            try:
-                root.error(response.text)
-            finally:
-                raise RuntimeError("Server failed to start")
-
-        # inject OTEL headers into default_headers
-        inject(default_headers)
-
-        # Use a standard OpenAI client to call the DRUM server. This mirrors the behavior of a deployed agent.
-        # Using the `chat.completions.create` method ensures the parameters are OpenAI compatible.
-        root.info("Executing Agent")
-        client = OpenAI(
-            base_url=drum_runner.url_server_address,
-            api_key="not-required",
-            default_headers=default_headers,
-            max_retries=0,
-        )
-        completion = client.chat.completions.create(**chat_completion)
-
-    # Continue outside the context manager to ensure the server is stopped and logs
-    # are flushed before we write the output
-    return completion
-
-
 def execute_drum_inline(
     chat_completion: CompletionCreateParamsBase,
     custom_model_dir: Path,
@@ -304,17 +246,10 @@ def run_agent_procedure(args: Any) -> None:
         root.info(f"Trace id: {trace_id}")
 
         root.info(f"Executing request in directory {args.custom_model_dir}")
-        if args.use_serverless:
-            result = execute_drum_inline(
-                chat_completion=chat_completion,
-                custom_model_dir=args.custom_model_dir,
-            )
-        else:
-            result = execute_drum(
-                chat_completion=chat_completion,
-                default_headers=default_headers,
-                custom_model_dir=args.custom_model_dir,
-            )
+        result = execute_drum_inline(
+            chat_completion=chat_completion,
+            custom_model_dir=args.custom_model_dir,
+        )
         store_result(
             result,
             trace_id,