Optimize mega flow by removing microservice wrapper (#582)

* refactor orchestrator

* fix

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* remove no_wrapper

* fix

* fix

* add align_gen

* add retriever and rerank params

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* add fake test for customize params

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix dep

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: Spycsh

comps/cores/mega/gateway.py

@@ -20,7 +20,7 @@
     EmbeddingRequest,
     UsageInfo,
 )
-from ..proto.docarray import LLMParams, LLMParamsDoc, RerankedDoc, TextDoc
+from ..proto.docarray import LLMParams, LLMParamsDoc, RerankedDoc, RerankerParms, RetrieverParms, TextDoc
 from .constants import MegaServiceEndpoint, ServiceRoleType, ServiceType
 from .micro_service import MicroService
 
@@ -167,8 +167,22 @@ async def handle_request(self, request: Request):
             streaming=stream_opt,
             chat_template=chat_request.chat_template if chat_request.chat_template else None,
         )
+        retriever_parameters = RetrieverParms(
+            search_type=chat_request.search_type if chat_request.search_type else "similarity",
+            k=chat_request.k if chat_request.k else 4,
+            distance_threshold=chat_request.distance_threshold if chat_request.distance_threshold else None,
+            fetch_k=chat_request.fetch_k if chat_request.fetch_k else 20,
+            lambda_mult=chat_request.lambda_mult if chat_request.lambda_mult else 0.5,
+            score_threshold=chat_request.score_threshold if chat_request.score_threshold else 0.2,
+        )
+        reranker_parameters = RerankerParms(
+            top_n=chat_request.top_n if chat_request.top_n else 1,
+        )
         result_dict, runtime_graph = await self.megaservice.schedule(
-            initial_inputs={"text": prompt}, llm_parameters=parameters
+            initial_inputs={"text": prompt},
+            llm_parameters=parameters,
+            retriever_parameters=retriever_parameters,
+            reranker_parameters=reranker_parameters,
         )
         for node, response in result_dict.items():
             if isinstance(response, StreamingResponse):

comps/cores/mega/orchestrator.py

@@ -4,6 +4,7 @@
 import asyncio
 import copy
 import json
+import os
 import re
 from typing import Dict, List
 
@@ -14,6 +15,10 @@
 from ..proto.docarray import LLMParams
 from .constants import ServiceType
 from .dag import DAG
+from .logger import CustomLogger
+
+logger = CustomLogger("comps-core-orchestrator")
+LOGFLAG = os.getenv("LOGFLAG", False)
 
 
 class ServiceOrchestrator(DAG):
@@ -36,18 +41,22 @@ def flow_to(self, from_service, to_service):
             self.add_edge(from_service.name, to_service.name)
             return True
         except Exception as e:
-            print(e)
+            logger.error(e)
             return False
 
-    async def schedule(self, initial_inputs: Dict, llm_parameters: LLMParams = LLMParams()):
+    async def schedule(self, initial_inputs: Dict, llm_parameters: LLMParams = LLMParams(), **kwargs):
         result_dict = {}
         runtime_graph = DAG()
         runtime_graph.graph = copy.deepcopy(self.graph)
+        if LOGFLAG:
+            logger.info(initial_inputs)
 
         timeout = aiohttp.ClientTimeout(total=1000)
         async with aiohttp.ClientSession(trust_env=True, timeout=timeout) as session:
             pending = {
-                asyncio.create_task(self.execute(session, node, initial_inputs, runtime_graph, llm_parameters))
+                asyncio.create_task(
+                    self.execute(session, node, initial_inputs, runtime_graph, llm_parameters, **kwargs)
+                )
                 for node in self.ind_nodes()
             }
             ind_nodes = self.ind_nodes()
@@ -67,11 +76,12 @@ async def schedule(self, initial_inputs: Dict, llm_parameters: LLMParams = LLMPa
                             for downstream in reversed(downstreams):
                                 try:
                                     if re.findall(black_node, downstream):
-                                        print(f"skip forwardding to {downstream}...")
+                                        if LOGFLAG:
+                                            logger.info(f"skip forwardding to {downstream}...")
                                         runtime_graph.delete_edge(node, downstream)
                                         downstreams.remove(downstream)
                                 except re.error as e:
-                                    print("Pattern invalid! Operation cancelled.")
+                                    logger.error("Pattern invalid! Operation cancelled.")
                             if len(downstreams) == 0 and llm_parameters.streaming:
                                 # turn the response to a StreamingResponse
                                 # to make the response uniform to UI
@@ -90,7 +100,7 @@ def fake_stream(text):
                             inputs = self.process_outputs(runtime_graph.predecessors(d_node), result_dict)
                             pending.add(
                                 asyncio.create_task(
-                                    self.execute(session, d_node, inputs, runtime_graph, llm_parameters)
+                                    self.execute(session, d_node, inputs, runtime_graph, llm_parameters, **kwargs)
                                 )
                             )
         nodes_to_keep = []
@@ -121,21 +131,33 @@ async def execute(
         inputs: Dict,
         runtime_graph: DAG,
         llm_parameters: LLMParams = LLMParams(),
+        **kwargs,
     ):
         # send the cur_node request/reply
         endpoint = self.services[cur_node].endpoint_path
         llm_parameters_dict = llm_parameters.dict()
-        for field, value in llm_parameters_dict.items():
-            if inputs.get(field) != value:
-                inputs[field] = value
+        if self.services[cur_node].service_type == ServiceType.LLM:
+            for field, value in llm_parameters_dict.items():
+                if inputs.get(field) != value:
+                    inputs[field] = value
+
+        # pre-process
+        inputs = self.align_inputs(inputs, cur_node, runtime_graph, llm_parameters_dict, **kwargs)
 
         if (
             self.services[cur_node].service_type == ServiceType.LLM
             or self.services[cur_node].service_type == ServiceType.LVM
         ) and llm_parameters.streaming:
             # Still leave to sync requests.post for StreamingResponse
+            if LOGFLAG:
+                logger.info(inputs)
             response = requests.post(
-                url=endpoint, data=json.dumps(inputs), proxies={"http": None}, stream=True, timeout=1000
+                url=endpoint,
+                data=json.dumps(inputs),
+                headers={"Content-type": "application/json"},
+                proxies={"http": None},
+                stream=True,
+                timeout=1000,
             )
             downstream = runtime_graph.downstream(cur_node)
             if downstream:
@@ -169,11 +191,32 @@ def generate():
                             else:
                                 yield chunk
 
-            return StreamingResponse(generate(), media_type="text/event-stream"), cur_node
+            return (
+                StreamingResponse(self.align_generator(generate(), **kwargs), media_type="text/event-stream"),
+                cur_node,
+            )
         else:
+            if LOGFLAG:
+                logger.info(inputs)
             async with session.post(endpoint, json=inputs) as response:
-                print(f"{cur_node}: {response.status}")
-                return await response.json(), cur_node
+                # Parse as JSON
+                data = await response.json()
+                # post process
+                data = self.align_outputs(data, cur_node, inputs, runtime_graph, llm_parameters_dict, **kwargs)
+
+                return data, cur_node
+
+    def align_inputs(self, inputs, *args, **kwargs):
+        """Override this method in megaservice definition."""
+        return inputs
+
+    def align_outputs(self, data, *args, **kwargs):
+        """Override this method in megaservice definition."""
+        return data
+
+    def align_generator(self, gen, *args, **kwargs):
+        """Override this method in megaservice definition."""
+        return gen
 
     def dump_outputs(self, node, response, result_dict):
         result_dict[node] = response

comps/cores/proto/docarray.py

@@ -173,6 +173,19 @@ class LLMParams(BaseDoc):
     )
 
 
+class RetrieverParms(BaseDoc):
+    search_type: str = "similarity"
+    k: int = 4
+    distance_threshold: Optional[float] = None
+    fetch_k: int = 20
+    lambda_mult: float = 0.5
+    score_threshold: float = 0.2
+
+
+class RerankerParms(BaseDoc):
+    top_n: int = 1
+
+
 class RAGASParams(BaseDoc):
     questions: DocList[TextDoc]
     answers: DocList[TextDoc]

tests/cores/mega/test_service_orchestrator_with_retriever_rerank_fake.py

@@ -0,0 +1,83 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+import json
+import unittest
+
+from comps import (
+    EmbedDoc,
+    Gateway,
+    RerankedDoc,
+    ServiceOrchestrator,
+    TextDoc,
+    opea_microservices,
+    register_microservice,
+)
+from comps.cores.mega.constants import ServiceType
+from comps.cores.proto.docarray import LLMParams, RerankerParms, RetrieverParms
+
+
+@register_microservice(name="s1", host="0.0.0.0", port=8083, endpoint="/v1/add", service_type=ServiceType.RETRIEVER)
+async def s1_add(request: EmbedDoc) -> TextDoc:
+    req = request.model_dump_json()
+    req_dict = json.loads(req)
+    text = req_dict["text"]
+    text += f"opea top_k {req_dict['k']}"
+    return {"text": text}
+
+
+@register_microservice(name="s2", host="0.0.0.0", port=8084, endpoint="/v1/add", service_type=ServiceType.RERANK)
+async def s2_add(request: TextDoc) -> TextDoc:
+    req = request.model_dump_json()
+    req_dict = json.loads(req)
+    text = req_dict["text"]
+    text += "project!"
+    return {"text": text}
+
+
+def align_inputs(self, inputs, cur_node, runtime_graph, llm_parameters_dict, **kwargs):
+    if self.services[cur_node].service_type == ServiceType.RETRIEVER:
+        inputs["k"] = kwargs["retriever_parameters"].k
+
+    return inputs
+
+
+def align_outputs(self, outputs, cur_node, inputs, runtime_graph, llm_parameters_dict, **kwargs):
+    if self.services[cur_node].service_type == ServiceType.RERANK:
+        top_n = kwargs["reranker_parameters"].top_n
+        outputs["text"] = outputs["text"][:top_n]
+    return outputs
+
+
+class TestServiceOrchestratorParams(unittest.IsolatedAsyncioTestCase):
+    def setUp(self):
+        self.s1 = opea_microservices["s1"]
+        self.s2 = opea_microservices["s2"]
+        self.s1.start()
+        self.s2.start()
+
+        ServiceOrchestrator.align_inputs = align_inputs
+        ServiceOrchestrator.align_outputs = align_outputs
+        self.service_builder = ServiceOrchestrator()
+
+        self.service_builder.add(opea_microservices["s1"]).add(opea_microservices["s2"])
+        self.service_builder.flow_to(self.s1, self.s2)
+        self.gateway = Gateway(self.service_builder, port=9898)
+
+    def tearDown(self):
+        self.s1.stop()
+        self.s2.stop()
+        self.gateway.stop()
+
+    async def test_retriever_schedule(self):
+        result_dict, _ = await self.service_builder.schedule(
+            initial_inputs={"text": "hello, ", "embedding": [1.0, 2.0, 3.0]},
+            retriever_parameters=RetrieverParms(k=8),
+            reranker_parameters=RerankerParms(top_n=20),
+        )
+        self.assertEqual(len(result_dict[self.s2.name]["text"]), 20)  # Check reranker top_n is accessed
+        self.assertTrue("8" in result_dict[self.s2.name]["text"])  # Check retriever k is accessed
+
+
+if __name__ == "__main__":
+    unittest.main()