ws-bom-robot-app 0.0.37__py3-none-any.whl → 0.0.103__py3-none-any.whl

ws_bom_robot_app/llm/vector_store/generator.py

@@ -1,137 +1,137 @@
-import os, gc, shutil, logging, traceback
-import asyncio, aiofiles, aiofiles.os
-from fastapi import HTTPException
-from fastapi.responses import StreamingResponse
-from langchain_core.documents import Document
-from ws_bom_robot_app.llm.vector_store.loader.base import Loader
-from ws_bom_robot_app.llm.models.api import RulesRequest, KbRequest, VectorDbResponse
-from ws_bom_robot_app.llm.vector_store.integration.manager import IntegrationManager
-from ws_bom_robot_app.llm.vector_store.db.manager import VectorDbManager
-from ws_bom_robot_app.config import config
-from ws_bom_robot_app.llm.models.kb import load_endpoints
-from ws_bom_robot_app.llm.utils.download import download_files
-
-async def _cleanup_directory(directory_path: str):
-  if os.path.exists(directory_path):
-    await asyncio.to_thread(shutil.rmtree, directory_path)
-
-#@timer
-async def rules(rq: RulesRequest) -> VectorDbResponse:
-  _config = rq.config()
-  db_name = rq.out_name()
-  store_path = os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_store, db_name)
-  try:
-    await VectorDbManager.get_strategy(rq.vector_type).create(rq.embeddings(),[Document(page_content=rule, metadata={"source": "rules"}) for rule in rq.rules], store_path) #type: ignore
-    db_file_path = shutil.make_archive(os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_out, db_name), "zip", store_path)
-    return VectorDbResponse(file = os.path.basename(db_file_path), vector_type=rq.vector_type)
-  except Exception as e:
-    try:
-      await _cleanup_directory(store_path)
-    finally:
-      return VectorDbResponse(success = False, error = str(e))
-  finally:
-    gc.collect()
-
-#@atimer
-async def kb(rq: KbRequest) -> VectorDbResponse:
-  os.environ['MPLCONFIGDIR'] = './tmp/.matplotlib'
-  _config = rq.config()
-  db_name = rq.out_name()
-  src_path = os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_src)
-  working_path = os.path.join(src_path, db_name)
-
-  if all([not rq.files,not rq.endpoints,not rq.integrations]):
-    return VectorDbResponse(success = False, error = "No files, endpoints or integrations provided")
-  else:
-    await aiofiles.os.makedirs(src_path, exist_ok=True)
-    await aiofiles.os.makedirs(working_path, exist_ok=True)
-
-  documents: list[Document] = []
-  # Download/copy all files
-  if rq.files:
-    try:
-      loaders = Loader(working_path)
-      filter_file_extensions = loaders.managed_file_extensions()
-      files_to_download = [file for file in rq.files if not os.path.exists(os.path.join(src_path, os.path.basename(file)))]
-      if files_to_download:
-        await download_files(
-          [f"{_config.robot_cms_host}/{_config.robot_cms_kb_folder}/{os.path.basename(file)}" for file in files_to_download if any([file.endswith(ext) for ext in filter_file_extensions])],
-          src_path, authorization=_config.robot_cms_auth)
-      # copy files to working tmp folder
-      for file in rq.files:
-        async with aiofiles.open(os.path.join(src_path, os.path.basename(file)), 'rb') as src_file:
-          async with aiofiles.open(os.path.join(working_path, os.path.basename(file)), 'wb') as dest_file:
-            await dest_file.write(await src_file.read())
-      #load files
-      try:
-        documents.extend(await loaders.load())
-      except Exception as e:
-        tb = traceback.format_exc()
-        _error = f"File loader failure: {e} | {tb}"
-        logging.warning(_error)
-        return VectorDbResponse(success = False, error = _error)
-    except Exception as e:
-      await _cleanup_directory(working_path)
-      return VectorDbResponse(success = False, error = f"Failed to download file {e}")
-
-  if rq.endpoints:
-    try:
-      documents.extend(await load_endpoints(rq.endpoints, working_path))
-    except Exception as e:
-      await _cleanup_directory(working_path)
-      tb = traceback.format_exc()
-      _error = f"Endpoint failure: {e} | {tb}"
-      logging.warning(_error)
-      return VectorDbResponse(success = False, error = _error)
-
-  if rq.integrations:
-    tasks = []
-    for integration in rq.integrations:
-      tasks.append(
-        IntegrationManager
-        .get_strategy(integration.type.lower(), working_path, integration.__pydantic_extra__) #type: ignore
-        .load()
-      )
-    try:
-      integration_documents = await asyncio.gather(*tasks)
-      for docs in integration_documents:
-        documents.extend(docs)
-    except Exception as e:
-      await _cleanup_directory(working_path)
-      tb = traceback.format_exc()
-      _error = f"Integration failure: {e} | {tb}"
-      logging.warning(_error)
-      return VectorDbResponse(success=False, error=_error)
-
-  #cleanup
-  await _cleanup_directory(working_path)
-
-  if documents and len(documents) > 0:
-    try:
-      store_path = os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_store, db_name)
-      db_file_path = await aiofiles.os.wrap(shutil.make_archive)(
-          os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_out, db_name),
-          "zip",
-          await VectorDbManager.get_strategy(rq.vector_type).create(rq.embeddings(), documents, store_path, return_folder_path=True)
-      )
-      return VectorDbResponse(file = os.path.basename(db_file_path), vector_type=rq.vector_type)
-    except Exception as e:
-      await _cleanup_directory(store_path)
-      return VectorDbResponse(success = False, error = str(e))
-    finally:
-      del documents
-      gc.collect()
-  else:
-    _error = "No documents found in the knowledgebase folder"
-    logging.warning(_error)
-    return VectorDbResponse(success = False, error = _error)
-
-async def kb_stream_file(filename: str):
-    file_path = os.path.join(config.robot_data_folder, config.robot_data_db_folder, config.robot_data_db_folder_out, filename)
-    if not os.path.isfile(file_path):
-        raise HTTPException(status_code=404, detail="File not found")
-    def iter_file():
-        with open(file_path, mode="rb") as file:
-            while chunk := file.read(1024*8):
-                yield chunk
-    return StreamingResponse(iter_file(), media_type="application/octet-stream", headers={"Content-Disposition": f"attachment; filename={filename}"})
+import os, gc, shutil, logging, traceback
+import asyncio, aiofiles, aiofiles.os
+from fastapi import HTTPException
+from fastapi.responses import StreamingResponse
+from langchain_core.documents import Document
+from ws_bom_robot_app.llm.vector_store.loader.base import Loader
+from ws_bom_robot_app.llm.models.api import RulesRequest, KbRequest, VectorDbResponse
+from ws_bom_robot_app.llm.vector_store.integration.manager import IntegrationManager
+from ws_bom_robot_app.llm.vector_store.db.manager import VectorDbManager
+from ws_bom_robot_app.config import config
+from ws_bom_robot_app.llm.models.kb import load_endpoints
+from ws_bom_robot_app.llm.utils.download import download_files
+
+async def _cleanup_directory(directory_path: str):
+  if os.path.exists(directory_path):
+    await asyncio.to_thread(shutil.rmtree, directory_path)
+
+#@timer
+async def rules(rq: RulesRequest) -> VectorDbResponse:
+  _config = rq.config()
+  db_name = rq.out_name()
+  store_path = os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_store, db_name)
+  try:
+    await VectorDbManager.get_strategy(rq.vector_type).create(rq.embeddings(),[Document(page_content=rule, metadata={"source": "rules"}) for rule in rq.rules], store_path) #type: ignore
+    db_file_path = shutil.make_archive(os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_out, db_name), "zip", store_path)
+    return VectorDbResponse(file = os.path.basename(db_file_path), vector_type=rq.vector_type)
+  except Exception as e:
+    try:
+      await _cleanup_directory(store_path)
+    finally:
+      return VectorDbResponse(success = False, error = str(e))
+  finally:
+    gc.collect()
+
+#@atimer
+async def kb(rq: KbRequest) -> VectorDbResponse:
+  os.environ['MPLCONFIGDIR'] = './tmp/.matplotlib'
+  _config = rq.config()
+  db_name = rq.out_name()
+  src_path = os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_src)
+  working_path = os.path.join(src_path, db_name)
+
+  if all([not rq.files,not rq.endpoints,not rq.integrations]):
+    return VectorDbResponse(success = False, error = "No files, endpoints or integrations provided")
+  else:
+    await aiofiles.os.makedirs(src_path, exist_ok=True)
+    await aiofiles.os.makedirs(working_path, exist_ok=True)
+
+  documents: list[Document] = []
+  # Download/copy all files
+  if rq.files:
+    try:
+      loaders = Loader(working_path)
+      filter_file_extensions = loaders.managed_file_extensions()
+      files_to_download = [file for file in rq.files if not os.path.exists(os.path.join(src_path, os.path.basename(file)))]
+      if files_to_download:
+        await download_files(
+          [f"{_config.robot_cms_host}/{_config.robot_cms_kb_folder}/{os.path.basename(file)}" for file in files_to_download if any([file.endswith(ext) for ext in filter_file_extensions])],
+          src_path, authorization=_config.robot_cms_auth)
+      # copy files to working tmp folder
+      for file in rq.files:
+        async with aiofiles.open(os.path.join(src_path, os.path.basename(file)), 'rb') as src_file:
+          async with aiofiles.open(os.path.join(working_path, os.path.basename(file)), 'wb') as dest_file:
+            await dest_file.write(await src_file.read())
+      #load files
+      try:
+        documents.extend(await loaders.load())
+      except Exception as e:
+        tb = traceback.format_exc()
+        _error = f"File loader failure: {e} | {tb}"
+        logging.warning(_error)
+        return VectorDbResponse(success = False, error = _error)
+    except Exception as e:
+      await _cleanup_directory(working_path)
+      return VectorDbResponse(success = False, error = f"Failed to download file {e}")
+
+  if rq.endpoints:
+    try:
+      documents.extend(await load_endpoints(rq.endpoints, working_path))
+    except Exception as e:
+      await _cleanup_directory(working_path)
+      tb = traceback.format_exc()
+      _error = f"Endpoint failure: {e} | {tb}"
+      logging.warning(_error)
+      return VectorDbResponse(success = False, error = _error)
+
+  if rq.integrations:
+    tasks = []
+    for integration in rq.integrations:
+      tasks.append(
+        IntegrationManager
+        .get_strategy(integration.type.lower(), working_path, integration.__pydantic_extra__) #type: ignore
+        .load()
+      )
+    try:
+      integration_documents = await asyncio.gather(*tasks)
+      for docs in integration_documents:
+        documents.extend(docs)
+    except Exception as e:
+      await _cleanup_directory(working_path)
+      tb = traceback.format_exc()
+      _error = f"Integration failure: {e} | {tb}"
+      logging.warning(_error)
+      return VectorDbResponse(success=False, error=_error)
+
+  #cleanup
+  await _cleanup_directory(working_path)
+
+  if documents and len(documents) > 0:
+    try:
+      store_path = os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_store, db_name)
+      db_file_path = await aiofiles.os.wrap(shutil.make_archive)(
+          os.path.join(_config.robot_data_folder, _config.robot_data_db_folder, _config.robot_data_db_folder_out, db_name),
+          "zip",
+          await VectorDbManager.get_strategy(rq.vector_type).create(rq.embeddings(), documents, store_path, return_folder_path=True)
+      )
+      return VectorDbResponse(file = os.path.basename(db_file_path), vector_type=rq.vector_type)
+    except Exception as e:
+      await _cleanup_directory(store_path)
+      return VectorDbResponse(success = False, error = str(e))
+    finally:
+      del documents
+      gc.collect()
+  else:
+    _error = "No documents found in the knowledgebase folder"
+    logging.warning(_error)
+    return VectorDbResponse(success = False, error = _error)
+
+async def kb_stream_file(filename: str):
+    file_path = os.path.join(config.robot_data_folder, config.robot_data_db_folder, config.robot_data_db_folder_out, filename)
+    if not os.path.isfile(file_path):
+        raise HTTPException(status_code=404, detail="File not found")
+    def iter_file():
+        with open(file_path, mode="rb") as file:
+            while chunk := file.read(1024*8):
+                yield chunk
+    return StreamingResponse(iter_file(), media_type="application/octet-stream", headers={"Content-Disposition": f"attachment; filename={filename}"})

ws_bom_robot_app/llm/vector_store/integration/api.py

@@ -0,0 +1,216 @@
+import asyncio, logging, aiohttp
+from ws_bom_robot_app.llm.vector_store.integration.base import IntegrationStrategy
+from langchain_core.documents import Document
+from ws_bom_robot_app.llm.vector_store.loader.base import Loader
+from typing import List, Union, Optional, Dict, Any, Literal
+from pydantic import BaseModel, Field, AliasChoices, field_validator
+import json
+import os
+
+
+class AuthConfig(BaseModel):
+  """
+  Configuration for API authentication.
+
+  Attributes:
+    type: Type of authentication (bearer, basic, api_key, custom, none)
+    token: Bearer token or API key value
+    username: Username for basic auth
+    password: Password for basic auth
+    header_name: Custom header name for API key
+    prefix: Prefix for the auth value (e.g., 'Bearer', 'Token')
+  """
+  type: Literal["bearer", "basic", "api_key", "custom", "none"] = Field(default="none")
+  token: Optional[str] = Field(default=None)
+  username: Optional[str] = Field(default=None)
+  password: Optional[str] = Field(default=None)
+  header_name: Optional[str] = Field(default=None, validation_alias=AliasChoices("headerName", "header_name"))
+  prefix: Optional[str] = Field(default=None)
+
+
+class ApiParams(BaseModel):
+  """
+  Generic API Integration Parameters.
+
+  Attributes:
+    url: The base URL of the API endpoint
+    method: HTTP method (GET, POST, PUT, DELETE, PATCH)
+    headers: Custom headers to include in the request
+    params: Query parameters for the request
+    body: Request body for POST/PUT/PATCH requests
+    auth: Authentication configuration
+    response_data_path: JSON path to extract data from response (e.g., 'data.items', 'results')
+    max_retries: Maximum number of retry attempts for failed requests
+    retry_delay: Base delay in seconds between retries (uses exponential backoff)
+    timeout: Request timeout in seconds
+  """
+  url: str = Field(validation_alias=AliasChoices("url", "endpoint"))
+  method: Literal["GET", "POST", "PUT", "DELETE", "PATCH"] = Field(default="GET")
+  headers: Optional[Dict[str, str]] = Field(default_factory=dict)
+  params: Optional[Dict[str, Any]] = Field(default_factory=dict)
+  body: Optional[Union[Dict[str, Any], str]] = Field(default=None)
+  auth: Optional[AuthConfig] = Field(default_factory=lambda: AuthConfig())
+  response_data_path: Optional[str] = Field(default=None, validation_alias=AliasChoices("responseDataPath", "response_data_path"))
+  max_retries: int = Field(default=5, validation_alias=AliasChoices("maxRetries", "max_retries"))
+  retry_delay: float = Field(default=1.0, validation_alias=AliasChoices("retryDelay", "retry_delay"))
+  timeout: int = Field(default=30)
+
+  @field_validator('auth', mode='before')
+  @classmethod
+  def parse_auth(cls, v):
+    """Parse auth config from dict if needed"""
+    if isinstance(v, dict):
+      return AuthConfig(**v)
+    return v or AuthConfig()
+
+
+class Api(IntegrationStrategy):
+  """
+  Generic API Integration that supports:
+  - Multiple HTTP methods (GET, POST, PUT, DELETE, PATCH)
+  - Various authentication types (Bearer, Basic, API Key, Custom)
+  - Custom headers and parameters
+  - Automatic retry with exponential backoff
+  - Flexible response data extraction
+  """
+
+  def __init__(self, knowledgebase_path: str, data: dict[str, Union[str, int, list]]):
+    super().__init__(knowledgebase_path, data)
+    self.__data = ApiParams.model_validate(self.data)
+
+  def working_subdirectory(self) -> str:
+    return 'api_integration'
+
+  async def run(self) -> None:
+    """Fetch data from the API and save to JSON file"""
+    _data = await self.__fetch_data()
+    json_file_path = os.path.join(self.working_directory, 'api_data.json')
+    with open(json_file_path, 'w', encoding='utf-8') as f:
+      json.dump(_data, f, ensure_ascii=False, indent=2)
+    logging.info(f"Saved {len(_data) if isinstance(_data, list) else 1} items to {json_file_path}")
+
+  async def load(self) -> list[Document]:
+    """Load data from API and convert to documents"""
+    await self.run()
+    await asyncio.sleep(1)
+    return await Loader(self.working_directory).load()
+
+  def __prepare_headers(self) -> Dict[str, str]:
+    """Prepare request headers with authentication"""
+    headers = self.__data.headers.copy() if self.__data.headers else {}
+
+    # Add Content-Type if not present
+    if 'Content-Type' not in headers and self.__data.method in ["POST", "PUT", "PATCH"]:
+      headers['Content-Type'] = 'application/json'
+
+    # Add authentication
+    auth = self.__data.auth
+    if auth.type == "bearer":
+      prefix = auth.prefix or "Bearer"
+      headers['Authorization'] = f"{prefix} {auth.token}"
+    elif auth.type == "basic":
+      import base64
+      credentials = f"{auth.username}:{auth.password}"
+      encoded = base64.b64encode(credentials.encode()).decode()
+      headers['Authorization'] = f"Basic {encoded}"
+    elif auth.type == "api_key" and auth.header_name:
+      prefix = f"{auth.prefix} " if auth.prefix else ""
+      headers[auth.header_name] = f"{prefix}{auth.token}"
+
+    return headers
+
+  def __get_nested_value(self, data: Any, path: Optional[str]) -> Any:
+    """Extract nested value from data using dot notation path"""
+    if not path:
+      return data
+
+    keys = path.split('.')
+    current = data
+    for key in keys:
+      if isinstance(current, dict):
+        current = current.get(key)
+      elif isinstance(current, list) and key.isdigit():
+        current = current[int(key)]
+      else:
+        return None
+
+      if current is None:
+        return None
+
+    return current
+
+  async def __make_request(
+    self,
+    url: str,
+    headers: Dict[str, str],
+    params: Optional[Dict[str, Any]] = None
+  ) -> Dict[str, Any]:
+    """Make HTTP request with retry logic"""
+    retry_count = 0
+
+    while retry_count <= self.__data.max_retries:
+      try:
+        timeout = aiohttp.ClientTimeout(total=self.__data.timeout)
+
+        async with aiohttp.ClientSession(timeout=timeout) as session:
+          request_kwargs = {
+            "headers": headers,
+            "params": params or self.__data.params
+          }
+
+          # Add body for POST/PUT/PATCH
+          if self.__data.method in ["POST", "PUT", "PATCH"] and self.__data.body:
+            if isinstance(self.__data.body, dict):
+              request_kwargs["json"] = self.__data.body
+            else:
+              request_kwargs["data"] = self.__data.body
+
+          async with session.request(
+            self.__data.method,
+            url,
+            **request_kwargs
+          ) as response:
+            # Check response status
+            if response.status == 429:  # Rate limit
+              retry_count += 1
+              if retry_count > self.__data.max_retries:
+                raise Exception("Rate limit exceeded. Maximum retries reached.")
+
+              wait_time = self.__data.retry_delay * (2 ** retry_count)
+              logging.warning(f"Rate limited. Waiting {wait_time}s (Attempt {retry_count}/{self.__data.max_retries})")
+              await asyncio.sleep(wait_time)
+              continue
+
+            response.raise_for_status()
+
+            # Parse response
+            try:
+              data = await response.json()
+              return data
+            except aiohttp.ContentTypeError:
+              text = await response.text()
+              logging.warning(f"Non-JSON response received: {text[:200]}")
+              return {"text": text}
+
+      except aiohttp.ClientError as e:
+        retry_count += 1
+        if retry_count > self.__data.max_retries:
+          raise Exception(f"Request failed after {self.__data.max_retries} retries: {e}")
+
+        wait_time = self.__data.retry_delay * (2 ** retry_count)
+        logging.warning(f"Request error: {e}. Retrying in {wait_time}s...")
+        await asyncio.sleep(wait_time)
+        continue
+
+    raise Exception("Maximum retries exceeded")
+
+  async def __fetch_data(self) -> Any:
+    """Fetch data from API"""
+    headers = self.__prepare_headers()
+    response = await self.__make_request(self.__data.url, headers)
+
+    # Extract data from response using path if specified
+    data = self.__get_nested_value(response, self.__data.response_data_path)
+    result = data if data is not None else response
+
+    return result

ws_bom_robot_app/llm/vector_store/integration/azure.py

@@ -1,6 +1,6 @@
 import asyncio
 from ws_bom_robot_app.llm.vector_store.integration.base import IntegrationStrategy, UnstructuredIngest
-from unstructured_ingest.v2.processes.connectors.fsspec.azure import AzureConnectionConfig, AzureAccessConfig, AzureDownloaderConfig, AzureIndexerConfig
+from unstructured_ingest.processes.connectors.fsspec.azure import AzureConnectionConfig, AzureAccessConfig, AzureDownloaderConfig, AzureIndexerConfig
 from langchain_core.documents import Document
 from ws_bom_robot_app.llm.vector_store.loader.base import Loader
 from typing import Union, Optional

ws_bom_robot_app/llm/vector_store/integration/base.py

@@ -1,10 +1,17 @@
-import os
+import os, copy
+from random import random
 from langchain_core.documents import Document
 from abc import ABC, abstractmethod
-from unstructured_ingest.v2.interfaces import ProcessorConfig
-from unstructured_ingest.v2.pipeline.pipeline import Pipeline, PartitionerConfig, FiltererConfig
+from unstructured_ingest.interfaces import ProcessorConfig
+from unstructured_ingest.pipeline.pipeline import (
+  Pipeline,
+  PartitionerConfig,
+  FiltererConfig
+)
+from unstructured_ingest.processes.connector_registry import source_registry
 from typing import Union
 from ws_bom_robot_app.llm.utils.secrets import Secrets
+from ws_bom_robot_app.config import config
 
 class IntegrationStrategy(ABC):
   @classmethod
@@ -32,23 +39,59 @@ class IntegrationStrategy(ABC):
     pass
 
 class UnstructuredIngest():
+  _PIPELINE: Pipeline = None
   def __init__(self, working_directory: str):
     self.working_directory = working_directory
-  def pipeline(self,indexer,downloader,connection,extension: list[str] = None) -> Pipeline:
-    return Pipeline.from_configs(
-      context=ProcessorConfig(
+    self._runtime_options = config.runtime_options()
+  def pipeline(self,indexer_config,downloader_config,connection_config,extension: list[str] = None) -> Pipeline:
+    def _default_processor_config() -> ProcessorConfig:
+      return ProcessorConfig(
         reprocess=False,
         verbose=False,
         tqdm=False,
-        num_processes=2,
+        num_processes=config.robot_ingest_max_threads, #safe choice to 1, avoid potential process-related issues with Docker
+        disable_parallelism=False,
         preserve_downloads=True,
         download_only=True,
-        raise_on_error=False
-      ),
-      indexer_config=indexer,
-      downloader_config=downloader,
-      source_connection_config=connection,
-      partitioner_config=PartitionerConfig(),
-      filterer_config=FiltererConfig(file_glob=[f"**/*{ext}" for ext in extension] if extension else None)
-    )
+        raise_on_error=False,
+        iter_delete=True,
+        delete_cache=False #already managed by the generator task
+      )
+    def _init_pipeline() -> Pipeline:
+      return Pipeline.from_configs(
+        context=_default_processor_config(),
+        indexer_config=indexer_config,
+        downloader_config=downloader_config,
+        source_connection_config=connection_config,
+        partitioner_config=PartitionerConfig(),
+        filterer_config=FiltererConfig(file_glob=[f"**/*{ext}" for ext in extension] if extension else None)
+      )
+    def _instance_pipeline() -> Pipeline:
+        from unstructured_ingest.pipeline.steps.index import  IndexStep
+        from unstructured_ingest.pipeline.steps.download import  DownloadStep
+        from unstructured_ingest.pipeline.steps.filter import Filterer, FilterStep
+        _context = _default_processor_config()
+        source_entry = {
+                    k: v
+                    for k, v in source_registry.items()
+                    if type(indexer_config) is v.indexer_config
+                    and type(downloader_config) is v.downloader_config
+                    and type(connection_config) is v.connection_config
+                }
+        source = list(source_entry.values())[0]
+        _pipeline = copy.deepcopy(UnstructuredIngest._PIPELINE)
+        _pipeline.context = _context
+        _pipeline.context.work_dir = f"{self.working_directory}_unstructured" # use sibling directory, cleaned up by the generator task
+        _pipeline.indexer_step = IndexStep(process=source.indexer(index_config=indexer_config, connection_config=connection_config), context=_context)
+        _pipeline.downloader_step = DownloadStep(process=source.downloader(download_config=downloader_config, connection_config=connection_config), context=_context)
+        _pipeline.filter_step = FilterStep(process=Filterer(config=FiltererConfig(file_glob=[f"**/*{ext}" for ext in extension] if extension else None)), context=_context) if extension else None
+        return _pipeline
 
+    if not UnstructuredIngest._PIPELINE:
+      import random
+      import time
+      time.sleep(random.uniform(0.2, 1))
+      if not UnstructuredIngest._PIPELINE:
+        UnstructuredIngest._PIPELINE = _init_pipeline()
+
+    return _instance_pipeline()