logging.getLogger("pipelines").setLevel(LOG_LEVEL)
logger = logging.getLogger("pipelines")
from pydantic import BaseConfig
BaseConfig.arbitrary_types_allowed = True
router = APIRouter()
PIPELINE = Pipeline.load_from_yaml(Path(PIPELINE_YAML_PATH),
pipeline_name=QUERY_PIPELINE_NAME)
DOCUMENT_STORE = PIPELINE.get_document_store()
logging.info(f"Loaded pipeline nodes: {PIPELINE.graph.nodes.keys()}")
concurrency_limiter = RequestLimiter(CONCURRENT_REQUEST_PER_WORKER)
logging.info("Concurrent requests per worker: {CONCURRENT_REQUEST_PER_WORKER}")
@router.get("/initialized")
def check_status():
"""
This endpoint can be used during startup to understand if the
server is ready to take any requests, or is still loading.
The recommended approach is to call this endpoint with a short timeout,
like 500ms, and in case of no reply, consider the server busy.
"""
return True
meta: Optional[Dict[str, Optional[str]]]
class AnswersToIndividualQuestion(BaseModel):
question: str
answers: List[Optional[Answer]]
class Answers(BaseModel):
results: List[AnswersToIndividualQuestion]
#############################################
# Endpoints
#############################################
doc_qa_limiter = RequestLimiter(CONCURRENT_REQUEST_PER_WORKER)
@router.post("/models/{model_id}/doc-qa",
response_model=Answers,
response_model_exclude_unset=True)
def doc_qa(model_id: int, request: Question):
with doc_qa_limiter.run():
start_time = time.time()
finder = FINDERS.get(model_id, None)
if not finder:
raise HTTPException(
status_code=404,
detail=
f"Couldn't get Finder with ID {model_id}. Available IDs: {list(FINDERS.keys())}"
)
offset_end: int
offset_start_in_doc: Optional[int]
offset_end_in_doc: Optional[int]
document_id: Optional[str] = None
meta: Optional[Dict[str, str]]
class Response(BaseModel):
query: str
answers: List[Answer]
PIPELINE = Pipeline.load_from_yaml(Path(PIPELINE_YAML_PATH),
pipeline_name=QUERY_PIPELINE_NAME)
logger.info(f"Loaded pipeline nodes: {PIPELINE.graph.nodes.keys()}")
concurrency_limiter = RequestLimiter(4)
@router.post("/query", response_model=Response)
def query(request: Request):
with concurrency_limiter.run():
result = _process_request(PIPELINE, request)
return result
def _process_request(pipeline, request) -> Response:
start_time = time.time()
filters = {}
if request.filters:
# put filter values into a list and remove filters with null value