mirror of
https://github.com/MODSetter/SurfSense.git
synced 2026-07-04 22:02:16 +02:00
feat: add support for Azure Document Intelligence in ETL pipeline
This commit is contained in:
parent
73a9c5fbd1
commit
1fa8d1220b
10 changed files with 248 additions and 5 deletions
|
|
@ -397,6 +397,10 @@ class Config:
|
|||
# LlamaCloud API Key
|
||||
LLAMA_CLOUD_API_KEY = os.getenv("LLAMA_CLOUD_API_KEY")
|
||||
|
||||
elif ETL_SERVICE == "AZURE_DI":
|
||||
AZURE_DI_ENDPOINT = os.getenv("AZURE_DI_ENDPOINT")
|
||||
AZURE_DI_KEY = os.getenv("AZURE_DI_KEY")
|
||||
|
||||
# Residential Proxy Configuration (anonymous-proxies.net)
|
||||
# Used for web crawling and YouTube transcript fetching to avoid IP bans.
|
||||
RESIDENTIAL_PROXY_USERNAME = os.getenv("RESIDENTIAL_PROXY_USERNAME")
|
||||
|
|
|
|||
|
|
@ -56,7 +56,7 @@ class EtlPipelineService:
|
|||
if not etl_service:
|
||||
raise EtlServiceUnavailableError(
|
||||
"No ETL_SERVICE configured. "
|
||||
"Set ETL_SERVICE to UNSTRUCTURED, LLAMACLOUD, or DOCLING in your .env"
|
||||
"Set ETL_SERVICE to UNSTRUCTURED, LLAMACLOUD, DOCLING, or AZURE_DI in your .env"
|
||||
)
|
||||
|
||||
ext = PurePosixPath(request.filename).suffix.lower()
|
||||
|
|
@ -80,6 +80,12 @@ class EtlPipelineService:
|
|||
content = await parse_with_llamacloud(
|
||||
request.file_path, request.estimated_pages
|
||||
)
|
||||
elif etl_service == "AZURE_DI":
|
||||
from app.etl_pipeline.parsers.azure_doc_intelligence import (
|
||||
parse_with_azure_doc_intelligence,
|
||||
)
|
||||
|
||||
content = await parse_with_azure_doc_intelligence(request.file_path)
|
||||
else:
|
||||
raise EtlServiceUnavailableError(f"Unknown ETL_SERVICE: {etl_service}")
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,93 @@
|
|||
import asyncio
|
||||
import logging
|
||||
import os
|
||||
import random
|
||||
|
||||
from app.config import config as app_config
|
||||
|
||||
MAX_RETRIES = 5
|
||||
BASE_DELAY = 10
|
||||
MAX_DELAY = 120
|
||||
|
||||
|
||||
async def parse_with_azure_doc_intelligence(file_path: str) -> str:
|
||||
from azure.ai.documentintelligence.aio import DocumentIntelligenceClient
|
||||
from azure.ai.documentintelligence.models import DocumentContentFormat
|
||||
from azure.core.credentials import AzureKeyCredential
|
||||
from azure.core.exceptions import (
|
||||
ClientAuthenticationError,
|
||||
HttpResponseError,
|
||||
ServiceRequestError,
|
||||
ServiceResponseError,
|
||||
)
|
||||
|
||||
file_size_mb = os.path.getsize(file_path) / (1024 * 1024)
|
||||
retryable_exceptions = (ServiceRequestError, ServiceResponseError)
|
||||
|
||||
last_exception = None
|
||||
attempt_errors: list[str] = []
|
||||
|
||||
for attempt in range(1, MAX_RETRIES + 1):
|
||||
try:
|
||||
client = DocumentIntelligenceClient(
|
||||
endpoint=app_config.AZURE_DI_ENDPOINT,
|
||||
credential=AzureKeyCredential(app_config.AZURE_DI_KEY),
|
||||
)
|
||||
async with client:
|
||||
with open(file_path, "rb") as f:
|
||||
poller = await client.begin_analyze_document(
|
||||
"prebuilt-read",
|
||||
body=f,
|
||||
output_content_format=DocumentContentFormat.MARKDOWN,
|
||||
)
|
||||
result = await poller.result()
|
||||
|
||||
if attempt > 1:
|
||||
logging.info(
|
||||
f"Azure Document Intelligence succeeded on attempt {attempt} "
|
||||
f"after {len(attempt_errors)} failures"
|
||||
)
|
||||
|
||||
if not result.content:
|
||||
return ""
|
||||
|
||||
return result.content
|
||||
|
||||
except ClientAuthenticationError:
|
||||
raise
|
||||
except HttpResponseError as e:
|
||||
if e.status_code and 400 <= e.status_code < 500:
|
||||
raise
|
||||
last_exception = e
|
||||
error_type = type(e).__name__
|
||||
error_msg = str(e)[:200]
|
||||
attempt_errors.append(f"Attempt {attempt}: {error_type} - {error_msg}")
|
||||
except retryable_exceptions as e:
|
||||
last_exception = e
|
||||
error_type = type(e).__name__
|
||||
error_msg = str(e)[:200]
|
||||
attempt_errors.append(f"Attempt {attempt}: {error_type} - {error_msg}")
|
||||
|
||||
if attempt < MAX_RETRIES:
|
||||
base_delay = min(BASE_DELAY * (2 ** (attempt - 1)), MAX_DELAY)
|
||||
jitter = base_delay * 0.25 * (2 * random.random() - 1)
|
||||
delay = base_delay + jitter
|
||||
|
||||
logging.warning(
|
||||
f"Azure Document Intelligence failed "
|
||||
f"(attempt {attempt}/{MAX_RETRIES}): "
|
||||
f"{attempt_errors[-1]}. File: {file_size_mb:.1f}MB. "
|
||||
f"Retrying in {delay:.0f}s..."
|
||||
)
|
||||
await asyncio.sleep(delay)
|
||||
else:
|
||||
logging.error(
|
||||
f"Azure Document Intelligence failed after {MAX_RETRIES} "
|
||||
f"attempts. File size: {file_size_mb:.1f}MB. "
|
||||
f"Errors: {'; '.join(attempt_errors)}"
|
||||
)
|
||||
|
||||
raise last_exception or RuntimeError(
|
||||
f"Azure Document Intelligence parsing failed after {MAX_RETRIES} retries. "
|
||||
f"File size: {file_size_mb:.1f}MB"
|
||||
)
|
||||
|
|
@ -93,6 +93,22 @@ UNSTRUCTURED_DOCUMENT_EXTENSIONS: frozenset[str] = frozenset(
|
|||
}
|
||||
)
|
||||
|
||||
AZURE_DI_DOCUMENT_EXTENSIONS: frozenset[str] = frozenset(
|
||||
{
|
||||
".pdf",
|
||||
".docx",
|
||||
".xlsx",
|
||||
".pptx",
|
||||
".png",
|
||||
".jpg",
|
||||
".jpeg",
|
||||
".bmp",
|
||||
".tiff",
|
||||
".tif",
|
||||
".heif",
|
||||
}
|
||||
)
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Union (used by classify_file for routing) + service lookup
|
||||
# ---------------------------------------------------------------------------
|
||||
|
|
@ -101,12 +117,14 @@ DOCUMENT_EXTENSIONS: frozenset[str] = (
|
|||
DOCLING_DOCUMENT_EXTENSIONS
|
||||
| LLAMAPARSE_DOCUMENT_EXTENSIONS
|
||||
| UNSTRUCTURED_DOCUMENT_EXTENSIONS
|
||||
| AZURE_DI_DOCUMENT_EXTENSIONS
|
||||
)
|
||||
|
||||
_SERVICE_MAP: dict[str, frozenset[str]] = {
|
||||
"DOCLING": DOCLING_DOCUMENT_EXTENSIONS,
|
||||
"LLAMACLOUD": LLAMAPARSE_DOCUMENT_EXTENSIONS,
|
||||
"UNSTRUCTURED": UNSTRUCTURED_DOCUMENT_EXTENSIONS,
|
||||
"AZURE_DI": AZURE_DI_DOCUMENT_EXTENSIONS,
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue