LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Python ai-agentsdata-engineeringdocument-aigenerative-aiidpjson-extractionllmmcp-serverocrpdf-extractionprompt-engineeringstructured-output
UN-3996 [FIX] Stop WebSockets exhausting the Gunicorn thread pool (#2241)
Kirtiman Mishra3be7483bbcd1
diff --git a/backend/entrypoint.sh b/backend/entrypoint.shindex 5d44345fd..20441c58f 100755--- a/backend/entrypoint.sh+++ b/backend/entrypoint.sh@@ -29,11 +29,15 @@ if [ "$migrate" = true ]; then.venv/bin/python manage.py migratefi-# Configure Gunicorn based on --dev flag+# Configure Gunicorn based on --dev flag.+# Threads must exceed concurrent browser tabs — each open Socket.IO WebSocket+# holds one thread for its lifetime. The pool spawns threads lazily, so a high+# ceiling is free at idle.gunicorn_args=(--bind 0.0.0.0:8000- --workers 2- --threads 2+ --workers "${GUNICORN_WORKERS:-2}"+ --threads "${GUNICORN_THREADS:-512}"+ --worker-class gthread--log-level debug--timeout 600--access-logfile -diff --git a/backend/sample.env b/backend/sample.envindex b57101813..700422d1a 100644--- a/backend/sample.env+++ b/backend/sample.env@@ -7,6 +7,11 @@ CSRF_COOKIE_SECURE=False# Default log levelDEFAULT_LOG_LEVEL="INFO"+# Gunicorn tuning. GUNICORN_THREADS must exceed concurrent browser tabs —+# each open Socket.IO WebSocket holds one thread for its lifetime.+GUNICORN_WORKERS=2+GUNICORN_THREADS=512+# CommonPATH_PREFIX="api/v1"