Switch to gevent workers to better handle IO-bound workloads

.github/workflows/tests.yml

@@ -121,7 +121,7 @@ jobs:
         run: >
           docker exec -e SELENIUM_DEBUG=1 -e SELENIUM_TIMEOUT=30 cl-django
           python /opt/courtlistener/manage.py test
-          cl --verbosity=2 ${{ matrix.tag_flags }} --parallel
+          cl --verbosity=2 ${{ matrix.tag_flags }} --parallel=1
       - name: cp selenium results from docker to host
         if: failure()
         run: |

docker/django/Dockerfile

@@ -113,12 +113,12 @@ FROM python-base as web-prod
 
 CMD gunicorn cl_wsgi:application \
     --chdir /opt/courtlistener/docker/django/wsgi-configs/ \
+    --config /opt/courtlistener/docker/django/wsgi-configs/gunicorn.conf.py \
     --user www-data \
     --group www-data \
     # Set high number of workers. Docs recommend 2-4× core count`
     --workers ${NUM_WORKERS:-48} \
-    --worker-class gthread \
-    --threads 10 \
+    --worker-class gevent \
     # Allow longer queries to solr.
     --limit-request-line 6000 \
     # Reset each worker once in a while

docker/django/wsgi-configs/gunicorn.conf.py

@@ -0,0 +1,5 @@
+from psycogreen.gevent import patch_psycopg
+
+
+def post_fork(server, worker):
+    patch_psycopg()

manage.py

@@ -1,4 +1,10 @@
 #!/usr/bin/env python
+from gevent import monkey
+
+monkey.patch_all()
+from psycogreen.gevent import patch_psycopg
+
+patch_psycopg()
 import os
 import sys