Make waitress thread count configurable.

This should fix most scaling problems without needing to start
more server instances. By default, waitress maintains at most
4 threads. This works fine if the database is small (sub 100k posts)
but causes a large Task queue depth to occur if the database is larger.

Letting users increase the amount of threads means that one server instance
is able to handle more requests without locking up the rest of the site.

This adds a new environment variable to .env, THREADS, which can be used to
configure the amount of threads to start and is by default set to 4
(the default amount used by waitress).

doc/example.env

@@ -10,6 +10,12 @@ BUILD_INFO=latest
 # otherwise the port specified here will be publicly accessible
 PORT=8080
 
+# How many waitress threads to start
+# 4 is the default amount of threads. If you experience performance
+# degradation with a large number of posts, increasing this may
+# improve performance, since waitress is most likely clogging up with Tasks.
+THREADS=4
+
 # URL base to run szurubooru under
 # See "Additional Features" section in INSTALL.md
 BASE_URL=/

docker-compose.yml

@@ -21,6 +21,7 @@ services:
       #POSTGRES_DB: defaults to same as POSTGRES_USER
       #POSTGRES_PORT: 5432
       #LOG_SQL: 0 (1 for verbose SQL logs)
+      THREADS:
     volumes:
       - "${MOUNT_DATA}:/data"
       - "./server/config.yaml:/opt/app/config.yaml"

server/Dockerfile

@@ -83,6 +83,9 @@ ARG PORT=6666
 ENV PORT=${PORT}
 EXPOSE ${PORT}
 
+ARG THREADS=4
+ENV THREADS=${THREADS}
+
 VOLUME ["/data/"]
 
 ARG DOCKER_REPO

server/docker-start.sh

@@ -4,5 +4,5 @@ cd /opt/app
 
 alembic upgrade head
 
-echo "Starting szurubooru API on port ${PORT}"
-exec waitress-serve-3 --port ${PORT} szurubooru.facade:app
+echo "Starting szurubooru API on port ${PORT} - Running on ${THREADS} threads"
+exec waitress-serve-3 --port ${PORT} --threads ${THREADS} szurubooru.facade:app