feat: improve idle memory usage (#1758)

* health check as python script

* install crfpp model via python

* drop curl from finale container

* use uvicorn by default w/ gunicorn as opt in

* recommend setting mem limit for container

Dockerfile

@@ -95,10 +95,9 @@ ENV TESTING=false
 ARG COMMIT
 ENV GIT_COMMIT_HASH=$COMMIT
 
-# curl for used by healthcheck
 RUN apt-get update \
     && apt-get install --no-install-recommends -y \
-    curl gosu \
+    gosu \
     tesseract-ocr-all \
     && apt-get autoremove \
     && rm -rf /var/lib/apt/lists/*
@@ -107,9 +106,6 @@ RUN apt-get update \
 COPY --from=builder-base $POETRY_HOME $POETRY_HOME
 COPY --from=builder-base $PYSETUP_PATH $PYSETUP_PATH
 
-# copy CRF++ Binary from crfpp
-ENV CRF_MODEL_URL=https://github.com/mealie-recipes/nlp-model/releases/download/v1.0.0/model.crfmodel
-
 ENV LD_LIBRARY_PATH=/usr/local/lib
 COPY --from=crfpp /usr/local/lib/ /usr/local/lib
 COPY --from=crfpp /usr/local/bin/crf_learn /usr/local/bin/crf_learn
@@ -130,14 +126,14 @@ RUN . $VENV_PATH/bin/activate && poetry install -E pgsql --no-dev
 WORKDIR /
 
 # Grab CRF++ Model Release
-RUN curl -L0 $CRF_MODEL_URL --output $MEALIE_HOME/mealie/services/parser_services/crfpp/model.crfmodel
+RUN python $MEALIE_HOME/mealie/scripts/install_model.py
 
 VOLUME [ "$MEALIE_HOME/data/" ]
 ENV APP_PORT=9000
 
 EXPOSE ${APP_PORT}
 
-HEALTHCHECK CMD curl -f http://localhost:${APP_PORT}/docs || exit 1
+HEALTHCHECK CMD python $MEALIE_HOME/mealie/scripts/healthcheck.py || exit 1
 
 RUN chmod +x $MEALIE_HOME/mealie/run.sh
 ENTRYPOINT $MEALIE_HOME/mealie/run.sh

docker-compose.yml

@@ -3,6 +3,10 @@ services:
   mealie-frontend:
     container_name: mealie-frontend
     image: mealie-frontend:dev
+    deploy:
+      resources:
+        limits:
+          memory: 500M
     build:
       context: ./frontend
       dockerfile: Dockerfile
@@ -34,6 +38,10 @@ services:
       - THEME_DARK_ERROR=#EF5350
   mealie:
     container_name: mealie-api
+    deploy:
+      resources:
+        limits:
+          memory: 1000M
     build:
       context: ./
       target: production
@@ -57,6 +65,7 @@ services:
 
       # =====================================
       # Web Concurrency
+      WEB_GUNICORN: true
       WORKERS_PER_CORE: 0.5
       MAX_WORKERS: 1
       WEB_CONCURRENCY: 1

docs/docs/documentation/getting-started/installation/backend-config.md

@@ -54,6 +54,7 @@ Changing the webworker settings may cause unforeseen memory leak issues with Mea
 
 | Variables        | Default | Description                                                                                                                       |
 | ---------------- | :-----: | --------------------------------------------------------------------------------------------------------------------------------- |
+| WEB_GUNICORN     |  false  | Enables Gunicorn to manage Uvicorn web for multiple works                                                                         |
 | WORKERS_PER_CORE |    1    | Set the number of workers to the number of CPU cores multiplied by this value (Value \* CPUs). More info [here][workers_per_core] |
 | MAX_WORKERS      |    1    | Set the maximum number of workers to use. Default is not set meaning unlimited. More info [here][max_workers]                     |
 | WEB_CONCURRENCY  |    1    | Override the automatic definition of number of workers. More info [here][web_concurrency]                                         |
@@ -68,5 +69,5 @@ Changing the webworker settings may cause unforeseen memory leak issues with Mea
 | LDAP_TLS_INSECURE   |  False  | Do not verify server certificate when using secure LDAP                                                            |
 | LDAP_TLS_CACERTFILE |  None   | File path to Certificate Authority used to verify server certificate (e.g. `/path/to/ca.crt`)                      |
 | LDAP_BIND_TEMPLATE  |  None   | Templated DN for users, `{}` will be replaced with the username (e.g. `cn={},dc=example,dc=com`, `{}@example.com`) |
-| LDAP_BASE_DN        |  None   | Starting point when searching for users authentication (e.g. `CN=Users,DC=xx,DC=yy,DC=de`)                         | 
+| LDAP_BASE_DN        |  None   | Starting point when searching for users authentication (e.g. `CN=Users,DC=xx,DC=yy,DC=de`)                         |
 | LDAP_ADMIN_FILTER   |  None   | Optional LDAP filter, which tells Mealie the LDAP user is an admin (e.g. `(memberOf=cn=admins,dc=example,dc=com)`) |

docs/docs/documentation/getting-started/installation/postgres.md

@@ -25,6 +25,10 @@ services:
   mealie-api:
     image: hkotel/mealie:api-v1.0.0beta-4
     container_name: mealie-api
+    deploy:
+      resources:
+        limits:
+          memory: 1000M # (4)
     depends_on:
       - postgres
     volumes:
@@ -66,3 +70,4 @@ volumes:
     <br/> <br/> **Note** that both containers must be on the same docker-network for this to work.
 2.  To access the mealie interface you only need to expose port 3000 on the mealie-frontend container. Here we expose port 9925 on the host, feel free to change this to any port you like.
 3.  Mounting the data directory to the frontend is now required to access the images/assets directory. This can be mounted read-only. Internally the frontend containers runs a Caddy proxy server that serves the assets requested to reduce load on the backend API.
+4.  Setting an explicit memory limit is recommended. Python can pre-allocate larger amounts of memory than is necessary if you have a machine with a lot of RAM. This can cause the container to idle at a high memory usage. Setting a memory limit will improve idle performance.

docs/docs/documentation/getting-started/installation/sqlite.md

@@ -25,6 +25,10 @@ services:
   mealie-api:
     image: hkotel/mealie:api-v1.0.0beta-4
     container_name: mealie-api
+    deploy:
+      resources:
+        limits:
+          memory: 1000M # (4)
     volumes:
       - mealie-data:/app/data/
     environment:
@@ -49,3 +53,4 @@ volumes:
     <br/> <br/> **Note** that both containers must be on the same docker-network for this to work.
 2.  To access the mealie interface you only need to expose port 3000 on the mealie-frontend container. Here we expose port 9925 on the host, feel free to change this to any port you like.
 3.  Mounting the data directory to the frontend is now required to access the images/assets directory. This can be mounted read-only. Internally the frontend containers runs a Caddy proxy server that serves the assets requested to reduce load on the backend API.
+4.  Setting an explicit memory limit is recommended. Python can pre-allocate larger amounts of memory than is necessary if you have a machine with a lot of RAM. This can cause the container to idle at a high memory usage. Setting a memory limit will improve idle performance.

mealie/run.sh

@@ -41,11 +41,6 @@ init() {
     poetry run python /app/mealie/db/init_db.py
 }
 
-# Migrations
-# TODO
-# Migrations
-# Set Port from ENV Variable
-
 if [ "$ARG1" == "reload" ]; then
     echo "Hot Reload!"
 
@@ -63,6 +58,11 @@ else
     GUNICORN_PORT=${API_PORT:-9000}
 
     # Start API
-    # uvicorn mealie.app:app --host 0.0.0.0 --port 9000
-    gunicorn mealie.app:app -b 0.0.0.0:$GUNICORN_PORT -k uvicorn.workers.UvicornWorker -c /app/gunicorn_conf.py --preload
+
+    if [ $WEB_GUNICORN == 'true' ]; then
+        echo "Starting Gunicorn"
+        gunicorn mealie.app:app -b 0.0.0.0:$GUNICORN_PORT -k uvicorn.workers.UvicornWorker -c /app/gunicorn_conf.py --preload
+    else
+        uvicorn mealie.app:app --host 0.0.0.0 --port $GUNICORN_PORT
+    fi
 fi

mealie/scripts/healthcheck.py

@@ -0,0 +1,23 @@
+import os
+
+import requests
+
+
+def main():
+    port = os.getenv("API_PORT")
+
+    if port is None:
+        port = 9000
+
+    url = f"http://127.0.0.1:{port}/api/app/about"
+
+    r = requests.get(url)
+
+    if r.status_code == 200:
+        exit(0)
+    else:
+        exit(1)
+
+
+if __name__ == "__main__":
+    main()

mealie/scripts/install_model.py

@@ -0,0 +1,21 @@
+import requests
+
+from mealie.services.parser_services import crfpp
+
+MODEL_URL = "https://github.com/mealie-recipes/nlp-model/releases/download/v1.0.0/model.crfmodel"
+
+
+def main():
+    """
+    Install the model into the crfpp directory
+    """
+
+    r = requests.get(MODEL_URL, stream=True, allow_redirects=True)
+    with open(crfpp.MODEL_PATH, "wb") as f:
+        for chunk in r.iter_content(chunk_size=1024):
+            if chunk:
+                f.write(chunk)
+
+
+if __name__ == "__main__":
+    main()