Fix PID lockfile issue for both Hugging Face Spaces and Docker

Dockerfile

@@ -27,11 +27,14 @@ COPY .git .git
 
 COPY build.gradle.kts gradle.properties gradlew settings.gradle.kts ./
 
-RUN ./gradlew clean install -Pprod --parallel --info --stacktrace
+RUN ./gradlew clean -Pprod --parallel --info --stacktrace
 RUN ./gradlew :backend:package -Pprod --parallel --info --stacktrace
 RUN ./gradlew :frontend:package -Pprod --parallel --info --stacktrace
-RUN ./gradlew :python-client:package -Pprod --parallel --info --stacktrace
-
+RUN pip install pdm==2.8.2 urllib3==1.26.15 certifi==2023.7.22 && \
+    cd /app/python-client && \
+    pdm install --prod -G:all && \
+    pdm build && \
+    cd /app
 
 # Create an environment and install giskard wheel. Some dependencies may require gcc which is only installed in build
 # stage, but not in the production one
@@ -70,6 +73,8 @@ COPY --from=build /app/python-client/dist $GSK_DIST_PATH/python-client
 COPY --from=build /app/python-client/.venv-prod $VENV_PATH
 COPY --from=build /app/backend/build/libs/backend*.jar $GSK_DIST_PATH/backend/giskard.jar
 COPY --from=build /app/frontend/dist $GSK_DIST_PATH/frontend/dist
+COPY scripts/file-guard.sh $GSK_DIST_PATH/file-guard.sh
+COPY scripts/start-*.sh $GSK_DIST_PATH/
 
 COPY supervisord.conf ./
 COPY packaging/nginx.conf.template $GSK_DIST_PATH/frontend/

packaging/nginx.conf.template

@@ -1,4 +1,4 @@
-error_log stderr info;
+error_log stderr notice;
 pid "${GSK_HOME}/run/nginx/nginx.pid";
 daemon off;
 working_directory "${GSK_HOME}/run/nginx";

scripts/file-guard.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+# Script to guard lockfile with timeout
+
+if [ -z "$1" ]
+then
+    echo "No lockfile to monitor"
+    exit 0
+fi
+
+MAX_RETRY=${2:-100}
+INTERNAL=${3:-0.2}
+retry=0
+LOCKFILE="$1"
+while [ -f "$LOCKFILE" ] && [ $retry -lt $MAX_RETRY ]
+do
+    echo "Waiting for \"$LOCKFILE\" file to be removed"
+    sleep $INTERNAL
+    retry=$(( $retry + 1 ))
+done
+
+if [ -f "$LOCKFILE" ]
+then
+    echo "Warning: \"$LOCKFILE\" is still there after $MAX_RETRY retries"
+    echo "         Force to remove $LOCKFILE"
+    rm "$LOCKFILE"
+fi

scripts/start-db.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+
+[ ! -d $PGDATA ] && initdb -D "$PGDATA" -U giskard
+bash $GSK_DIST_PATH/file-guard.sh "$PGDATA/postmaster.pid"
+postgres -D "$PGDATA"

scripts/start-frontend.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+
+envsubst '\$GSK_HOME \$GSK_DIST_PATH' < $GSK_DIST_PATH/frontend/nginx.conf.template > $GSK_DIST_PATH/frontend/nginx.conf
+nginx -c $GSK_DIST_PATH/frontend/nginx.conf

scripts/start-worker.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+
+python -m giskard.cli worker stop -s
+python -m giskard.cli worker start -s

supervisord.conf

@@ -30,19 +30,20 @@ command=java -jar %(ENV_GSK_DIST_PATH)s/backend/giskard.jar
 stdout_logfile=%(ENV_GSK_HOME)s/run/%(program_name)s.log
 autorestart=true
 redirect_stderr=true
+stopsignal=INT
 startsecs=5
-command=/bin/bash -c '[ ! -d $PGDATA ] && initdb -D "$PGDATA" -U giskard; postgres -D "$PGDATA"'
+command=/bin/bash %(ENV_GSK_DIST_PATH)s/start-db.sh
 
 [program:worker]
 stdout_logfile=%(ENV_GSK_HOME)s/run/%(program_name)s.log
 autorestart=true
 redirect_stderr=true
 startsecs=5
-command=python -m giskard.cli worker start -s
+command=/bin/bash %(ENV_GSK_DIST_PATH)s/start-worker.sh
 
 [program:frontend]
 stdout_logfile=%(ENV_GSK_HOME)s/run/%(program_name)s.log
 autorestart=true
 redirect_stderr=true
 startsecs=5
-command=/bin/bash -c "envsubst '\$GSK_HOME \$GSK_DIST_PATH' < $GSK_DIST_PATH/frontend/nginx.conf.template > $GSK_DIST_PATH/frontend/nginx.conf && nginx -c $GSK_DIST_PATH/frontend/nginx.conf"
+command=/bin/bash %(ENV_GSK_DIST_PATH)s/start-frontend.sh