Update liveness for minute replication

Author: Rub21

images/replication-job/Dockerfile

@@ -50,5 +50,6 @@ RUN ldconfig
 ENV PATH="/osmdbt/build:/usr/local/bin:$PATH"
 
 COPY start.sh /start.sh
+COPY liveness.sh /liveness.sh
 ENTRYPOINT ["/bin/bash","-c"]
 CMD ["/start.sh"]

images/replication-job/liveness.sh

@@ -1,44 +1,40 @@
 #!/usr/bin/env bash
-# Script to check if Osmosis (Java) is running and also check how old processed_files.log is.
-# If processed_files.log is older than MAX_AGE_MINUTES, then kill Osmosis.
+# Liveness check for OSM replication process
+# Checks if main process is running, PostgreSQL connection, and if log file is being updated
+# Exit codes: 0=healthy, 1=process not running or DB unreachable, 2=process stuck
 
-LOG_FILE="/mnt/data/processed_files.log"
-MAX_AGE_MINUTES=10
+LOG_FILE="${WORKING_DIRECTORY:-/mnt/data}/logs/processed_files.log"
+MAX_AGE_MINUTES="${MAX_LOG_AGE_MINUTES:-10}"
 
-get_file_age_in_minutes() {
-  local file="$1"
-  if [ ! -f "$file" ]; then
-    echo 999999
-    return
-  fi
-  local now
-  local mtime
-  now=$(date +%s)
-  mtime=$(stat -c %Y "$file")
-  local diff=$(( (now - mtime) / 60 )) 
-  echo "$diff"
-}
+# Check if main process (start.sh) is running
+if ! pgrep -f "start.sh" >/dev/null 2>&1; then
+  echo "Main process (start.sh) is not running!"
+  exit 1
+fi
 
-# Check if Osmosis (Java) is running
-OSMOSIS_COUNT=$(ps -ef | grep -E 'java.*osmosis' | grep -v grep | wc -l)
+# Check PostgreSQL connection
+export PGPASSWORD="${POSTGRES_PASSWORD:-}"
+if ! pg_isready -h "${POSTGRES_HOST:-localhost}" -p "${POSTGRES_PORT:-5432}" -U "${POSTGRES_USER:-osm}" -d "${POSTGRES_DB:-osm}" >/dev/null 2>&1; then
+  echo "PostgreSQL is not reachable!"
+  exit 1
+fi
 
-if [ "$OSMOSIS_COUNT" -ge 1 ]; then
-  echo "Osmosis is running."
-  # Check how old the processed_files.log file is
-  file_age=$(get_file_age_in_minutes "$LOG_FILE")
-  echo "processed_files.log file age in minutes: $file_age"
-  if [ "$file_age" -ge "$MAX_AGE_MINUTES" ]; then
-    echo "processed_files.log is older than $MAX_AGE_MINUTES minutes. Attempting to kill Osmosis and restart the container..."
-    # Kill the Osmosis process
-    pkill -f "java.*osmosis" || true
-    echo "Osmosis is not terminating. Force-killing the container..."
-    echo "Container force-restart triggered."
-    exit 2
-  else
-    echo "processed_files.log is not too old. No action needed."
-    exit 0
-  fi
-else
-  echo "Osmosis is not running!"
+# Check log file age
+if [ ! -f "$LOG_FILE" ]; then
+  echo "Log file not found: $LOG_FILE"
   exit 1
 fi
+
+# Get file age in minutes
+now=$(date +%s)
+mtime=$(stat -c %Y "$LOG_FILE" 2>/dev/null || stat -f %m "$LOG_FILE" 2>/dev/null || echo "0")
+file_age=$(( (now - mtime) / 60 ))
+
+# If log is too old, process might be stuck
+if [ "$file_age" -ge "$MAX_AGE_MINUTES" ]; then
+  echo "Log file is older than $MAX_AGE_MINUTES minutes (age: $file_age min). Process may be stuck."
+  exit 2
+fi
+
+echo "Process is healthy (log age: $file_age min)"
+exit 0

osm-seed/templates/jobs/replication-job-deployment.yaml

@@ -55,6 +55,8 @@ spec:
               value: {{ quote .Values.db.env.POSTGRES_PASSWORD }}
             - name: POSTGRES_USER
               value: {{ .Values.db.env.POSTGRES_USER }}
+            - name: POSTGRES_PORT
+              value: {{ .Values.db.env.POSTGRES_PORT | quote}}
             - name: REPLICATION_FOLDER
               value: replication/minute
             - name: CLOUDPROVIDER