merge develop into main

aio/s6-rc.d/postgres/finish

@@ -1,7 +1,7 @@
 #!/command/with-contenv bash
 set -eu
 
-STATE_DIR="/run/postgres"
+STATE_DIR="/run/service/postgres"
 COUNT_FILE="$STATE_DIR/restart-count"
 MAX_RESTARTS="${MAX_RESTARTS:-3}"
 
@@ -12,12 +12,15 @@ count=$(cat "$COUNT_FILE" 2>/dev/null || echo 0)
 count=$((count+1))
 echo -n "$count" > "$COUNT_FILE"
 
+# Record last restart timestamp
+date +%s > "$STATE_DIR/last-restart"
+
 echo "[postgres] crash count: $count/${MAX_RESTARTS}"
 
 if [ "$count" -ge "$MAX_RESTARTS" ]; then
   echo "[postgres] too many failures, terminating supervision tree"
   # Ask s6-svscan to exit; this will stop the container
-  exec /command/s6-svscanctl -t /run/service
+exec /command/s6-svscanctl -t /run/service
 fi
 
 exit 0

aio/s6-rc.d/price-aggregator/finish

@@ -1,7 +1,7 @@
 #!/command/with-contenv bash
 set -eu
 
-STATE_DIR="/run/price-aggregator"
+STATE_DIR="/run/service/price-aggregator"
 COUNT_FILE="$STATE_DIR/restart-count"
 MAX_RESTARTS="${MAX_RESTARTS:-3}"
 
@@ -14,12 +14,15 @@ count=$(cat "$COUNT_FILE" 2>/dev/null || echo 0)
 count=$((count+1))
 echo "$count" > "$COUNT_FILE"
 
+# Record last restart timestamp
+date +%s > "$STATE_DIR/last-restart"
+
 echo "[price-aggregator] crash count: $count/${MAX_RESTARTS}"
 
 if [ "$count" -ge "$MAX_RESTARTS" ]; then
   echo "[price-aggregator] too many failures, terminating supervision tree"
   # Ask s6-svscan to exit; this will stop the container
-  exec /command/s6-svscanctl -t /run/service
+exec /command/s6-svscanctl -t /run/service
 fi
 
 exit 0

aio/s6-rc.d/price-aggregator/run

@@ -31,4 +31,6 @@ fi
 mkdir -p /var/log/price-aggregator
 exec > >(s6-log n10 s10485760 /var/log/price-aggregator) 2>&1
 
+export PROCESS_METRICS_TTL_SECONDS=259200
+
 exec node dist/main

aio/s6-rc.d/process-metrics/finish

@@ -0,0 +1,22 @@
+#!/command/with-contenv bash
+set -eu
+
+STATE_DIR="/run/process-metrics"
+COUNT_FILE="$STATE_DIR/restart-count"
+MAX_RESTARTS="${MAX_RESTARTS:-3}"
+
+mkdir -p "$STATE_DIR"
+
+count=$(cat "$COUNT_FILE" 2>/dev/null || echo 0)
+count=$((count+1))
+echo "$count" > "$COUNT_FILE"
+date +%s > "$STATE_DIR/last-restart"
+
+echo "[process-metrics] crash count: $count/${MAX_RESTARTS}"
+
+if [ "$count" -ge "$MAX_RESTARTS" ]; then
+  echo "[process-metrics] too many failures, terminating supervision tree"
+  exec /command/s6-svscanctl -t /run/service
+fi
+
+exit 0

aio/s6-rc.d/process-metrics/run

@@ -0,0 +1,33 @@
+#!/command/with-contenv bash
+set -euo pipefail
+
+echo "[process-metrics] Starting exporter..."
+
+GLOBAL_CONFIG_JSON_PATH="/sp/configurations/configuration.json"
+
+mkdir -p /etc/process-metrics
+CONFIG_FILE_PATH=/etc/process-metrics/config.json
+if [ -f "$GLOBAL_CONFIG_JSON_PATH" ]; then
+  if ! jq -r '.solution.priceAggregatorConfig.metricsPush' "$GLOBAL_CONFIG_JSON_PATH" > "$CONFIG_FILE_PATH" 2>/dev/null; then
+    echo "[process-metrics] WARN: failed to extract priceAggregatorConfig.metricsPush from $GLOBAL_CONFIG_JSON_PATH" >&2
+  fi
+else
+  echo "[process-metrics] WARN: global config not found at $GLOBAL_CONFIG_JSON_PATH; using existing $CONFIG_FILE_PATH if present" >&2
+fi
+
+# Force jobName and groupingLabels.service to "process-metrics"
+if [ -f "$CONFIG_FILE_PATH" ]; then
+  TMP_FILE="$(mktemp)"
+  if ! jq '.jobName = "process-metrics" | .groupingLabels.service = "process-metrics"' "$CONFIG_FILE_PATH" > "$TMP_FILE" 2>/dev/null; then
+    echo "[process-metrics] WARN: failed to rewrite jobName/groupingLabels.service in $CONFIG_FILE_PATH" >&2
+    rm -f "$TMP_FILE"
+  else
+    mv "$TMP_FILE" "$CONFIG_FILE_PATH"
+  fi
+fi
+
+# Route stdout/stderr through s6-log with rotation
+mkdir -p /var/log/process-metrics
+exec > >(s6-log n10 s10485760 /var/log/process-metrics) 2>&1
+
+exec node /scripts/process-metrics-exporter/index.js

aio/s6-rc.d/process-metrics/type

@@ -0,0 +1 @@
+longrun

scripts/bash/entrypoint-allinone.sh

@@ -83,14 +83,15 @@ set -eu
 # Fixed service name using node index from generator time
 SERVICE_NAME="chainlink-node-${i}"
 
-STATE_DIR="/run/\${SERVICE_NAME}"
+STATE_DIR="/run/service/\${SERVICE_NAME}"
 COUNT_FILE="\${STATE_DIR}/restart-count"
 MAX_RESTARTS="\${MAX_RESTARTS:-3}"
 
 mkdir -p "\$STATE_DIR"
 count=\$(cat "\$COUNT_FILE" 2>/dev/null || echo 0)
 count=\$((count+1))
 echo "\$count" > "\$COUNT_FILE"
+date +%s > "\$STATE_DIR/last-restart"
 
 echo "[\${SERVICE_NAME}] crash count: \$count/\${MAX_RESTARTS}"