Merge pull request #58 from sfudeus/add_refresh_age_metric

Add refresh age metric and healthcheck

Dockerfile

@@ -1,9 +1,13 @@
 FROM python:3.12-slim-bookworm
 COPY requirements.txt /tmp
+RUN apt-get update && apt-get install curl -y
 RUN pip3 install --no-cache-dir -r /tmp/requirements.txt
 
 COPY exporter.py /usr/local/bin/homematic_exporter
+COPY healthcheck.sh /usr/local/bin/healthcheck.sh
 
 ENTRYPOINT [ "/usr/local/bin/homematic_exporter" ]
 
 EXPOSE 8010
+HEALTHCHECK --interval=20s --timeout=3s \
+    CMD bash /usr/local/bin/healthcheck.sh

exporter.py

@@ -152,6 +152,7 @@ def run(self):
             try:
                 with generate_metrics_summary.labels(self.ccu_host).time():
                     self.generate_metrics()
+                    self.refresh_time = time.time()
             except OSError as os_error:
                 logging.info("Failed to generate metrics: {0}".format(os_error))
                 error_counter.labels(self.ccu_host).inc()
@@ -183,6 +184,11 @@ def __init__(self, ccu_host, ccu_port, auth, gathering_interval, reload_names_in
         self.gathering_interval = int(gathering_interval)
         self.reload_names_interval = int(reload_names_interval)
         self.devicecount = Gauge('devicecount', 'Number of processed/supported devices', labelnames=['ccu'], namespace=self.METRICS_NAMESPACE)
+        # Upon request export the seconds since the last successful update.
+        # This is robust against internal crashes and can be used by the healthcheck.
+        self.refresh_time = time.time()
+        self.refresh_age = Gauge("refresh_age_seconds", "Seconds since the last successful refresh.", labelnames=["ccu"], namespace=self.METRICS_NAMESPACE)
+        self.refresh_age.labels(self.ccu_host).set_function(lambda: time.time() - self.refresh_time)
 
     def generate_metrics(self):
         logging.info("Gathering metrics")

healthcheck.sh

@@ -0,0 +1,14 @@
+#!/bin/bash
+
+# First argument is the maximum age in seconds (default: 60)
+maxage=${1:-60}
+
+# Get the age of the last successful metrics refresh in seconds
+age=$(curl -s http://localhost:9040/metrics | grep 'homematic_refresh_age_seconds{' | cut -d ' ' -f2 | cut -d '.' -f1)
+
+if [[ $age -lt $maxage ]]; then
+    exit 0
+else
+    # Maximum age exceeded → unhealthy
+    exit 1
+fi