Merge branch '379-ignore-kubejobcompletion-kubejobfailed' into 'master'

Ignore failing pods and jobs from NC installation

Closes #379

See merge request openappstack/openappstack!184

test/pytest/test_prometheus.py

 import json
 import pytest
-import requests
+import re
+
 
 @pytest.mark.prometheus
 def test_prometheus_alerts(host):
+    def summarize_alerts(alerts):
+        """Print a alert summary."""
+        print('Total alerts: %s' % len(alerts))
+        print(json.dumps(alerts, indent=2))
 
     print("Starting prometheus test...")
 
@@ -12,15 +17,37 @@ def test_prometheus_alerts(host):
     status = alert_json["status"]
     alerts = alert_json["data"]["alerts"]
 
-    # Filter out the ever firing "Dead mans switch"
-    real_alerts = [alert for alert in alerts
-                   if alert["labels"]["severity"] != "none"]
+    real_alerts = []
+    ignored_alerts = []
+
+    for alert in alerts:
+        # Filter out the ever firing "Dead mans switch" test alert
+        if (alert["labels"]["severity"] == "none" or \
+            # Filter out failing Nextcloud installation jobs since a lot of
+            # them fail until they succeed during installation
+            (re.match(r'(KubeJobFailed|KubeJobCompletion)',
+                      alert["labels"]["alertname"]) and
+             "nextcloud" in alert["labels"]["job_name"]) or
+            # Filter out failing Nextcloud pods since a lot of pods fail
+            # during installation
+            # We use python-behave tests to check for functionality
+            (re.match(r'(KubePodNotReady)',
+                      alert["labels"]["alertname"]) and
+             "nextcloud" in alert["labels"]["pod"])):
+            ignored_alerts.append(alert)
+        else:
+            real_alerts.append(alert)
+
+    print('\n\n\n========= Ignored ==========')
+    summarize_alerts(ignored_alerts)
+
+    print('\n\n\n========= Firing ==========')
+    summarize_alerts(real_alerts)
 
-    alert_names = list(map(lambda alert: alert["labels"]["alertname"], real_alerts))
     count = len(real_alerts)
 
     assert status == "success", "Failure queriying the prometheus api at" + url
-    assert count == 0, "Firing alerts: {0}".format(str(alert_names))
+    assert count == 0, "Firing alerts: {0}".format(count)
 
 
 if __name__ == "__main__":