From ca07c1413249a5f427926c848e3942eb3e42d493 Mon Sep 17 00:00:00 2001 From: Denis GERMAIN Date: Sun, 4 Dec 2016 12:18:51 +0100 Subject: [PATCH] Corrected a CRITICAL state returned as 1 (WARNING state from nagios POV) Added a safeguard when information of the local node aren't found --- check_rhcs/{check_rhcs => check_rhcs.py} | 25 ++++++++++++------------ 1 file changed, 13 insertions(+), 12 deletions(-) rename check_rhcs/{check_rhcs => check_rhcs.py} (90%) diff --git a/check_rhcs/check_rhcs b/check_rhcs/check_rhcs.py similarity index 90% rename from check_rhcs/check_rhcs rename to check_rhcs/check_rhcs.py index a235498..5a8db20 100644 --- a/check_rhcs/check_rhcs +++ b/check_rhcs/check_rhcs.py @@ -1,6 +1,4 @@ #!/bin/env python - -# # Gather the cluster state and the current node state # # Output example: @@ -133,7 +131,7 @@ def main(): if typeCheck == 'cluster': # First we query for the state of the cluster itself. - # Should it be found tha the cluste ris not quorate we alert and exit immediately + # Should it be found that the cluster is not quorate we alert and exit immediately cluster = getClusterName(dom) qState = getQuorumState(dom) @@ -145,15 +143,18 @@ def main(): # Now we find the status of the local node from clustat. # We only care about the local state since this way we can tie the alert to the host. nodeStates = getLocalNodeState(dom) - if nodeStates['state'] != "1": - print "WARNING: Local node state is offline!" - sys.exit(1) - elif nodeStates['rgmanager'] != "1": - print "CRITICAL: RGManager service not running on " + nodeStates['name'] + "!" - sys.exit(1) - else: - print "OK: Cluster node " + nodeStates['name'] + " is online and cluster is quorate." - sys.exit(0) + if nodeStates == {}: + print "UNKNOWN: Local node informations couldn't be found!" + sys.exit(3) + if nodeStates['state'] != "1": + print "WARNING: Local node state is offline!" + sys.exit(1) + elif nodeStates['rgmanager'] != "1": + print "CRITICAL: RGManager service not running on " + nodeStates['name'] + "!" + sys.exit(2) + else: + print "OK: Cluster node " + nodeStates['name'] + " is online and cluster is quorate." + sys.exit(0) elif typeCheck == 'service': serviceState = getServiceState(dom, serviceName)