diff --git a/check_rhcs/check_rhcs b/check_rhcs/check_rhcs.py similarity index 83% rename from check_rhcs/check_rhcs rename to check_rhcs/check_rhcs.py index a235498..66fd565 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: @@ -23,6 +21,14 @@ # # # Frank Clements +# +# INFO : In RHEL 5, there is a bug in clustat preventing non-root users to use +# clustat. See https://bugzilla.redhat.com/show_bug.cgi?id=531273 +# You might need to use setuid on clustat to change this if rgmanager cannot be +# upgraded to 3.0.7+ +# $chown root:nagios /usr/sbin/clustat +# $chmod u+s /usr/sbin/clustat + import xml.dom.minidom import os @@ -112,6 +118,7 @@ def main(): sys.exit(2) check_suspend = False + typeCheck = None for o, a in opts: if o in ('-c', '--cluster'): typeCheck = 'cluster' @@ -124,6 +131,10 @@ def main(): usage() sys.exit() + if typeCheck == None: + usage() + sys.exit() + try: clustatOutput = os.popen('/usr/sbin/clustat -fx') dom = xml.dom.minidom.parse(clustatOutput) @@ -133,7 +144,7 @@ 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 +156,18 @@ 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)