Module: check_mk Branch: master Commit: bf04855b1fd1f64882229a25559fa04e5074b805 URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=bf04855b1fd1f6… Author: Mathias Kettner &lt;mk(a)mathias-kettner.de&gt; Date: Fri Feb 20 14:10:38 2015 +0100 check_cluster_nodes: added cluster check from Karl-Heinz Fiebig --- doc/treasures/check_cluster_nodes | 145 +++++++++++++++++++++++++++++++++++++ 1 file changed, 145 insertions(+) diff --git a/doc/treasures/check_cluster_nodes b/doc/treasures/check_cluster_nodes new file mode 100755 index 0000000..40b8f47 --- /dev/null +++ b/doc/treasures/check_cluster_nodes @@ -0,0 +1,145 @@ +#!/usr/bin/python +# -*- encoding: utf-8; py-indent-offset: 4 -*- +############################################################################### +# name : ClusterNodesState.py +# to do : check state of all cluster nodes +# start : nagios plugin +# docu : needs a check_mk environment +# parameter : -n Clustername (use -h to see all posibilities) +# created : +# created by : +# +# modifed by date comment +############################################################################### + +import os, sys, getopt +import livestatus + +def usage(): + usage = """ + -h --help Prints this + -n --clustername cluster name + -w --warning warning at % of lost hosts + -c --critical critical at % of lost hosts + """ + print usage + +def main(argv): + clustername = 'NoClusterName' + warning = float(5) + critical = float(50) + + if len(argv) == 0 or not sys.argv[1].startswith('-'): + usage() + sys.exit(2) + + try: + opt, args = getopt.getopt(argv, "h:n:w:c:", [ 'help', 'clustername=', 'warning=', 'critical=' ] ) + except getopt.GetoptError, err: + print str(err) + usage() + sys.exit(2) + output = None + verbose = False + for o, a in opt: + if o in ("-h", "--help"): + usage() + sys.exit() + elif o in ("-n", "--clustername"): + clustername = a + elif o in ("-w", "--warning"): + warning = float(a) + elif o in ("-c", "--critical"): + critical = float(a) + else: + assert False, "unhandled option" + + try: + omd_root = os.getenv("OMD_ROOT") + socket_path = "unix:" + omd_root + "/tmp/run/live" + except: + sys.stderr.write("This example is indented to run in an OMD site\n") + sys.stderr.write("Please change socket_path in this example, if you are\n") + sys.stderr.write("not using OMD.\n") + sys.exit(1) + + try: + clusternodes = {} + up_list = [] + unreachable_list = [] + down_list = [] + unknown_list = [] + + parents = livestatus.SingleSiteConnection(socket_path).query_table("GET hosts\nColumns: parents\nFilter: host_name = %s\n" % clustername) + for i in range(len(parents[0][0])): + state = livestatus.SingleSiteConnection(socket_path).query_table("GET hosts\nColumns: name hard_state\nFilter: host_name = %s\n" % parents[0][0][i]) + cmk_state = livestatus.SingleSiteConnection(socket_path).query_table("GET services\nColumns: host_name description state\nFilter: description ~ Check_MK$\nFilter: host_name = %s\n" % parents[0][0][i]) + nodename = state[0][0] + if state[0][1] == 0 and cmk_state[0][2] == 0: + nodestate = "up" + elif state[0][1] == 1: + nodestate = "unreachable" + elif state[0][1] == 2: + nodestate = "down" + elif cmk_state[0][2] == 2: + nodestate = "unknown" + clusternodes[nodename] = nodestate + + #clusternodes['zbghvm42'] = 'unknown' + #clusternodes['zbghvm43'] = 'unreachable' + #clusternodes['zbghvm44'] = 'up' + #clusternodes['zbghvm45'] = 'up' + #clusternodes['zbghvm46'] = 'down' + #clusternodes['zbghvm47'] = 'down' + #clusternodes['zbghvm48'] = 'up' + #clusternodes['zbghvm49'] = 'unknown' + + for key, value in clusternodes.items(): + if value.count('up') > 0: + up_list.append(key) + if value.count('unreachable') > 0: + unreachable_list.append(key) + if value.count('down') > 0: + down_list.append(key) + if value.count('unknown') > 0: + unknown_list.append(key) + + all_nodes = float(len(clusternodes)) + up_nodes = float(len(up_list)) + unreachable_nodes = float(len(unreachable_list)) + down_nodes = float(len(down_list)) + unknown_nodes = float(len(unknown_list)) + + failedpercent = ((down_nodes + unreachable_nodes + unknown_nodes) / all_nodes) * 100.0 + + str_up = "" + str_down = "" + str_unreachable = "" + str_unknown = "" + if len(up_list) > 0: + str_up = (" %s " % ([str(item) for item in up_list])).translate(None, "'[,]") + if len(down_list) > 0: + str_down = ("(!!)%s " % ([str(item) for item in down_list])).translate(None, "'[,]") + if len(unreachable_list) > 0: + str_unreachable = ("(ur)%s " % ([str(item) for item in unreachable_list])).translate(None, "'[,]") + if len(unknown_list) > 0: + str_unknown = ("(?)%s " % ([str(item) for item in unknown_list])).translate(None, "'[,]") + + if failedpercent > warning and failedpercent < critical: + print "WARN - (.)" + str_up + str_down + str_unreachable + str_unknown + sys.exit(1) + elif failedpercent >= critical: + if str_up == "": + print "CRIT - " + str_up + str_down + str_unreachable + str_unknown + else: + print "CRIT - (.)" + str_up + str_down + str_unreachable + str_unknown + sys.exit(2) + else: + print "OK - (.)" + str_up + sys.exit(0) + + except Exception, e: # livestatus.MKLivestatusException, e: + print "Livestatus error: %s" % str(e) + +if __name__ == "__main__": + main(sys.argv[1:]) # [1:] slices off the first argument which is the name of the program