[checkmk-commits] Check_MK Git: check_mk: #2119 omd_status: check can now work in a cluster environment

Module: check_mk Branch: master Commit: 30d4a5ba7fc5ae7de00d96d3a381bf7d0e402660 URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=30d4a5ba7fc5ae… Author: Goetz Golla <gg(a)mathias-kettner.de> Date: Fri Jun 5 18:14:15 2015 +0200 #2119 omd_status: check can now work in a cluster environment --- .werks/2119 | 9 ++++++ ChangeLog | 1 + checkman/omd_status | 5 +++ checks/omd_status | 86 +++++++++++++++++++++++++++++++++++++++------------ 4 files changed, 81 insertions(+), 20 deletions(-) diff --git a/.werks/2119 b/.werks/2119 new file mode 100644 index 0000000..df120d8 --- /dev/null +++ b/.werks/2119 @@ -0,0 +1,9 @@ +Title: omd_status: check can now work in a cluster environment +Level: 1 +Component: checks +Compatible: compat +Version: 1.2.7i2 +Date: 1433520823 +Class: feature + + diff --git a/ChangeLog b/ChangeLog index 244c288..fa36cee 100644 --- a/ChangeLog +++ b/ChangeLog @@ -6,6 +6,7 @@ * 2117 postfix_mailq: agent and check now retrieve and monitor active queue data as well * 2325 oracle_tablespaces: correctly handle case where check runs in clustered mode... * 2216 raritan_pdu_ocprot: New check to monitor overcurrent protectors of Raritan PDUs... + * 2119 omd_status: check can now work in a cluster environment * 2323 FIX: df: Fix new graphs for all filesystem checks in case of existing inode information * 2305 FIX: agent_vsphere, esx_vsphere_sensors: now able to handle sensor names with semicolon... * 2118 FIX: aix_sap_processlist: agent plugin now is more general to deal with various AIX versions... diff --git a/checkman/omd_status b/checkman/omd_status index 362aad7..a51a9fb 100644 --- a/checkman/omd_status +++ b/checkman/omd_status @@ -10,6 +10,11 @@ description: otherwise. If the site is only partially running then the check output shows the names of the stopped services. + The check also works in a cluster environment, listing all node where the + sites are running on and their states. The total state of the service is + then only critical when one of the sites on a node is partially running, + or when all sites on the nodes are stopped. + item: The name of the site (string). diff --git a/checks/omd_status b/checks/omd_status index b3b56d0..4e47cf9 100644 --- a/checks/omd_status +++ b/checks/omd_status @@ -41,33 +41,79 @@ # OVERALL 2 def inventory_omd_status(info): - for line in info: - if line[0][0] == '[': - yield line[0][1:-1], None + for site in info.keys(): + yield site, None -def check_omd_status(item, _no_params, info): +def parse_omd_status(info): active = False + parsed = {} + for line in info: - if line[0] == '[' + item + ']': + if line[1][0] == '[': + item = line[1][1:-1] + # items may appear several times in clusters + # so dont overwrite the previous node result + if item not in parsed: + parsed[item] = {} + node = line[0] + parsed[item][node] = {} + parsed[item][node]["stopped"] = [] active = True stopped = [] - elif active: - if line[0] == 'OVERALL': - if line[1] == '0': - return (0, 'all services are running') - elif line[1] == '1': - return (2, 'site is stopped') - else: - return (2, 'partially running! stopped services: %s' % ", ".join(stopped)) - elif line[1] != '0': - stopped.append(line[0]) - return (3, "site not existing or AUTOSTART off") + elif active and line[1] == 'OVERALL': + if line[2] == '0': + parsed[item][node]["overall"] = "running" + elif line[2] == '1': + parsed[item][node]["overall"] = "stopped" + active = False + elif active and line[2] != '0': + parsed[item][node]["stopped"].append(line[1]) + parsed[item][node]["overall"] = "partially" + + return parsed + + +def check_omd_status(item, _no_params, info): + + parsed_site = info[item] + number_nodes = len(parsed_site) + stopped_nodes = 0 + + for node, services in parsed_site.iteritems(): + if services["overall"] == "stopped": + stopped_nodes += 1 + + # stopped sites are only CRIT when all are stopped + if stopped_nodes == number_nodes: + state = 2 + else: + state = 0 + for node, services in parsed_site.iteritems(): + if node: + node_text = " on %s" % node + else: + node_text = "" + if services["overall"] == "running": + infotext = "running%s" % node_text + # running sites are always OK + yield 0, infotext + elif services["overall"] == "stopped": + infotext = "stopped%s" % node_text + # stopped sites are only CRIT when all are stopped + yield state, infotext + else: + infotext = "partially running%s, stopped services: " % node_text + infotext += ", ".join(services["stopped"]) + # partially running sites are always CRIT + yield 2, infotext check_info["omd_status"] = { - 'check_function': check_omd_status, - 'inventory_function': inventory_omd_status, - 'service_description': 'OMD %s status', - 'group': 'omd_status', + 'check_function' : check_omd_status, + 'inventory_function' : inventory_omd_status, + 'parse_function' : parse_omd_status, + 'service_description' : 'OMD %s status', + 'group' : 'omd_status', + 'node_info' : True, }