[checkmk-commits] 4170 local: local check is now clusteraware. Best or worst state in cluster mode can be configured

Module: check_mk Branch: master Commit: 98968cdbbde6ba8f3a4cdc3f5373e47f844ab0b7 URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=98968cdbbde6ba… Author: Simon Betz &lt;si(a)mathias-kettner.de&gt; Date: Mon Dec 19 08:05:34 2016 +0100 4170 local: local check is now clusteraware. Best or worst state in cluster mode can be configured Change-Id: Iea613ba04877ed54fab98f4846c1015b5bf51142 --- .werks/4170 | 11 +++ ChangeLog | 1 + checks/local | 138 +++++++++++++++++++++++------------ web/plugins/wato/check_parameters.py | 20 +++++ 4 files changed, 125 insertions(+), 45 deletions(-) diff --git a/.werks/4170 b/.werks/4170 new file mode 100644 index 0000000..1f0eb6d --- /dev/null +++ b/.werks/4170 @@ -0,0 +1,11 @@ +Title: local: local check is now clusteraware. Best or worst state in cluster mode can be configured +Level: 1 +Component: checks +Compatible: compat +Version: 1.4.0i4 +Date: 1482131073 +Class: feature + +Now your're able to determine the state of local checks on clusters via {{Settings for local checks}} +in {{Host & Service parameters}}. Within these settings you can choose between best or worst state. +Default setting is worst state. diff --git a/ChangeLog b/ChangeLog index 598017a..eb242a5 100644 --- a/ChangeLog +++ b/ChangeLog @@ -40,6 +40,7 @@ * 4098 ups_cps_battery, ups_cps_battery.temp, ups_cps_inphase, ups_cps_outphase: Several checks to monitor CPS UPS devices * 4118 check_bi_aggr: changed check and WATO rule to support Kerberos auth... * 4104 veeam_tapejobs: New check to monitor Veeam tape jobs + * 4170 local: local check is now clusteraware. Best or worst state in cluster mode can be configured * 3987 FIX: Check_MK Agent Access: Windows agent reported incorrect only from value * 3952 FIX: diskstat: fixed bug if multipath devices having an alias... * 3939 FIX: f5_bigip_conns: readded performance data and graphs... diff --git a/checks/local b/checks/local index db71272..a73f548 100644 --- a/checks/local +++ b/checks/local @@ -99,61 +99,109 @@ def inventory_local(info): inventory = [] # Lines with P do not need to supply a text for line in info: - if len(line) >= 4 or len(line) == 3 and line[0] == 'P': - inventory.append( (line[1], None) ) + nodename = line[0] + stripped_line = line[1:] + if len(stripped_line) >= 4 or len(stripped_line) == 3 and stripped_line[0] == 'P': + inventory.append( (stripped_line[1], None) ) else: - raise MKGeneralException("Invalid line in agent section <<<local>>>: %s" % " ".join(line)) + raise MKGeneralException("Invalid line in agent section <<<local>>>: %s" % " ".join(stripped_line)) return inventory +# Some helper functions +def _parse_local_line(line): + if not (len(line) >= 4 or (len(line) == 3 and line[0] == 'P')): + return 3, "Incomplete line in local check output: %s" % " ".join(line), [] + + statechar = line[0] + perftxt = line[2] + + # convert eventually escaped newinfo_line chars to real newinfo_lines + # (will be converted back later individually for the different cores) + output = " ".join(line[3:]).replace("\\n", "\n") + + perfdata = [] + if perftxt != "-": + for entry in perftxt.split('|'): + try: + varname, valuetxt = entry.split('=') + values = valuetxt.split(';') + perfdata.append( tuple([varname] + values) ) + + except ValueError: + return 3, "Invalid performance data %s in local check output %s" % \ + (perftxt, " ".join(line)), [] + + if statechar == 'P': + state, texts = local_compute_state(perfdata) + if output: + texts = [output] + texts + output = ", ".join(texts) + + else: + try: + state = int(statechar) + except: + return 3, "Invalid state %s in local check output %s: must be P, 0, 1, 2 or 3" % \ + (statechar, " ".join(line)), [] + + if state not in range(0, 4): + output += ", local check has sent invalid state %d" % state + state = 3 + + return state, output, perfdata + + +def _calculate_local_best_state(collected_stats): + states = [] + infotexts = [] + perfdatas = [] + for nodename, attrs in collected_stats.items(): + for itemname, (state, output, perfdata) in attrs.items(): + if nodename is not None: + output = "On node %s: %s" % (nodename, output) + states.append(state) + infotexts.append(output) + perfdatas += perfdata + return min(states), ", ".join(infotexts), perfdatas + + +def _calculate_local_worst_state(collected_stats): + for nodename, attrs in collected_stats.items(): + for itemname, (state, output, perfdata) in attrs.items(): + if nodename is not None: + output = "On node %s: %s" % (nodename, output) + yield state, output, perfdata + + def check_local(item, params, info): + collected_stats = {} for line in info: + nodename = line[0] + stripped_line = line[1:] # Ignore invalid lines, tolerate bugs in local checks # of unexperienced users - if len(line) >= 2 and line[1] == item: - if not (len(line) >= 4 or (len(line) == 3 and line[0] == 'P')): - return 3, "Incomplete line in local check output: %s" % " ".join(line) - - statechar = line[0] - perftxt = line[2] - - output = " ".join(line[3:]) - # convert eventually escaped newline chars to real newlines - # (will be converted back later individually for the different cores) - output = output.replace("\\n", "\n") - - perfdata = [] - if perftxt != "-": - # new: allow multiple perfdata by using | as separator - for entry in perftxt.split('|'): - try: - varname, valuetxt = entry.split('=') - values = valuetxt.split(';') - perfdata.append(tuple( [varname] + values )) - except ValueError: - return 3, "Invalid performance data %s in local check output %s" % \ - (perftxt, " ".join(line)) - if statechar == 'P': - state, texts = local_compute_state(perfdata) - if output: - texts = [output] + texts - output = ", ".join(texts) - else: - try: - state = int(statechar) - except: - return 3, "Invalid state %s in local check output %s: must be P, 0, 1, 2 or 3" % \ - (statechar, " ".join(line)) + if len(stripped_line) >= 2 and stripped_line[1] == item: + collected_stats.setdefault(nodename, {}) + collected_stats[nodename].setdefault(item, _parse_local_line(stripped_line)) + + if collected_stats == {}: + yield 3, "No data found in agent output" + return - if state not in range(0, 4): - output += ", local check has sent invalid state %d" % state - state = 3 - return (state, output, perfdata) + if params is not None and params.get("outcome_on_cluster", "worst") == "best": + yield _calculate_local_best_state(collected_stats) + return + else: + for res in _calculate_local_worst_state(collected_stats): + yield res check_info["local"] = { - 'inventory_function' : inventory_local, - 'check_function' : check_local, - 'service_description' : '%s', - 'has_perfdata' : True, + 'check_function' : check_local, + 'inventory_function' : inventory_local, + 'service_description' : '%s', + 'has_perfdata' : True, + 'node_info' : True, + 'group' : 'local', } diff --git a/web/plugins/wato/check_parameters.py b/web/plugins/wato/check_parameters.py index 6cb567c..92a622e 100644 --- a/web/plugins/wato/check_parameters.py +++ b/web/plugins/wato/check_parameters.py @@ -1423,6 +1423,26 @@ register_rule(group + '/' + subgroup_applications, ) +register_check_parameters( + subgroup_applications, + "local", + _("Settings for local checks"), + Dictionary( + elements = [ + ("outcome_on_cluster", DropdownChoice(choices = [ + ("worst", _("Worst state")), + ("best", _("Best state")), + ], + title = _("Clusters: Prefered check result of local checks"), + help = _("If you're running local checks on clusters via clustered services rule " + "you can influence the check result with this rule. You can choose between " + "best or worst state. Default setting is worst state."), + default_value = "worst")) + ] + ), + TextAscii(title = _("Name of local item")), + "dict" +) register_check_parameters(