[checkmk-commits] WATO rules for k8s network interfaces

Module: check_mk Branch: master Commit: 688610983aaf6d29ea57de323a0c142967d16fec URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=688610983aaf6d… Author: Óscar Nájera &lt;on(a)mathias-kettner.de&gt; Date: Wed Jan 23 18:11:16 2019 +0100 WATO rules for k8s network interfaces CMK-1542 Change-Id: I9dd74194a09e96822dbfa473b751bb21fc9c559d --- checks/k8s_stats | 64 +++++++++++++++++++++------- cmk/gui/plugins/wato/check_parameters/if.py | 47 ++++++++++++++++++++ tests/unit/cmk/gui/watolib/test_rulespecs.py | 33 +++++++++++++- 3 files changed, 128 insertions(+), 16 deletions(-) diff --git a/checks/k8s_stats b/checks/k8s_stats index d07b0c3..fe3c5ed 100644 --- a/checks/k8s_stats +++ b/checks/k8s_stats @@ -39,6 +39,29 @@ def discover_k8s_stats_fs(key, value): return [x for x in devices if x not in ignore_fs] +def k8s_network_err_pac(device, params, now): + for name, mway, pway in [('Input', 'rx', 'in'), ('Output', 'tx', 'out')]: + pac_rate = get_rate('if.%s.if_%s_pkts' % (device['name'], pway), now, + device['%s_packets' % mway]) + err_rate = get_rate('if.%s.if_%s_errors' % (device['name'], pway), now, + device['%s_errors' % mway]) + warn, crit = params.get('errors', (None, None)) + perf_data = [('if_%s_pkts' % pway, pac_rate), ('if_%s_errors' % pway, err_rate)] + if isinstance(warn, float): + try: + err_perc = 100.0 * err_rate / (err_rate + pac_rate) + except ZeroDivisionError: + err_perc = 0 + + status, infotext, _perf = check_levels( + err_perc, 'err_perc', (warn, crit), unit='%', infoname=name + " errors percentage") + else: # absolute levels or no levels + status, infotext, _perf = check_levels( + err_rate, 'if_errors', (warn, crit), unit='/s', infoname=name + " error rate") + + yield status, infotext, perf_data + + def check_k8s_stats_network(item, params, metrics): def get_if(): for device in metrics['network']['interfaces']: @@ -46,21 +69,31 @@ def check_k8s_stats_network(item, params, metrics): return device device = get_if() - - perf_names = [ - ('In', 'rx_bytes', 'in', lambda x: "%s/s" % get_bytes_human_readable(x)), - ('Out', 'tx_bytes', 'out', lambda x: "%s/s" % get_bytes_human_readable(x)), - ('Input Packets', 'rx_packets', 'if_in_pkts', None), - ('Output Packets', 'tx_packets', 'if_out_pkts', None), - ('Input Errors', 'rx_errors', 'if_in_errors', None), - ('Output Errors', 'tx_errors', 'if_out_errors', None), - ('Input Discards', 'rx_dropped', 'if_in_discards', None), - ('Output Discards', 'tx_dropped', 'if_out_discards', None), - ] - - for name, met, dsname, hrf in perf_names: - rate = get_rate('if.%s.%s' % (item, dsname), time.time(), device[met]) - yield check_levels(rate, dsname, params, unit='/s', human_readable_func=hrf, infoname=name) + now = time.strptime(metrics['timestamp'][:-4], "%Y-%m-%dT%H:%M:%S.%f") + now = time.mktime(now) - time.timezone + + # Bandwidth + for name, met, dsname in [('In', 'rx_bytes', 'in'), ('Out', 'tx_bytes', 'out')]: + rate = get_rate('if.%s.%s' % (item, dsname), now, device[met]) + yield check_levels( + rate, + dsname, + None, + unit='/s', + human_readable_func=lambda x: "%s/s" % get_bytes_human_readable(x), + infoname=name) + + # Errors / Packets + for check_result in k8s_network_err_pac(device, params, now): + yield check_result + + # Discards + for name, met, dsname in [ + ('Input Discards', 'rx_dropped', 'if_in_discards'), + ('Output Discards', 'tx_dropped', 'if_out_discards'), + ]: + rate = get_rate('if.%s.%s' % (item, dsname), now, device[met]) + yield check_levels(rate, dsname, params.get('discards'), unit='/s', infoname=name) def check_k8s_stats_fs(item, params, metrics): @@ -91,6 +124,7 @@ check_info['k8s_stats.network'] = { 'check_function': check_k8s_stats_network, 'service_description': 'Interface %s', "has_perfdata": True, + "group": "k8s_if", 'includes': ['k8s.include'], } diff --git a/cmk/gui/plugins/wato/check_parameters/if.py b/cmk/gui/plugins/wato/check_parameters/if.py index e11e369..7a9647d 100644 --- a/cmk/gui/plugins/wato/check_parameters/if.py +++ b/cmk/gui/plugins/wato/check_parameters/if.py @@ -583,3 +583,50 @@ register_check_parameters( TextAscii(title=_("port specification"), allow_empty=False), "dict", ) + +register_check_parameters( + RulespecGroupCheckParametersNetworking, + "k8s_if", + _("Kubernetes Network interfaces"), + Dictionary(elements=[ + ("errors", + Alternative( + title=_("Levels for error rates"), + help= + _("These levels make the check go warning or critical whenever the " + "<b>percentual error rate</b> or the <b>absolute error rate</b> of the monitored interface reaches " + "the given bounds. The percentual error rate is computed by dividing number of " + "errors by the total number of packets (successful plus errors)."), + elements=[ + Tuple( + title=_("Percentual levels for error rates"), + elements=[ + Percentage( + title=_("Warning at"), + unit=_("percent errors"), + default_value=0.01, + display_format='%.3f'), + Percentage( + title=_("Critical at"), + unit=_("percent errors"), + default_value=0.1, + display_format='%.3f') + ]), + Tuple( + title=_("Absolute levels for error rates"), + elements=[ + Integer(title=_("Warning at"), unit=_("errors")), + Integer(title=_("Critical at"), unit=_("errors")) + ]) + ])), + ("discards", + Tuple( + title=_("Absolute levels for discards rates"), + elements=[ + Integer(title=_("Warning at"), unit=_("discards")), + Integer(title=_("Critical at"), unit=_("discards")) + ])), + ]), + TextAscii(title=_("port specification"), allow_empty=False), + "dict", +) diff --git a/tests/unit/cmk/gui/watolib/test_rulespecs.py b/tests/unit/cmk/gui/watolib/test_rulespecs.py index fd35786..a6f662d 100644 --- a/tests/unit/cmk/gui/watolib/test_rulespecs.py +++ b/tests/unit/cmk/gui/watolib/test_rulespecs.py @@ -31,7 +31,6 @@ from cmk.gui.plugins.wato.utils import ( ) - def test_rulespec_sub_group(): class TestGroup(RulespecGroup): @property @@ -143,6 +142,7 @@ def test_grouped_rulespecs(): 'static_checks:bluecat_dns', 'static_checks:bluecat_ha', 'static_checks:steelhead_connections', + 'static_checks:k8s_if', ], 'eventconsole': [ 'extra_host_conf:_ec_event_limit', @@ -242,6 +242,7 @@ def test_grouped_rulespecs(): 'checkgroup_parameters:cisco_qos', 'if_groups', 'if_disable_if64_hosts', + 'checkgroup_parameters:k8s_if', 'checkgroup_parameters:adva_ifs', 'checkgroup_parameters:bluecat_ntp', 'checkgroup_parameters:bluecat_dhcp', @@ -6079,6 +6080,21 @@ expected_rulespecs = { 'title': u'JVM uptime (since last reboot)', 'valuespec_class_name': 'TimeperiodValuespec' }, + 'checkgroup_parameters:k8s_if': { + 'factory_default': [], + 'group_name': 'checkparams/networking', + 'help': None, + 'is_deprecated': False, + 'is_optional': False, + 'item_enum': None, + 'item_help': None, + 'item_name': u'port specification', + 'item_spec_class_name': 'TextAscii', + 'item_type': 'item', + 'match_type': 'dict', + 'title': u'Kubernetes Network interfaces', + 'valuespec_class_name': 'TimeperiodValuespec' + }, 'checkgroup_parameters:k8s_pods_cpu': { 'factory_default': [], 'group_name': 'checkparams/applications', @@ -14154,6 +14170,21 @@ expected_rulespecs = { 'title': u'JVM uptime (since last reboot)', 'valuespec_class_name': 'Tuple' }, + 'static_checks:k8s_if': { + 'factory_default': [], + 'group_name': 'static/networking', + 'help': None, + 'is_deprecated': False, + 'is_optional': False, + 'item_enum': None, + 'item_help': None, + 'item_name': None, + 'item_spec_class_name': 'TextAscii', + 'item_type': None, + 'match_type': 'all', + 'title': u'Kubernetes Network interfaces', + 'valuespec_class_name': 'Tuple' + }, 'static_checks:k8s_pods_cpu': { 'factory_default': [], 'group_name': 'static/applications',