[checkmk-commits] 3411 mysql_slave: now max time and related state can be configured if slave-IO and slave-SQL are not running

Module: check_mk Branch: master Commit: 3873a8291346e2c5a68909fe9808a0109f6c379e URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=3873a8291346e2… Author: Simon Betz &lt;si(a)mathias-kettner.de&gt; Date: Fri Apr 15 11:46:57 2016 +0200 3411 mysql_slave: now max time and related state can be configured if slave-IO and slave-SQL are not running --- .werks/3411 | 9 +++++ ChangeLog | 1 + checks/mysql_slave | 73 +++++++++++++++++++++++----------- web/plugins/wato/check_parameters.py | 13 ++++++ 4 files changed, 72 insertions(+), 24 deletions(-) diff --git a/.werks/3411 b/.werks/3411 new file mode 100644 index 0000000..daefe79 --- /dev/null +++ b/.werks/3411 @@ -0,0 +1,9 @@ +Title: mysql_slave: now max time and related state can be configured if slave-IO and slave-SQL are not running +Level: 1 +Component: checks +Compatible: compat +Version: 1.2.9i1 +Date: 1460713517 +Class: feature + + diff --git a/ChangeLog b/ChangeLog index a671347..0aedf0c 100644 --- a/ChangeLog +++ b/ChangeLog @@ -60,6 +60,7 @@ NOTE: Please refer to the migration notes! * 3027 mtr (Matt's traceroute): New agent plugin... * 3410 ipmi: now monitors compact/discrete sensors... + * 3411 mysql_slave: now max time and related state can be configured if slave-IO and slave-SQL are not running * 3073 FIX: windows agent: relative paths to mrpe scripts are now treated as relative to the agent installation directory... * 3061 FIX: mk_jolokia: Fixed debugging of the agent plugin * 3074 FIX: windows agent: fixed incorrect values for 32-bit performance counters diff --git a/checks/mysql_slave b/checks/mysql_slave index 6d89d6e..aa73d9f 100644 --- a/checks/mysql_slave +++ b/checks/mysql_slave @@ -78,40 +78,65 @@ def inventory_mysql_slave(parsed): def check_mysql_slave(item, params, parsed): if item in parsed: data = parsed[item] - state = 0 perfdata = [] - output = [] - if data['Slave_IO_Running']: - output.append('Slave-IO: running') - else: - output.append('Slave-IO: not running(!!)') - state = 2 - - if data['Slave_SQL_Running']: - output.append('Slave-SQL: running') + if data['Slave_IO_Running'] and data['Slave_SQL_Running']: + infotext = 'Slave-IO: running, Slave-SQL: running' + seconds_behind_master = data['Seconds_Behind_Master'] + warn, crit = params.get('seconds_behind_master', (None, None)) + set_item_state("slave_running", time.time()) # Makes only sense to monitor the age when the SQL slave is running - if data['Seconds_Behind_Master'] == 'NULL': - output.append('Time behind master: NULL (Lost connection?)(!!)') + if seconds_behind_master == 'NULL': + infotext += ', Time behind master: NULL (Lost connection?)(!!)' state = 2 else: - out = 'Time behind Master: %s' % get_age_human_readable(data['Seconds_Behind_Master']) - warn, crit = params.get('seconds_behind_master', (None, None)) - if crit != None and data['Seconds_Behind_Master'] > crit: + infotext += ', Time behind Master: %s' % get_age_human_readable(seconds_behind_master) + if crit != None and seconds_behind_master > crit: state = 2 - out += '(!!)' - elif warn != None and data['Seconds_Behind_Master'] > warn: - state = max(state, 1) - out += '(!)' - output.append(out) - perfdata.append(('sync_latency', data['Seconds_Behind_Master'], warn, crit)) - else: - output.append('Slave-SQL: not running(!!)') + elif warn != None and seconds_behind_master > warn: + state = 1 + + if state: + infotext += " (warn crit at %s/%s)" % \ + (get_age_human_readable(warn), + get_age_human_readable(crit)) + + perfdata.append(('sync_latency', seconds_behind_master, warn, crit)) + + elif not data['Slave_IO_Running'] and not data['Slave_SQL_Running']: + old_value = get_item_state("slave_running") + not_running_state, not_running_max_seconds = \ + params.get("slave_not_running", (None, None)) + + if old_value != None and not_running_state != None and \ + not_running_max_seconds != None: + + not_running_age = time.time() - old_value + infotext = 'Slave-IO: not running, Slave-SQL: not running, Age: %s (crit at %s)' % \ + (get_age_human_readable(not_running_age), + get_age_human_readable(not_running_max_seconds)) + + if not_running_age < not_running_max_seconds: + state = not_running_state + else: + state = 2 + + else: + infotext = "Slave-IO: not running(!!), Slave-SQL: not running(!!)" + state = 2 + + elif not data['Slave_IO_Running'] and data['Slave_SQL_Running']: + infotext = 'Slave-IO: not running(!!), Slave-SQL: running' + state = 2 + + elif data['Slave_IO_Running'] and not data['Slave_SQL_Running']: + infotext = 'Slave-IO: running, Slave-SQL: not running(!!)' state = 2 - return state, ', '.join(output), perfdata + return state, infotext, perfdata + check_info['mysql_slave'] = { "parse_function" : parse_mysql_slave, diff --git a/web/plugins/wato/check_parameters.py b/web/plugins/wato/check_parameters.py index 9d717e1..f2db844 100644 --- a/web/plugins/wato/check_parameters.py +++ b/web/plugins/wato/check_parameters.py @@ -6695,6 +6695,19 @@ register_check_parameters( ] ) ), + ( "slave_not_running", + Tuple( + title = _("State and max. time if slave is not running"), + help = _("This rule allows you to configure the state which will be " + "reported until the configured time will be exceeded."), + elements = [ + MonitoringState( + default_value = 2, + ), + Age(), + ] + ), + ), ], optional_keys = None), TextAscii(