Module: check_mk Branch: master Commit: 7d9cb8e4d7d214446e978c9e82ab20e39922c092 URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=7d9cb8e4d7d214… Author: Bastian Kuhn &lt;bk(a)mathias-kettner.de&gt; Date: Fri Oct 24 15:51:23 2014 +0200 #1190 statgrab_cpu: Check can now handle parameters --- .werks/1190 | 9 +++++++ ChangeLog | 1 + checks/cpu_util.include | 58 +++++++++++++++++++++++++++++++++++++++ checks/kernel | 69 +++++++---------------------------------------- checks/statgrab_cpu | 51 ++++++++++++++--------------------- 5 files changed, 98 insertions(+), 90 deletions(-) diff --git a/.werks/1190 b/.werks/1190 new file mode 100644 index 0000000..3cda189 --- /dev/null +++ b/.werks/1190 @@ -0,0 +1,9 @@ +Title: statgrab_cpu: Check can now handle parameters +Level: 1 +Component: checks +Compatible: compat +Version: 1.2.5i6 +Date: 1414158659 +Class: feature + + diff --git a/ChangeLog b/ChangeLog index 1c45197..59f6890 100644 --- a/ChangeLog +++ b/ChangeLog @@ -74,6 +74,7 @@ * 1442 ups_socomec_out_source: New check for checking the power source of out phases for Socomec UPSs * 0662 domino_mailqueues: new check to monitor mail queues in Lotus Domino * 1188 veeam_client: Check now also outputs ReadSize and TransferedSize... + * 1190 statgrab_cpu: Check can now handle parameters * 1051 FIX: tcp_conn_stats: fix missing performance data... * 1142 FIX: winperf_ts_sessions: fix computation, check has never really worked * 1090 FIX: zfsget: fixed exception which happened on incomplete zfs entries diff --git a/checks/cpu_util.include b/checks/cpu_util.include index 410d24a..0ef034b 100644 --- a/checks/cpu_util.include +++ b/checks/cpu_util.include @@ -77,3 +77,61 @@ def check_cpu_util(util, params, this_time = None): perfdata += extraperf # reference curve for predictive levels return state, infotext, perfdata +# This one can handle user, system and wait +def check_cpu_util_unix(values, params): + this_time = int(time.time()) + # Compute jiffi-differences of all relevant counters + diff_values = [] + n = 0 + global g_counters + for v in values: + n += 1 + countername = "cpu.util.%d" % n + last_time, last_val = g_counters.get(countername, (0, 0)) + diff_values.append(v - last_val) + g_counters[countername] = (this_time, v) + + sum_jiffies = sum(diff_values) # do not account for steal! + if sum_jiffies == 0: + raise MKCounterWrapped(None, "Too short time difference since last check") + print sum_jiffies + + user = diff_values[0] + diff_values[1] # add user + nice + system = diff_values[2] + wait = diff_values[3] + user_perc = 100.0 * float(user) / float(sum_jiffies) + system_perc = 100.0 * float(system) / float(sum_jiffies) + wait_perc = 100.0 * float(wait) / float(sum_jiffies) + perfdata = [ + ( "user", "%.3f" % user_perc ), + ( "system", "%.3f" % system_perc ), + ( "wait", "%.3f" % wait_perc ) ] + + yield 0, "user: %.1f%%, system: %.1f%%" % (user_perc, system_perc), perfdata + + # Handle level on iowait + state = 0 + if "iowait" in params and params["iowait"] != None: + warn, crit = params["iowait"] + if wait_perc >= crit: + state = 2 + elif wait_perc >= warn: + state = 1 + yield state, "wait: %.1f%%" % (wait_perc) + + # Total utilization + util_total_perc = user_perc + system_perc + wait_perc + state = 0 + levelstext = "" + if "util" in params: + warn, crit = params["util"] + if util_total_perc >= crit: + state = 2 + elif util_total_perc >= warn: + state = 1 + else: + state = 0 + if state: + levelstext = " (warn/crit at %.1f%%/%.1f%%)" % (warn, crit) + + yield state, "total: %.1f%%" % util_total_perc + levelstext diff --git a/checks/kernel b/checks/kernel index 6135420..e773fee 100644 --- a/checks/kernel +++ b/checks/kernel @@ -129,8 +129,7 @@ def kernel_check_cpu_utilization(item, params, info): # Look for entry beginning with "cpu" f = [ l for l in info if l[0] == "cpu" ] if len(f) != 1: - yield 3, "More than one line with CPU info found. This check is not cluster-enabled." - return + return 3, "More than one line with CPU info found. This check is not cluster-enabled." line = f[0] if len(line) < 8: @@ -140,63 +139,14 @@ def kernel_check_cpu_utilization(item, params, info): # 'cpu' user nice system idle wait hw-int sw-int (steal ...) # convert number to int values = [ int(x) for x in line[1:8] ] - this_time = int(time.time()) - - # Compute jiffi-differences of all relevant counters - diff_values = [] - n = 0 - global g_counters - for v in values: - n += 1 - countername = "cpu.util.%d" % n - last_time, last_val = g_counters.get(countername, (0, 0)) - diff_values.append(v - last_val) - g_counters[countername] = (this_time, v) - - sum_jiffies = sum(diff_values[0:7]) # do not account for steal! - if sum_jiffies == 0: - raise MKCounterWrapped(None, "Too short time difference since last check") - - user = diff_values[0] + diff_values[1] # add user + nice - system = diff_values[2] - wait = diff_values[4] - user_perc = 100.0 * float(user) / float(sum_jiffies) - system_perc = 100.0 * float(system) / float(sum_jiffies) - wait_perc = 100.0 * float(wait) / float(sum_jiffies) - perfdata = [ - ( "user", "%.3f" % user_perc ), - ( "system", "%.3f" % system_perc ), - ( "wait", "%.3f" % wait_perc ) ] - - yield 0, "user: %.1f%%, system: %.1f%%" % (user_perc, system_perc), perfdata - - # Handle level on iowait - state = 0 - if "iowait" in params and params["iowait"] != None: - warn, crit = params["iowait"] - if wait_perc >= crit: - state = 2 - elif wait_perc >= warn: - state = 1 - yield state, "wait: %.1f%%" % (wait_perc) - - # Total utilization - util_total_perc = user_perc + system_perc + wait_perc - state = 0 - levelstext = "" - if "util" in params: - warn, crit = params["util"] - if util_total_perc >= crit: - state = 2 - elif util_total_perc >= warn: - state = 1 - else: - state = 0 - if state: - levelstext = " (warn/crit at %.1f%%/%.1f%%)" % (warn, crit) - - yield state, "total: %.1f%%" % util_total_perc + levelstext - + user = values[0] + nice = values[1] + system = values[2] + wait = values[4] + hw_int = values[5] + sw_int = values[6] + idle = 100 - user - system - nice - wait + return check_cpu_util_unix([user, nice, system, wait, hw_int, sw_int, idle], params) check_info["kernel.util"] = { 'check_function': kernel_check_cpu_utilization, @@ -205,5 +155,6 @@ check_info["kernel.util"] = { 'has_perfdata': True, 'default_levels_variable': 'kernel_util_default_levels', 'group': 'cpu_iowait', + 'includes': ['cpu_util.include'], } diff --git a/checks/statgrab_cpu b/checks/statgrab_cpu index 7bb0880..b52e0e2 100644 --- a/checks/statgrab_cpu +++ b/checks/statgrab_cpu @@ -28,49 +28,38 @@ def inventory_statgrab_cpu(info): if len(info) > 1: - return [(None, None)] + return [(None, {})] def check_statgrab_cpu(item, params, info): - global g_counters + if not params: + params = {} user = 0 + nice = 0 for var, value in info: if var == 'iowait': wait = int(value) elif var == 'kernel': system = int(value) - elif var == 'nice' or var == 'user': - user += int(value) + elif var == 'nice': + nice = int(value) + elif var == 'user': + user = int(value) elif var == 'total': total = int(value) - - values = [ user, system, wait, total ] - this_time = int(time.time()) - diff_values = [ ] - n = 0 - for v in values: - n += 1 - countername = "cpu.util.%d" % n - last_time, last_val = g_counters.get(countername, (0, 0)) - diff_values.append(v - last_val) - g_counters[countername] = (this_time, v) - - diff_total = diff_values[3] - if diff_total == 0: - return (0, "too short interval") - user_perc = 100.0 * float(diff_values[0]) / float(diff_total) - system_perc = 100.0 * float(diff_values[1]) / float(diff_total) - wait_perc = 100.0 * float(diff_values[2]) / float(diff_total) - perfdata = [ - ( "user", "%.3f" % user_perc ), - ( "system", "%.3f" % system_perc ), - ( "wait", "%.3f" % wait_perc ) ] - return (0, "user: %2.0f%%, system: %2.0f%%, wait: %2.0f%%" % (user_perc, system_perc, wait_perc), perfdata) + idle = 100 - total + # user, nice, system, wait, hw-int, sw-int, idle + values = [ user, nice, system, wait, 0, 0, idle ] + + # No return cause of the use of yield + return check_cpu_util_unix(values, params) check_info["statgrab_cpu"] = { - 'check_function': check_statgrab_cpu, - 'inventory_function': inventory_statgrab_cpu, - 'service_description': 'CPU utilization', - 'has_perfdata': True, + 'check_function': check_statgrab_cpu, + 'inventory_function': inventory_statgrab_cpu, + 'service_description': 'CPU utilization', + 'has_perfdata': True, + 'includes': ['cpu_util.include'], + 'group': 'cpu_iowait', }