double Checkable::GetLastCheck(void) const
{
CheckResult::Ptr cr = GetLastCheckResult();
double schedule_end = -1;
if (cr)
schedule_end = cr->GetScheduleEnd();
return schedule_end;
}
double Checkable::CalculateLatency(const CheckResult::Ptr& cr)
{
if (!cr)
return 0;
double latency = (cr->GetScheduleEnd() - cr->GetScheduleStart()) - CalculateExecutionTime(cr);
if (latency < 0)
latency = 0;
return latency;
}
void Checkable::UpdateStatistics(const CheckResult::Ptr& cr, CheckableType type)
{
time_t ts = cr->GetScheduleEnd();
if (type == CheckableHost) {
if (cr->GetActive())
CIB::UpdateActiveHostChecksStatistics(ts, 1);
else
CIB::UpdatePassiveHostChecksStatistics(ts, 1);
} else if (type == CheckableService) {
if (cr->GetActive())
CIB::UpdateActiveServiceChecksStatistics(ts, 1);
else
CIB::UpdatePassiveServiceChecksStatistics(ts, 1);
} else {
Log(LogWarning, "Checkable", "Unknown checkable type for statistic update.");
}
}
void Checkable::ProcessCheckResult(const CheckResult::Ptr& cr, const MessageOrigin::Ptr& origin)
{
{
ObjectLock olock(this);
m_CheckRunning = false;
}
double now = Utility::GetTime();
if (cr->GetScheduleStart() == 0)
cr->SetScheduleStart(now);
if (cr->GetScheduleEnd() == 0)
cr->SetScheduleEnd(now);
if (cr->GetExecutionStart() == 0)
cr->SetExecutionStart(now);
if (cr->GetExecutionEnd() == 0)
cr->SetExecutionEnd(now);
if (!origin || origin->IsLocal()) {
Log(LogDebug, "Checkable")
<< "No origin or local origin for object '" << GetName()
<< "', setting " << IcingaApplication::GetInstance()->GetNodeName()
<< " as check_source.";
cr->SetCheckSource(IcingaApplication::GetInstance()->GetNodeName());
}
Endpoint::Ptr command_endpoint = GetCommandEndpoint();
/* override check source if command_endpoint was defined */
if (command_endpoint && !GetExtension("agent_check")) {
Log(LogDebug, "Checkable")
<< "command_endpoint found for object '" << GetName()
<< "', setting " << command_endpoint->GetName()
<< " as check_source.";
cr->SetCheckSource(command_endpoint->GetName());
}
/* agent checks go through the api */
if (command_endpoint && GetExtension("agent_check")) {
ApiListener::Ptr listener = ApiListener::GetInstance();
if (listener) {
/* send message back to its origin */
Dictionary::Ptr message = ClusterEvents::MakeCheckResultMessage(this, cr);
listener->SyncSendMessage(command_endpoint, message);
}
return;
}
bool reachable = IsReachable();
bool notification_reachable = IsReachable(DependencyNotification);
ASSERT(!OwnsLock());
ObjectLock olock(this);
CheckResult::Ptr old_cr = GetLastCheckResult();
ServiceState old_state = GetStateRaw();
StateType old_stateType = GetStateType();
long old_attempt = GetCheckAttempt();
bool recovery = false;
if (old_cr && cr->GetExecutionStart() < old_cr->GetExecutionStart())
return;
/* The ExecuteCheck function already sets the old state, but we need to do it again
* in case this was a passive check result. */
SetLastStateRaw(old_state);
SetLastStateType(old_stateType);
SetLastReachable(reachable);
long attempt = 1;
std::set<Checkable::Ptr> children = GetChildren();
if (!old_cr) {
SetStateType(StateTypeHard);
} else if (cr->GetState() == ServiceOK) {
if (old_state == ServiceOK && old_stateType == StateTypeSoft) {
SetStateType(StateTypeHard); // SOFT OK -> HARD OK
recovery = true;
}
if (old_state != ServiceOK)
recovery = true; // NOT OK -> SOFT/HARD OK
ResetNotificationNumbers();
SetLastStateOK(Utility::GetTime());
/* update reachability for child objects in OK state */
if (!children.empty())
OnReachabilityChanged(this, cr, children, origin);
} else {
if (old_attempt >= GetMaxCheckAttempts()) {
SetStateType(StateTypeHard);
} else if (old_stateType == StateTypeSoft && old_state != ServiceOK) {
SetStateType(StateTypeSoft);
attempt = old_attempt + 1; //NOT-OK -> NOT-OK counter
} else if (old_state == ServiceOK) {
SetStateType(StateTypeSoft);
attempt = 1; //OK -> NOT-OK transition, reset the counter
} else {
attempt = old_attempt;
}
switch (cr->GetState()) {
case ServiceOK:
/* Nothing to do here. */
break;
case ServiceWarning:
SetLastStateWarning(Utility::GetTime());
break;
case ServiceCritical:
SetLastStateCritical(Utility::GetTime());
break;
case ServiceUnknown:
SetLastStateUnknown(Utility::GetTime());
break;
}
/* update reachability for child objects in NOT-OK state */
if (!children.empty())
OnReachabilityChanged(this, cr, children, origin);
}
if (!reachable)
SetLastStateUnreachable(Utility::GetTime());
SetCheckAttempt(attempt);
ServiceState new_state = cr->GetState();
SetStateRaw(new_state);
bool stateChange = (old_state != new_state);
if (stateChange) {
SetLastStateChange(now);
/* remove acknowledgements */
if (GetAcknowledgement() == AcknowledgementNormal ||
(GetAcknowledgement() == AcknowledgementSticky && new_state == ServiceOK)) {
ClearAcknowledgement();
}
/* reschedule direct parents */
BOOST_FOREACH(const Checkable::Ptr& parent, GetParents()) {
if (parent.get() == this)
continue;
ObjectLock olock(parent);
parent->SetNextCheck(Utility::GetTime());
}
}
bool remove_acknowledgement_comments = false;
if (GetAcknowledgement() == AcknowledgementNone)
remove_acknowledgement_comments = true;
bool hardChange = (GetStateType() == StateTypeHard && old_stateType == StateTypeSoft);
if (stateChange && old_stateType == StateTypeHard && GetStateType() == StateTypeHard)
hardChange = true;
bool is_volatile = GetVolatile();
if (hardChange || is_volatile) {
SetLastHardStateRaw(new_state);
SetLastHardStateChange(now);
}
if (new_state != ServiceOK)
TriggerDowntimes();
Host::Ptr host;
Service::Ptr service;
tie(host, service) = GetHostService(this);
CheckableType checkable_type = CheckableHost;
if (service)
checkable_type = CheckableService;
/* statistics for external tools */
Checkable::UpdateStatistics(cr, checkable_type);
bool in_downtime = IsInDowntime();
bool send_notification = hardChange && notification_reachable && !in_downtime && !IsAcknowledged();
if (!old_cr)
send_notification = false; /* Don't send notifications for the initial state change */
if (old_state == ServiceOK && old_stateType == StateTypeSoft)
send_notification = false; /* Don't send notifications for SOFT-OK -> HARD-OK. */
if (is_volatile && old_state == ServiceOK && new_state == ServiceOK)
send_notification = false; /* Don't send notifications for volatile OK -> OK changes. */
bool send_downtime_notification = (GetLastInDowntime() != in_downtime);
SetLastInDowntime(in_downtime);
olock.Unlock();
if (remove_acknowledgement_comments)
RemoveCommentsByType(CommentAcknowledgement);
Dictionary::Ptr vars_after = new Dictionary();
vars_after->Set("state", new_state);
vars_after->Set("state_type", GetStateType());
vars_after->Set("attempt", GetCheckAttempt());
vars_after->Set("reachable", reachable);
if (old_cr)
cr->SetVarsBefore(old_cr->GetVarsAfter());
cr->SetVarsAfter(vars_after);
olock.Lock();
SetLastCheckResult(cr);
bool was_flapping, is_flapping;
was_flapping = IsFlapping();
if (GetStateType() == StateTypeHard)
UpdateFlappingStatus(stateChange);
is_flapping = IsFlapping();
olock.Unlock();
// Log(LogDebug, "Checkable")
// << "Flapping: Checkable " << GetName()
// << " was: " << (was_flapping)
// << " is: " << is_flapping)
// << " threshold: " << GetFlappingThreshold()
// << "% current: " + GetFlappingCurrent()) << "%.";
OnNewCheckResult(this, cr, origin);
/* signal status updates to for example db_ido */
OnStateChanged(this);
String old_state_str = (service ? Service::StateToString(old_state) : Host::StateToString(Host::CalculateState(old_state)));
String new_state_str = (service ? Service::StateToString(new_state) : Host::StateToString(Host::CalculateState(new_state)));
if (hardChange || is_volatile) {
OnStateChange(this, cr, StateTypeHard, origin);
Log(LogNotice, "Checkable")
<< "State Change: Checkable " << GetName() << " hard state change from " << old_state_str << " to " << new_state_str << " detected." << (is_volatile ? " Checkable is volatile." : "");
} else if (stateChange) {
OnStateChange(this, cr, StateTypeSoft, origin);
Log(LogNotice, "Checkable")
<< "State Change: Checkable " << GetName() << " soft state change from " << old_state_str << " to " << new_state_str << " detected.";
}
if (GetStateType() == StateTypeSoft || hardChange || recovery || is_volatile)
ExecuteEventHandler();
if (send_downtime_notification)
OnNotificationsRequested(this, in_downtime ? NotificationDowntimeStart : NotificationDowntimeEnd, cr, "", "");
if (!was_flapping && is_flapping) {
OnNotificationsRequested(this, NotificationFlappingStart, cr, "", "");
Log(LogNotice, "Checkable")
<< "Flapping: Checkable " << GetName() << " started flapping (" << GetFlappingThreshold() << "% < " << GetFlappingCurrent() << "%).";
NotifyFlapping(origin);
} else if (was_flapping && !is_flapping) {
OnNotificationsRequested(this, NotificationFlappingEnd, cr, "", "");
Log(LogNotice, "Checkable")
<< "Flapping: Checkable " << GetName() << " stopped flapping (" << GetFlappingThreshold() << "% >= " << GetFlappingCurrent() << "%).";
NotifyFlapping(origin);
} else if (send_notification)
OnNotificationsRequested(this, recovery ? NotificationRecovery : NotificationProblem, cr, "", "");
}
Dictionary::Ptr ServiceDbObject::GetStatusFields(void) const
{
Dictionary::Ptr fields = make_shared<Dictionary>();
Service::Ptr service = static_pointer_cast<Service>(GetObject());
CheckResult::Ptr cr = service->GetLastCheckResult();
if (cr) {
fields->Set("output", CompatUtility::GetCheckResultOutput(cr));
fields->Set("long_output", CompatUtility::GetCheckResultLongOutput(cr));
fields->Set("perfdata", CompatUtility::GetCheckResultPerfdata(cr));
fields->Set("check_source", cr->GetCheckSource());
}
fields->Set("current_state", service->GetState());
fields->Set("has_been_checked", CompatUtility::GetCheckableHasBeenChecked(service));
fields->Set("should_be_scheduled", service->GetEnableActiveChecks());
fields->Set("current_check_attempt", service->GetCheckAttempt());
fields->Set("max_check_attempts", service->GetMaxCheckAttempts());
if (cr)
fields->Set("last_check", DbValue::FromTimestamp(cr->GetScheduleEnd()));
fields->Set("next_check", DbValue::FromTimestamp(service->GetNextCheck()));
fields->Set("check_type", CompatUtility::GetCheckableCheckType(service));
fields->Set("last_state_change", DbValue::FromTimestamp(service->GetLastStateChange()));
fields->Set("last_hard_state_change", DbValue::FromTimestamp(service->GetLastHardStateChange()));
fields->Set("last_time_ok", DbValue::FromTimestamp(static_cast<int>(service->GetLastStateOK())));
fields->Set("last_time_warning", DbValue::FromTimestamp(static_cast<int>(service->GetLastStateWarning())));
fields->Set("last_time_critical", DbValue::FromTimestamp(static_cast<int>(service->GetLastStateCritical())));
fields->Set("last_time_unknown", DbValue::FromTimestamp(static_cast<int>(service->GetLastStateUnknown())));
fields->Set("state_type", service->GetStateType());
fields->Set("last_notification", DbValue::FromTimestamp(CompatUtility::GetCheckableNotificationLastNotification(service)));
fields->Set("next_notification", DbValue::FromTimestamp(CompatUtility::GetCheckableNotificationNextNotification(service)));
fields->Set("no_more_notifications", Empty);
fields->Set("notifications_enabled", CompatUtility::GetCheckableNotificationsEnabled(service));
fields->Set("problem_has_been_acknowledged", CompatUtility::GetCheckableProblemHasBeenAcknowledged(service));
fields->Set("acknowledgement_type", CompatUtility::GetCheckableAcknowledgementType(service));
fields->Set("current_notification_number", CompatUtility::GetCheckableNotificationNotificationNumber(service));
fields->Set("passive_checks_enabled", CompatUtility::GetCheckablePassiveChecksEnabled(service));
fields->Set("active_checks_enabled", CompatUtility::GetCheckableActiveChecksEnabled(service));
fields->Set("event_handler_enabled", CompatUtility::GetCheckableEventHandlerEnabled(service));
fields->Set("flap_detection_enabled", CompatUtility::GetCheckableFlapDetectionEnabled(service));
fields->Set("is_flapping", CompatUtility::GetCheckableIsFlapping(service));
fields->Set("percent_state_change", CompatUtility::GetCheckablePercentStateChange(service));
if (cr) {
fields->Set("latency", Convert::ToString(Service::CalculateLatency(cr)));
fields->Set("execution_time", Convert::ToString(Service::CalculateExecutionTime(cr)));
}
fields->Set("scheduled_downtime_depth", service->GetDowntimeDepth());
fields->Set("process_performance_data", CompatUtility::GetCheckableProcessPerformanceData(service));
fields->Set("event_handler", CompatUtility::GetCheckableEventHandler(service));
fields->Set("check_command", CompatUtility::GetCheckableCheckCommand(service));
fields->Set("normal_check_interval", CompatUtility::GetCheckableCheckInterval(service));
fields->Set("retry_check_interval", CompatUtility::GetCheckableRetryInterval(service));
fields->Set("check_timeperiod_object_id", service->GetCheckPeriod());
fields->Set("modified_service_attributes", service->GetModifiedAttributes());
return fields;
}
Dictionary::Ptr HostDbObject::GetStatusFields(void) const
{
Dictionary::Ptr fields = new Dictionary();
Host::Ptr host = static_pointer_cast<Host>(GetObject());
CheckResult::Ptr cr = host->GetLastCheckResult();
if (cr) {
fields->Set("output", CompatUtility::GetCheckResultOutput(cr));
fields->Set("long_output", CompatUtility::GetCheckResultLongOutput(cr));
fields->Set("perfdata", CompatUtility::GetCheckResultPerfdata(cr));
fields->Set("check_source", cr->GetCheckSource());
}
fields->Set("current_state", host->IsReachable() ? host->GetState() : 2);
fields->Set("has_been_checked", CompatUtility::GetCheckableHasBeenChecked(host));
fields->Set("should_be_scheduled", host->GetEnableActiveChecks());
fields->Set("current_check_attempt", host->GetCheckAttempt());
fields->Set("max_check_attempts", host->GetMaxCheckAttempts());
if (cr)
fields->Set("last_check", DbValue::FromTimestamp(cr->GetScheduleEnd()));
fields->Set("next_check", DbValue::FromTimestamp(host->GetNextCheck()));
fields->Set("check_type", CompatUtility::GetCheckableCheckType(host));
fields->Set("last_state_change", DbValue::FromTimestamp(host->GetLastStateChange()));
fields->Set("last_hard_state_change", DbValue::FromTimestamp(host->GetLastHardStateChange()));
fields->Set("last_time_up", DbValue::FromTimestamp(static_cast<int>(host->GetLastStateUp())));
fields->Set("last_time_down", DbValue::FromTimestamp(static_cast<int>(host->GetLastStateDown())));
fields->Set("last_time_unreachable", DbValue::FromTimestamp(static_cast<int>(host->GetLastStateUnreachable())));
fields->Set("state_type", host->GetStateType());
fields->Set("last_notification", DbValue::FromTimestamp(CompatUtility::GetCheckableNotificationLastNotification(host)));
fields->Set("next_notification", DbValue::FromTimestamp(CompatUtility::GetCheckableNotificationNextNotification(host)));
fields->Set("no_more_notifications", Empty);
fields->Set("notifications_enabled", CompatUtility::GetCheckableNotificationsEnabled(host));
{
ObjectLock olock(host);
fields->Set("problem_has_been_acknowledged", CompatUtility::GetCheckableProblemHasBeenAcknowledged(host));
fields->Set("acknowledgement_type", CompatUtility::GetCheckableAcknowledgementType(host));
}
fields->Set("current_notification_number", CompatUtility::GetCheckableNotificationNotificationNumber(host));
fields->Set("passive_checks_enabled", CompatUtility::GetCheckablePassiveChecksEnabled(host));
fields->Set("active_checks_enabled", CompatUtility::GetCheckableActiveChecksEnabled(host));
fields->Set("event_handler_enabled", CompatUtility::GetCheckableEventHandlerEnabled(host));
fields->Set("flap_detection_enabled", CompatUtility::GetCheckableFlapDetectionEnabled(host));
fields->Set("is_flapping", CompatUtility::GetCheckableIsFlapping(host));
fields->Set("percent_state_change", CompatUtility::GetCheckablePercentStateChange(host));
if (cr) {
fields->Set("latency", Convert::ToString(Service::CalculateLatency(cr)));
fields->Set("execution_time", Convert::ToString(Service::CalculateExecutionTime(cr)));
}
fields->Set("scheduled_downtime_depth", host->GetDowntimeDepth());
fields->Set("failure_prediction_enabled", Empty);
fields->Set("process_performance_data", 0); /* this is a host which does not process any perf data */
fields->Set("obsess_over_host", Empty);
fields->Set("modified_host_attributes", host->GetModifiedAttributes());
fields->Set("event_handler", CompatUtility::GetCheckableEventHandler(host));
fields->Set("check_command", CompatUtility::GetCheckableCheckCommand(host));
fields->Set("normal_check_interval", CompatUtility::GetCheckableCheckInterval(host));
fields->Set("retry_check_interval", CompatUtility::GetCheckableRetryInterval(host));
fields->Set("check_timeperiod_object_id", host->GetCheckPeriod());
fields->Set("is_reachable", CompatUtility::GetCheckableIsReachable(host));
return fields;
}
void ElasticsearchWriter::AddCheckResult(const Dictionary::Ptr& fields, const Checkable::Ptr& checkable, const CheckResult::Ptr& cr)
{
String prefix = "check_result.";
fields->Set(prefix + "output", cr->GetOutput());
fields->Set(prefix + "check_source", cr->GetCheckSource());
fields->Set(prefix + "exit_status", cr->GetExitStatus());
fields->Set(prefix + "command", cr->GetCommand());
fields->Set(prefix + "state", cr->GetState());
fields->Set(prefix + "vars_before", cr->GetVarsBefore());
fields->Set(prefix + "vars_after", cr->GetVarsAfter());
fields->Set(prefix + "execution_start", FormatTimestamp(cr->GetExecutionStart()));
fields->Set(prefix + "execution_end", FormatTimestamp(cr->GetExecutionEnd()));
fields->Set(prefix + "schedule_start", FormatTimestamp(cr->GetScheduleStart()));
fields->Set(prefix + "schedule_end", FormatTimestamp(cr->GetScheduleEnd()));
/* Add extra calculated field. */
fields->Set(prefix + "latency", cr->CalculateLatency());
fields->Set(prefix + "execution_time", cr->CalculateExecutionTime());
if (!GetEnableSendPerfdata())
return;
Array::Ptr perfdata = cr->GetPerformanceData();
CheckCommand::Ptr checkCommand = checkable->GetCheckCommand();
if (perfdata) {
ObjectLock olock(perfdata);
for (const Value& val : perfdata) {
PerfdataValue::Ptr pdv;
if (val.IsObjectType<PerfdataValue>())
pdv = val;
else {
try {
pdv = PerfdataValue::Parse(val);
} catch (const std::exception&) {
Log(LogWarning, "ElasticsearchWriter")
<< "Ignoring invalid perfdata for checkable '"
<< checkable->GetName() << "' and command '"
<< checkCommand->GetName() << "' with value: " << val;
continue;
}
}
String escapedKey = pdv->GetLabel();
boost::replace_all(escapedKey, " ", "_");
boost::replace_all(escapedKey, ".", "_");
boost::replace_all(escapedKey, "\\", "_");
boost::algorithm::replace_all(escapedKey, "::", ".");
String perfdataPrefix = prefix + "perfdata." + escapedKey;
fields->Set(perfdataPrefix + ".value", pdv->GetValue());
if (pdv->GetMin())
fields->Set(perfdataPrefix + ".min", pdv->GetMin());
if (pdv->GetMax())
fields->Set(perfdataPrefix + ".max", pdv->GetMax());
if (pdv->GetWarn())
fields->Set(perfdataPrefix + ".warn", pdv->GetWarn());
if (pdv->GetCrit())
fields->Set(perfdataPrefix + ".crit", pdv->GetCrit());
if (!pdv->GetUnit().IsEmpty())
fields->Set(perfdataPrefix + ".unit", pdv->GetUnit());
}
}
}
