void InfluxdbWriter::SendPerfdata(const Dictionary::Ptr& tmpl, const Checkable::Ptr& checkable, const CheckResult::Ptr& cr, double ts)
{
Array::Ptr perfdata = cr->GetPerformanceData();
if (!perfdata)
return;
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, "InfluxdbWriter")
<< "Ignoring invalid perfdata value: " << val;
continue;
}
}
Dictionary::Ptr fields = new Dictionary();
fields->Set(String("value"), pdv->GetValue());
if (GetEnableSendThresholds()) {
if (pdv->GetCrit())
fields->Set(String("crit"), pdv->GetCrit());
if (pdv->GetWarn())
fields->Set(String("warn"), pdv->GetWarn());
if (pdv->GetMin())
fields->Set(String("min"), pdv->GetMin());
if (pdv->GetMax())
fields->Set(String("max"), pdv->GetMax());
}
if (GetEnableSendMetadata()) {
Host::Ptr host;
Service::Ptr service;
boost::tie(host, service) = GetHostService(checkable);
if (service)
fields->Set(String("state"), FormatInteger(service->GetState()));
else
fields->Set(String("state"), FormatInteger(host->GetState()));
fields->Set(String("current_attempt"), FormatInteger(checkable->GetCheckAttempt()));
fields->Set(String("max_check_attempts"), FormatInteger(checkable->GetMaxCheckAttempts()));
fields->Set(String("state_type"), FormatInteger(checkable->GetStateType()));
fields->Set(String("reachable"), FormatBoolean(checkable->IsReachable()));
fields->Set(String("downtime_depth"), FormatInteger(checkable->GetDowntimeDepth()));
fields->Set(String("acknowledgement"), FormatInteger(checkable->GetAcknowledgement()));
fields->Set(String("latency"), cr->CalculateLatency());
fields->Set(String("execution_time"), cr->CalculateExecutionTime());
}
SendMetric(tmpl, pdv->GetLabel(), fields, ts);
}
}
void OpenTsdbWriter::CheckResultHandler(const Checkable::Ptr& checkable, const CheckResult::Ptr& cr)
{
CONTEXT("Processing check result for '" + checkable->GetName() + "'");
if (!IcingaApplication::GetInstance()->GetEnablePerfdata() || !checkable->GetEnablePerfdata())
return;
Service::Ptr service = dynamic_pointer_cast<Service>(checkable);
Host::Ptr host;
if (service)
host = service->GetHost();
else
host = static_pointer_cast<Host>(checkable);
String metric;
std::map<String, String> tags;
String escaped_hostName = EscapeTag(host->GetName());
tags["host"] = escaped_hostName;
double ts = cr->GetExecutionEnd();
if (service) {
String serviceName = service->GetShortName();
String escaped_serviceName = EscapeMetric(serviceName);
metric = "icinga.service." + escaped_serviceName;
SendMetric(metric + ".state", tags, service->GetState(), ts);
} else {
metric = "icinga.host";
SendMetric(metric + ".state", tags, host->GetState(), ts);
}
SendMetric(metric + ".state_type", tags, checkable->GetStateType(), ts);
SendMetric(metric + ".reachable", tags, checkable->IsReachable(), ts);
SendMetric(metric + ".downtime_depth", tags, checkable->GetDowntimeDepth(), ts);
SendMetric(metric + ".acknowledgement", tags, checkable->GetAcknowledgement(), ts);
SendPerfdata(metric, tags, cr, ts);
metric = "icinga.check";
if (service) {
tags["type"] = "service";
String serviceName = service->GetShortName();
String escaped_serviceName = EscapeTag(serviceName);
tags["service"] = escaped_serviceName;
} else {
tags["type"] = "host";
}
SendMetric(metric + ".current_attempt", tags, checkable->GetCheckAttempt(), ts);
SendMetric(metric + ".max_check_attempts", tags, checkable->GetMaxCheckAttempts(), ts);
SendMetric(metric + ".latency", tags, cr->CalculateLatency(), ts);
SendMetric(metric + ".execution_time", tags, cr->CalculateExecutionTime(), ts);
}
Value ServicesTable::LatencyAccessor(const Value& row)
{
Service::Ptr service = static_cast<Service::Ptr>(row);
if (!service)
return Empty;
CheckResult::Ptr cr = service->GetLastCheckResult();
if (!cr)
return Empty;
return cr->CalculateLatency();
}
void LogstashWriter::CheckResultHandler(const Checkable::Ptr& checkable, const CheckResult::Ptr& cr)
{
CONTEXT("LOGSTASH Processing check result for '" + checkable->GetName() + "'");
Log(LogDebug, "LogstashWriter")
<< "Processing check result for '" << checkable->GetName() << "'";
Host::Ptr host;
Service::Ptr service;
tie(host, service) = GetHostService(checkable);
Dictionary::Ptr fields = new Dictionary();
if (service) {
fields->Set("service_name", service->GetShortName());
fields->Set("service_state", Service::StateToString(service->GetState()));
fields->Set("last_state", service->GetLastState());
fields->Set("last_hard_state", service->GetLastHardState());
} else {
fields->Set("last_state", host->GetLastState());
fields->Set("last_hard_state", host->GetLastHardState());
}
fields->Set("host_name", host->GetName());
fields->Set("type", "CheckResult");
fields->Set("state", service ? Service::StateToString(service->GetState()) : Host::StateToString(host->GetState()));
fields->Set("current_check_attempt", checkable->GetCheckAttempt());
fields->Set("max_check_attempts", checkable->GetMaxCheckAttempts());
fields->Set("latency", cr->CalculateLatency());
fields->Set("execution_time", cr->CalculateExecutionTime());
fields->Set("reachable", checkable->IsReachable());
double ts = Utility::GetTime();
if (cr) {
fields->Set("plugin_output", cr->GetOutput());
fields->Set("check_source", cr->GetCheckSource());
ts = cr->GetExecutionEnd();
}
Array::Ptr perfdata = cr->GetPerformanceData();
if (perfdata) {
Dictionary::Ptr perfdataItems = new Dictionary();
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, "LogstashWriter")
<< "Ignoring invalid perfdata value: '" << val << "' for object '"
<< checkable->GetName() << "'.";
continue;
}
}
Dictionary::Ptr perfdataItem = new Dictionary();
perfdataItem->Set("value", pdv->GetValue());
if (pdv->GetMin())
perfdataItem->Set("min", pdv->GetMin());
if (pdv->GetMax())
perfdataItem->Set("max", pdv->GetMax());
if (pdv->GetWarn())
perfdataItem->Set("warn", pdv->GetWarn());
if (pdv->GetCrit())
perfdataItem->Set("crit", pdv->GetCrit());
String escaped_key = EscapeMetricLabel(pdv->GetLabel());
perfdataItems->Set(escaped_key, perfdataItem);
}
fields->Set("performance_data", perfdataItems);
}
SendLogMessage(ComposeLogstashMessage(fields, GetSource(), ts));
}
void StatusDataWriter::DumpCheckableStatusAttrs(std::ostream& fp, const Checkable::Ptr& checkable)
{
CheckResult::Ptr cr = checkable->GetLastCheckResult();
EventCommand::Ptr eventcommand = checkable->GetEventCommand();
CheckCommand::Ptr checkcommand = checkable->GetCheckCommand();
fp << "\t" << "check_command=" << CompatUtility::GetCommandName(checkcommand) << "!" << CompatUtility::GetCheckableCommandArgs(checkable) << "\n"
"\t" "event_handler=" << CompatUtility::GetCommandName(eventcommand) << "\n"
"\t" "check_interval=" << (checkable->GetCheckInterval() / 60.0) << "\n"
"\t" "retry_interval=" << (checkable->GetRetryInterval() / 60.0) << "\n"
"\t" "has_been_checked=" << Convert::ToLong(checkable->HasBeenChecked()) << "\n"
"\t" "should_be_scheduled=" << checkable->GetEnableActiveChecks() << "\n"
"\t" "event_handler_enabled=" << Convert::ToLong(checkable->GetEnableEventHandler()) << "\n";
TimePeriod::Ptr checkPeriod = checkable->GetCheckPeriod();
if (checkPeriod)
fp << "\t" "check_period" "\t" << checkPeriod->GetName() << "\n";
if (cr) {
fp << "\t" << "check_execution_time=" << Convert::ToString(cr->CalculateExecutionTime()) << "\n"
"\t" "check_latency=" << Convert::ToString(cr->CalculateLatency()) << "\n";
}
Host::Ptr host;
Service::Ptr service;
tie(host, service) = GetHostService(checkable);
if (service) {
fp << "\t" "current_state=" << service->GetState() << "\n"
"\t" "last_hard_state=" << service->GetLastHardState() << "\n"
"\t" "last_time_ok=" << static_cast<int>(service->GetLastStateOK()) << "\n"
"\t" "last_time_warn=" << static_cast<int>(service->GetLastStateWarning()) << "\n"
"\t" "last_time_critical=" << static_cast<int>(service->GetLastStateCritical()) << "\n"
"\t" "last_time_unknown=" << static_cast<int>(service->GetLastStateUnknown()) << "\n";
} else {
int currentState = host->GetState();
if (currentState != HostUp && !host->IsReachable())
currentState = 2; /* hardcoded compat state */
fp << "\t" "current_state=" << currentState << "\n"
"\t" "last_hard_state=" << host->GetLastHardState() << "\n"
"\t" "last_time_up=" << static_cast<int>(host->GetLastStateUp()) << "\n"
"\t" "last_time_down=" << static_cast<int>(host->GetLastStateDown()) << "\n";
}
fp << "\t" "state_type=" << checkable->GetStateType() << "\n"
"\t" "last_check=" << static_cast<long>(host->GetLastCheck()) << "\n";
if (cr) {
fp << "\t" "plugin_output=" << CompatUtility::GetCheckResultOutput(cr) << "\n"
"\t" "long_plugin_output=" << CompatUtility::GetCheckResultLongOutput(cr) << "\n"
"\t" "performance_data=" << PluginUtility::FormatPerfdata(cr->GetPerformanceData()) << "\n";
}
fp << "\t" << "next_check=" << static_cast<long>(checkable->GetNextCheck()) << "\n"
"\t" "current_attempt=" << checkable->GetCheckAttempt() << "\n"
"\t" "max_attempts=" << checkable->GetMaxCheckAttempts() << "\n"
"\t" "last_state_change=" << static_cast<long>(checkable->GetLastStateChange()) << "\n"
"\t" "last_hard_state_change=" << static_cast<long>(checkable->GetLastHardStateChange()) << "\n"
"\t" "last_update=" << static_cast<long>(Utility::GetTime()) << "\n"
"\t" "notifications_enabled=" << Convert::ToLong(checkable->GetEnableNotifications()) << "\n"
"\t" "active_checks_enabled=" << Convert::ToLong(checkable->GetEnableActiveChecks()) << "\n"
"\t" "passive_checks_enabled=" << Convert::ToLong(checkable->GetEnablePassiveChecks()) << "\n"
"\t" "flap_detection_enabled=" << Convert::ToLong(checkable->GetEnableFlapping()) << "\n"
"\t" "is_flapping=" << Convert::ToLong(checkable->IsFlapping()) << "\n"
"\t" "percent_state_change=" << checkable->GetFlappingCurrent() << "\n"
"\t" "problem_has_been_acknowledged=" << (checkable->GetAcknowledgement() != AcknowledgementNone ? 1 : 0) << "\n"
"\t" "acknowledgement_type=" << checkable->GetAcknowledgement() << "\n"
"\t" "acknowledgement_end_time=" << checkable->GetAcknowledgementExpiry() << "\n"
"\t" "scheduled_downtime_depth=" << checkable->GetDowntimeDepth() << "\n"
"\t" "last_notification=" << CompatUtility::GetCheckableNotificationLastNotification(checkable) << "\n"
"\t" "next_notification=" << CompatUtility::GetCheckableNotificationNextNotification(checkable) << "\n"
"\t" "current_notification_number=" << CompatUtility::GetCheckableNotificationNotificationNumber(checkable) << "\n"
"\t" "is_reachable=" << Convert::ToLong(checkable->IsReachable()) << "\n";
}
void InfluxdbWriter::CheckResultHandlerWQ(const Checkable::Ptr& checkable, const CheckResult::Ptr& cr)
{
AssertOnWorkQueue();
CONTEXT("Processing check result for '" + checkable->GetName() + "'");
if (!IcingaApplication::GetInstance()->GetEnablePerfdata() || !checkable->GetEnablePerfdata())
return;
Host::Ptr host;
Service::Ptr service;
tie(host, service) = GetHostService(checkable);
MacroProcessor::ResolverList resolvers;
if (service)
resolvers.emplace_back("service", service);
resolvers.emplace_back("host", host);
resolvers.emplace_back("icinga", IcingaApplication::GetInstance());
String prefix;
double ts = cr->GetExecutionEnd();
// Clone the template and perform an in-place macro expansion of measurement and tag values
Dictionary::Ptr tmpl_clean = service ? GetServiceTemplate() : GetHostTemplate();
Dictionary::Ptr tmpl = static_pointer_cast<Dictionary>(tmpl_clean->Clone());
tmpl->Set("measurement", MacroProcessor::ResolveMacros(tmpl->Get("measurement"), resolvers, cr));
Dictionary::Ptr tags = tmpl->Get("tags");
if (tags) {
ObjectLock olock(tags);
for (const Dictionary::Pair& pair : tags) {
String missing_macro;
Value value = MacroProcessor::ResolveMacros(pair.second, resolvers, cr, &missing_macro);
if (!missing_macro.IsEmpty())
continue;
tags->Set(pair.first, value);
}
}
Array::Ptr perfdata = cr->GetPerformanceData();
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, "InfluxdbWriter")
<< "Ignoring invalid perfdata value: " << val;
continue;
}
}
Dictionary::Ptr fields = new Dictionary();
fields->Set("value", pdv->GetValue());
if (GetEnableSendThresholds()) {
if (pdv->GetCrit())
fields->Set("crit", pdv->GetCrit());
if (pdv->GetWarn())
fields->Set("warn", pdv->GetWarn());
if (pdv->GetMin())
fields->Set("min", pdv->GetMin());
if (pdv->GetMax())
fields->Set("max", pdv->GetMax());
}
if (!pdv->GetUnit().IsEmpty()) {
fields->Set("unit", pdv->GetUnit());
}
SendMetric(tmpl, pdv->GetLabel(), fields, ts);
}
}
if (GetEnableSendMetadata()) {
Host::Ptr host;
Service::Ptr service;
tie(host, service) = GetHostService(checkable);
Dictionary::Ptr fields = new Dictionary();
if (service)
fields->Set("state", new InfluxdbInteger(service->GetState()));
else
fields->Set("state", new InfluxdbInteger(host->GetState()));
fields->Set("current_attempt", new InfluxdbInteger(checkable->GetCheckAttempt()));
fields->Set("max_check_attempts", new InfluxdbInteger(checkable->GetMaxCheckAttempts()));
fields->Set("state_type", new InfluxdbInteger(checkable->GetStateType()));
fields->Set("reachable", checkable->IsReachable());
fields->Set("downtime_depth", new InfluxdbInteger(checkable->GetDowntimeDepth()));
fields->Set("acknowledgement", new InfluxdbInteger(checkable->GetAcknowledgement()));
fields->Set("latency", cr->CalculateLatency());
fields->Set("execution_time", cr->CalculateExecutionTime());
SendMetric(tmpl, Empty, fields, ts);
}
}
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", CompatUtility::GetHostCurrentState(host));
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_hard_state", host->GetLastHardState());
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));
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(cr->CalculateLatency()));
fields->Set("execution_time", Convert::ToString(cr->CalculateExecutionTime()));
}
fields->Set("scheduled_downtime_depth", host->GetDowntimeDepth());
fields->Set("failure_prediction_enabled", Empty);
fields->Set("process_performance_data", CompatUtility::GetCheckableProcessPerformanceData(host));
fields->Set("obsess_over_host", Empty);
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));
fields->Set("original_attributes", JsonEncode(host->GetOriginalAttributes()));
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());
}
}
}