// RunHistory for a single alert key. Returns true if notifications were altered.
func (s *Schedule) runHistory(r *RunHistory, ak models.AlertKey, event *models.Event, silenced SilenceTester) (checkNotify bool, err error) {
event.Time = r.Start
a := s.Conf.Alerts[ak.Name()]
if a.UnknownsNormal && event.Status == models.StUnknown {
event.Status = models.StNormal
}
data := s.DataAccess.State()
err = data.TouchAlertKey(ak, utcNow())
if err != nil {
return
}
si := silenced(ak)
// get existing open incident if exists
var incident *models.IncidentState
incident, err = data.GetOpenIncident(ak)
if err != nil {
return
}
defer func() {
// save unless incident is new and closed (log alert)
if incident != nil && (incident.Id != 0 || incident.Open) {
_, err = data.UpdateIncidentState(incident)
} else {
err = data.SetUnevaluated(ak, event.Unevaluated) // if nothing to save, at least store the unevaluated state
}
}()
// If nothing is out of the ordinary we are done
if event.Status <= models.StNormal && incident == nil {
return
}
// if event is unevaluated, we are done also.
if incident != nil {
incident.Unevaluated = event.Unevaluated
}
if event.Unevaluated {
return
}
shouldNotify := false
newIncident := false
if incident == nil {
incident = NewIncident(ak)
newIncident = true
shouldNotify = true
}
// VICTOROPS INTEGRATION: Enables notification of incidents which have returned to normal (Sends normNotification defined in config)
if event.Status <= models.StNormal && (incident.CurrentStatus == models.StWarning || incident.CurrentStatus == models.StCritical) {
slog.Infof("TRIGGER_RESOLVED: from %s to %s", incident.CurrentStatus, event.Status)
shouldNotify = true
}
// VICTOROPS INTEGRATION: Enables notification of Incidents which have returned to normal but are now back to warning or critical. i.e. enable Flapping
if incident.CurrentStatus == models.StNormal && (event.Status == models.StCritical || event.Status == models.StWarning) {
slog.Infof("TRIGGER_REALERT: from %s to %s", incident.CurrentStatus, event.Status)
shouldNotify = true
}
// set state.Result according to event result
if event.Status == models.StCritical {
incident.Result = event.Crit
} else if event.Status == models.StWarning {
incident.Result = event.Warn
}
if event.Status > models.StNormal {
incident.LastAbnormalStatus = event.Status
incident.LastAbnormalTime = event.Time.UTC().Unix()
}
if event.Status > incident.WorstStatus {
incident.WorstStatus = event.Status
shouldNotify = true
}
if event.Status != incident.CurrentStatus {
incident.Events = append(incident.Events, *event)
}
incident.CurrentStatus = event.Status
//run a preliminary save on new incidents to get an id
if newIncident {
if a.Log || silencedOrIgnored(a, event, si) {
//a log or silenced/ignored alert will not need to be saved
} else {
incident.Id, err = s.DataAccess.State().UpdateIncidentState(incident)
if err != nil {
return
}
}
}
//render templates and open alert key if abnormal
if event.Status > models.StNormal {
s.executeTemplates(incident, event, a, r)
incident.Open = true
if a.Log {
incident.Open = false
}
}
// On state increase, clear old notifications and notify current.
// Do nothing if state did not change.
notify := func(ns *conf.Notifications) {
if a.Log {
lastLogTime := s.lastLogTimes[ak]
now := utcNow()
if now.Before(lastLogTime.Add(a.MaxLogFrequency)) {
return
}
s.lastLogTimes[ak] = now
}
nots := ns.Get(s.Conf, incident.AlertKey.Group())
for _, n := range nots {
s.Notify(incident, n)
checkNotify = true
}
}
notifyCurrent := func() {
//Auto close ignoreUnknowns for new incident.
if silencedOrIgnored(a, event, si) {
incident.Open = false
return
}
// VICTOROPS INTEGRATION
incident.NeedAck = false
switch event.Status {
case models.StCritical, models.StUnknown:
notify(a.CritNotification)
case models.StWarning:
notify(a.WarnNotification)
case models.StNormal:
// VICTOROPS INTEGRATION
incident.NeedAck = false
notify(a.NormNotification)
}
}
// lock while we change notifications.
s.Lock("RunHistory")
if shouldNotify {
incident.NeedAck = false
if err = s.DataAccess.Notifications().ClearNotifications(ak); err != nil {
return
}
notifyCurrent()
}
// finally close an open alert with silence once it goes back to normal.
if si := silenced(ak); si != nil && event.Status == models.StNormal {
go func(ak models.AlertKey) {
slog.Infof("auto close %s because was silenced", ak)
err := s.Action("bosun", "Auto close because was silenced.", models.ActionClose, ak)
if err != nil {
slog.Errorln(err)
}
}(ak)
}
s.Unlock()
return checkNotify, nil
}
// RunHistory for a single alert key. Returns true if notifications were altered.
func (s *Schedule) runHistory(r *RunHistory, ak models.AlertKey, event *models.Event, silenced SilenceTester) (checkNotify bool, err error) {
event.Time = r.Start
data := s.DataAccess.State()
err = data.TouchAlertKey(ak, time.Now())
if err != nil {
return
}
// get existing open incident if exists
incident, err := data.GetOpenIncident(ak)
if err != nil {
return
}
defer func() {
// save unless incident is new and closed (log alert)
if incident != nil && (incident.Id != 0 || incident.Open) {
err = data.UpdateIncidentState(incident)
} else {
err = data.SetUnevaluated(ak, event.Unevaluated) // if nothing to save, at least store the unevaluated state
}
}()
// If nothing is out of the ordinary we are done
if event.Status <= models.StNormal && incident == nil {
return
}
// if event is unevaluated, we are done also.
if incident != nil {
incident.Unevaluated = event.Unevaluated
}
if event.Unevaluated {
return
}
shouldNotify := false
if incident == nil {
incident = NewIncident(ak)
shouldNotify = true
}
// set state.Result according to event result
if event.Status == models.StCritical {
incident.Result = event.Crit
} else if event.Status == models.StWarning {
incident.Result = event.Warn
}
if event.Status > models.StNormal {
incident.LastAbnormalStatus = event.Status
incident.LastAbnormalTime = event.Time.UTC().Unix()
}
if event.Status > incident.WorstStatus {
incident.WorstStatus = event.Status
shouldNotify = true
}
if event.Status != incident.CurrentStatus {
incident.Events = append(incident.Events, *event)
}
incident.CurrentStatus = event.Status
a := s.Conf.Alerts[ak.Name()]
//render templates and open alert key if abnormal
if event.Status > models.StNormal {
s.executeTemplates(incident, event, a, r)
incident.Open = true
if a.Log {
incident.Open = false
}
}
// On state increase, clear old notifications and notify current.
// Do nothing if state did not change.
notify := func(ns *conf.Notifications) {
if a.Log {
lastLogTime := s.lastLogTimes[ak]
now := time.Now()
if now.Before(lastLogTime.Add(a.MaxLogFrequency)) {
return
}
s.lastLogTimes[ak] = now
}
nots := ns.Get(s.Conf, incident.AlertKey.Group())
for _, n := range nots {
s.Notify(incident, n)
checkNotify = true
}
}
notifyCurrent := func() {
si := silenced(ak)
//Auto close ignoreUnknowns for new incident.
if a.IgnoreUnknown && event.Status == models.StUnknown {
incident.Open = false
return
} else if si != nil && si.Forget && event.Status == models.StUnknown {
incident.Open = false
return
}
incident.NeedAck = true
switch event.Status {
case models.StCritical, models.StUnknown:
notify(a.CritNotification)
case models.StWarning:
notify(a.WarnNotification)
}
}
clearOld := func() {
incident.NeedAck = false
delete(s.Notifications, ak)
}
// lock while we change notifications.
s.Lock("RunHistory")
if shouldNotify {
clearOld()
notifyCurrent()
}
// finally close an open alert with silence once it goes back to normal.
if si := silenced(ak); si != nil && event.Status == models.StNormal {
go func(ak models.AlertKey) {
slog.Infof("auto close %s because was silenced", ak)
err := s.Action("bosun", "Auto close because was silenced.", models.ActionClose, ak)
if err != nil {
slog.Errorln(err)
}
}(ak)
}
s.Unlock()
return checkNotify, nil
}
