gatus/watchdog/alerting.go

85 lines
3.5 KiB
Go
Raw Permalink Normal View History

package watchdog
2020-09-17 01:26:19 +02:00
import (
"errors"
"log"
"os"
2022-06-21 03:25:14 +02:00
"github.com/TwiN/gatus/v4/alerting"
"github.com/TwiN/gatus/v4/core"
2020-09-17 01:26:19 +02:00
)
// HandleAlerting takes care of alerts to resolve and alerts to trigger based on result success or failure
func HandleAlerting(endpoint *core.Endpoint, result *core.Result, alertingConfig *alerting.Config, debug bool) {
if alertingConfig == nil {
2020-09-17 01:26:19 +02:00
return
}
if result.Success {
handleAlertsToResolve(endpoint, result, alertingConfig, debug)
2020-09-17 01:26:19 +02:00
} else {
handleAlertsToTrigger(endpoint, result, alertingConfig, debug)
2020-09-17 01:26:19 +02:00
}
}
func handleAlertsToTrigger(endpoint *core.Endpoint, result *core.Result, alertingConfig *alerting.Config, debug bool) {
endpoint.NumberOfSuccessesInARow = 0
endpoint.NumberOfFailuresInARow++
for _, endpointAlert := range endpoint.Alerts {
// If the alert hasn't been triggered, move to the next one
if !endpointAlert.IsEnabled() || endpointAlert.FailureThreshold > endpoint.NumberOfFailuresInARow {
2020-09-17 01:26:19 +02:00
continue
}
if endpointAlert.Triggered {
if debug {
log.Printf("[watchdog][handleAlertsToTrigger] Alert for endpoint=%s with description='%s' has already been TRIGGERED, skipping", endpoint.Name, endpointAlert.GetDescription())
2020-09-17 01:26:19 +02:00
}
continue
}
alertProvider := alertingConfig.GetAlertingProviderByAlertType(endpointAlert.Type)
if alertProvider != nil && alertProvider.IsValid() {
log.Printf("[watchdog][handleAlertsToTrigger] Sending %s alert because alert for endpoint=%s with description='%s' has been TRIGGERED", endpointAlert.Type, endpoint.Name, endpointAlert.GetDescription())
var err error
if os.Getenv("MOCK_ALERT_PROVIDER") == "true" {
if os.Getenv("MOCK_ALERT_PROVIDER_ERROR") == "true" {
err = errors.New("error")
}
} else {
err = alertProvider.Send(endpoint, endpointAlert, result, false)
}
2020-09-17 01:26:19 +02:00
if err != nil {
log.Printf("[watchdog][handleAlertsToTrigger] Failed to send an alert for endpoint=%s: %s", endpoint.Name, err.Error())
2020-09-17 01:26:19 +02:00
} else {
endpointAlert.Triggered = true
2020-09-17 01:26:19 +02:00
}
} else {
log.Printf("[watchdog][handleAlertsToResolve] Not sending alert of type=%s despite being TRIGGERED, because the provider wasn't configured properly", endpointAlert.Type)
2020-09-17 01:26:19 +02:00
}
}
}
func handleAlertsToResolve(endpoint *core.Endpoint, result *core.Result, alertingConfig *alerting.Config, debug bool) {
endpoint.NumberOfSuccessesInARow++
for _, endpointAlert := range endpoint.Alerts {
if !endpointAlert.IsEnabled() || !endpointAlert.Triggered || endpointAlert.SuccessThreshold > endpoint.NumberOfSuccessesInARow {
2020-09-17 01:26:19 +02:00
continue
}
// Even if the alert provider returns an error, we still set the alert's Triggered variable to false.
// Further explanation can be found on Alert's Triggered field.
endpointAlert.Triggered = false
if !endpointAlert.IsSendingOnResolved() {
2020-09-17 01:26:19 +02:00
continue
}
alertProvider := alertingConfig.GetAlertingProviderByAlertType(endpointAlert.Type)
if alertProvider != nil && alertProvider.IsValid() {
log.Printf("[watchdog][handleAlertsToResolve] Sending %s alert because alert for endpoint=%s with description='%s' has been RESOLVED", endpointAlert.Type, endpoint.Name, endpointAlert.GetDescription())
2021-12-03 03:05:17 +01:00
err := alertProvider.Send(endpoint, endpointAlert, result, true)
2020-09-17 01:26:19 +02:00
if err != nil {
log.Printf("[watchdog][handleAlertsToResolve] Failed to send an alert for endpoint=%s: %s", endpoint.Name, err.Error())
2020-09-17 01:26:19 +02:00
}
} else {
log.Printf("[watchdog][handleAlertsToResolve] Not sending alert of type=%s despite being RESOLVED, because the provider wasn't configured properly", endpointAlert.Type)
2020-09-17 01:26:19 +02:00
}
}
endpoint.NumberOfFailuresInARow = 0
2020-09-17 01:26:19 +02:00
}