in pkg/monitor/sqsevent/asg-lifecycle-event.go [60:115]
func (m SQSMonitor) asgTerminationToInterruptionEvent(event *EventBridgeEvent, message *sqs.Message) (*monitor.InterruptionEvent, error) {
lifecycleDetail := &LifecycleDetail{}
err := json.Unmarshal(event.Detail, lifecycleDetail)
if err != nil {
return nil, err
}
nodeInfo, err := m.getNodeInfo(lifecycleDetail.EC2InstanceID)
if err != nil {
return nil, err
}
interruptionEvent := monitor.InterruptionEvent{
EventID: fmt.Sprintf("asg-lifecycle-term-%x", event.ID),
Kind: SQSTerminateKind,
AutoScalingGroupName: lifecycleDetail.AutoScalingGroupName,
StartTime: event.getTime(),
NodeName: nodeInfo.Name,
IsManaged: nodeInfo.IsManaged,
InstanceID: lifecycleDetail.EC2InstanceID,
Description: fmt.Sprintf("ASG Lifecycle Termination event received. Instance will be interrupted at %s \n", event.getTime()),
}
interruptionEvent.PostDrainTask = func(interruptionEvent monitor.InterruptionEvent, _ node.Node) error {
_, err := m.ASG.CompleteLifecycleAction(&autoscaling.CompleteLifecycleActionInput{
AutoScalingGroupName: &lifecycleDetail.AutoScalingGroupName,
LifecycleActionResult: aws.String("CONTINUE"),
LifecycleHookName: &lifecycleDetail.LifecycleHookName,
LifecycleActionToken: &lifecycleDetail.LifecycleActionToken,
InstanceId: &lifecycleDetail.EC2InstanceID,
})
if err != nil {
if aerr, ok := err.(awserr.RequestFailure); ok && aerr.StatusCode() != 400 {
return err
}
}
log.Info().Msgf("Completed ASG Lifecycle Hook (%s) for instance %s",
lifecycleDetail.LifecycleHookName,
lifecycleDetail.EC2InstanceID)
errs := m.deleteMessages([]*sqs.Message{message})
if errs != nil {
return errs[0]
}
return nil
}
interruptionEvent.PreDrainTask = func(interruptionEvent monitor.InterruptionEvent, n node.Node) error {
err := n.TaintASGLifecycleTermination(interruptionEvent.NodeName, interruptionEvent.EventID)
if err != nil {
log.Err(err).Msgf("Unable to taint node with taint %s:%s", node.ASGLifecycleTerminationTaint, interruptionEvent.EventID)
}
return nil
}
return &interruptionEvent, nil
}