// monitorReachability is a hostMonitoringFunc responsible for seeing if
// hosts are reachable or not. returns a slice of any errors that occur
func monitorReachability(settings *evergreen.Settings) []error {
evergreen.Logger.Logf(slogger.INFO, "Running reachability checks...")
// used to store any errors that occur
var errors []error
// fetch all hosts that have not been checked recently
// (> 10 minutes ago)
threshold := time.Now().Add(-ReachabilityCheckInterval)
hosts, err := host.Find(host.ByNotMonitoredSince(threshold))
if err != nil {
errors = append(errors, fmt.Errorf("error finding hosts not"+
" monitored recently: %v", err))
return errors
}
// check all of the hosts. continue on error so that other hosts can be
// checked successfully
for _, host := range hosts {
if err := checkHostReachability(host, settings); err != nil {
errors = append(errors, fmt.Errorf("error checking reachability"+
" for host %v: %v", host.Id, err))
continue
}
}
evergreen.Logger.Logf(slogger.INFO, "Finished running host reachability checks")
return errors
}
// monitorReachability is a hostMonitoringFunc responsible for seeing if
// hosts are reachable or not. returns a slice of any errors that occur
func monitorReachability(settings *evergreen.Settings) []error {
evergreen.Logger.Logf(slogger.INFO, "Running reachability checks...")
// used to store any errors that occur
var errors []error
// fetch all hosts that have not been checked recently
// (> 10 minutes ago)
threshold := time.Now().Add(-ReachabilityCheckInterval)
hosts, err := host.Find(host.ByNotMonitoredSince(threshold))
if err != nil {
errors = append(errors, fmt.Errorf("error finding hosts not monitored recently: %v", err))
return errors
}
workers := NumReachabilityWorkers
if len(hosts) < workers {
workers = len(hosts)
}
wg := sync.WaitGroup{}
wg.Add(workers)
hostsChan := make(chan host.Host, workers)
errChan := make(chan error, workers)
for i := 0; i < workers; i++ {
go func() {
defer wg.Done()
for host := range hostsChan {
if err := checkHostReachability(host, settings); err != nil {
errChan <- err
}
}
}()
}
errDone := make(chan struct{})
go func() {
defer close(errDone)
for err := range errChan {
errors = append(errors, fmt.Errorf("error checking reachability: %v", err))
}
}()
// check all of the hosts. continue on error so that other hosts can be
// checked successfully
for _, host := range hosts {
hostsChan <- host
}
close(hostsChan)
wg.Wait()
close(errChan)
<-errDone
return errors
}
