bosun: making save rely on fewer goroutines needing to acquire the lock.

.travis.yml

@@ -1,7 +1,7 @@
 language: go
 sudo: false
 go:
-  - tip
+  - 1.4.2
 
 notifications:
   email: false

cmd/bosun/sched/bolt.go

@@ -20,18 +20,27 @@ import (
 	"bosun.org/opentsdb"
 )
 
-var savePending bool
-
 func (s *Schedule) Save() {
-	go func() {
+	select {
+	case s.saveNeeded <- struct{}{}:
+	default:
+	}
+}
+
+func (s *Schedule) performSave() {
+	for range s.saveNeeded {
+		time.Sleep(5 * time.Second) // wait 5 seconds to throttle.
+
+		// if channel has an item on it, pull it off now to avoid re-saving later.
+		select {
+		case <-s.saveNeeded:
+		default:
+		}
+
 		s.Lock("Save")
 		defer s.Unlock()
-		if savePending {
-			return
-		}
-		savePending = true
-		time.AfterFunc(time.Second*5, s.save)
-	}()
+		s.save()
+	}
 }
 
 type counterWriter struct {
@@ -60,9 +69,6 @@ const (
 )
 
 func (s *Schedule) save() {
-	defer func() {
-		savePending = false
-	}()
 	if s.db == nil {
 		return
 	}

cmd/bosun/sched/sched.go

@@ -50,6 +50,7 @@ type Schedule struct {
 	maxIncidentId uint64
 	incidentLock  sync.Mutex
 	db            *bolt.DB
+	saveNeeded    chan struct{}
 }
 
 func init() {
@@ -412,6 +413,7 @@ func (s *Schedule) Init(c *conf.Conf) error {
 	s.Incidents = make(map[uint64]*Incident)
 	s.status = make(States)
 	s.Search = search.NewSearch()
+	s.saveNeeded = make(chan struct{}, 1)
 	if c.StateFile != "" {
 		s.db, err = bolt.Open(c.StateFile, 0600, nil)
 		if err != nil {
@@ -436,6 +438,9 @@ func Close() {
 }
 
 func (s *Schedule) Close() {
+	s.Lock("Close")
+	defer s.Unlock()
+	close(s.saveNeeded)
 	s.save()
 	if s.db != nil {
 		s.db.Close()
@@ -451,6 +456,7 @@ func (s *Schedule) Run() error {
 		go s.PingHosts()
 	}
 	go s.Poll()
+	go s.performSave()
 	interval := uint64(0)
 	for {
 		wait := time.After(s.Conf.CheckFrequency)