Address most comments and don't shuffle disappered nodes.

partitions.go

@@ -25,7 +25,7 @@ type partitions struct {
 	selected        map[int]bool
 	local           map[int]bool
 	remote          map[int][]string
-	old             map[int][]string
+	disappeared     map[int][]string
 	numMissing      int
 	ready           chan bool
 	readyClosed     bool
@@ -45,7 +45,7 @@ func watchPartitions(zkWatcher *zkWatcher, peers *peers, db, version string, num
 		replication:   replication,
 		local:         make(map[int]bool),
 		remote:        make(map[int][]string),
-		old:           make(map[int][]string, 1024),
+		disappeared:   make(map[int][]string, 1024),
 		ready:         make(chan bool),
 	}
 
@@ -113,6 +113,19 @@ func (p *partitions) updateLocalPartitions(local map[int]bool) {
 	}
 }
 
+func (p *partitions) deDupe(nodes []string) []string {
+	found := map[string]bool{}
+	dedupedNodes := make([]string, len(nodes))
+	for _, node := range nodes {
+		if !found[node] {
+			found[node] = true
+
+			dedupedNodes = append(dedupedNodes, node)
+		}
+	}
+	return dedupedNodes
+}
+
 func (p *partitions) updateRemotePartitions(nodes []string) {
 	if p.peers == nil {
 		return
@@ -131,23 +144,14 @@ func (p *partitions) updateRemotePartitions(nodes []string) {
 		}
 	}
 
-	// Keep track of old peers in case zookeeper goes away.
 	for partitionId, partition := range remote {
 		newPartition := make([]string, len(partition))
 		copy(newPartition, partition)
 
-		unDedupedPartition := append(newPartition, p.old[partitionId]...)
-		found := map[string]bool{}
-
-		// Shitty dedupe, iterate though the remote peers
-		for _, node := range unDedupedPartition {
-			if !found[node] {
-				found[node] = true
-				p.old[partitionId] = append([]string{node}, p.old[partitionId]...)
-			}
-		}
-		if len(p.old[partitionId]) >= 1024 {
-			p.old[partitionId] = p.old[partitionId][:1024]
+		unDedupedPartition := append(newPartition, p.disappeared[partitionId]...)
+		p.disappeared[partitionId] = p.deDupe(unDedupedPartition)
+		if len(p.disappeared[partitionId]) >= 1024 {
+			p.disappeared[partitionId] = p.disappeared[partitionId][:1024]
 		}
 	}
 
@@ -243,23 +247,21 @@ func (p *partitions) partitionZKNode(partition int) string {
 }
 
 // getPeers returns the list of peers who have the given partition available.
-func (p *partitions) getPeers(partition int) []string {
-	if p.peers == nil {
-		return nil
-	}
-
+func (p *partitions) getPeers(partition int) ([]string, []string) {
 	p.lock.RLock()
 	defer p.lock.RUnlock()
 
+	disappearedPeers := make([]string, 1024)
+	copy(disappearedPeers, p.disappeared[partition])
+
+	if p.peers == nil {
+		return nil, disappearedPeers
+	}
+
 	peers := make([]string, len(p.remote[partition]))
 	copy(peers, p.remote[partition])
 
-	oldPeers := make([]string, 1024)
-	copy(oldPeers, p.old[partition])
-	// Append old peers to peer list, in case of Zookeeper issues.
-	peers = append(peers, oldPeers...)
-
-	return peers
+	return peers, disappearedPeers
 }
 
 // partitionId returns a string id for the given partition, to be used for the

serve.go

@@ -120,12 +120,11 @@ func (vs *version) serveError(w http.ResponseWriter, key string, err error) {
 	w.WriteHeader(http.StatusInternalServerError)
 }
 
-func shuffle(vs []string) []string {
+func shuffle(vs []string, disappeared []string) []string {
 	shuffled := make([]string, len(vs))
 	perm := rand.Perm(len(vs))
 	for i, v := range perm {
 		shuffled[v] = vs[i]
 	}
-
-	return shuffled
+	return append(shuffled, disappeared...)
 }