Merge pull request #58 from getlantern/metrics

Added facility for tracking and exposing basic database-level stats

cluster_follow.go

@@ -9,6 +9,7 @@ import (
 	"github.com/getlantern/wal"
 	"github.com/getlantern/zenodb/common"
 	"github.com/getlantern/zenodb/encoding"
+	"github.com/getlantern/zenodb/metrics"
 	"github.com/spaolacci/murmur3"
 	"hash"
 	"runtime"
@@ -36,9 +37,10 @@ type followSpec struct {
 
 type follower struct {
 	common.Follow
-	cb        func(data []byte, offset wal.Offset) error
-	entries   chan *walEntry
-	hasFailed int32
+	followerId int
+	cb         func(data []byte, offset wal.Offset) error
+	entries    chan *walEntry
+	hasFailed  int32
 }
 
 func (f *follower) read() {
@@ -69,6 +71,7 @@ func (f *follower) submit(entry *walEntry) {
 
 func (f *follower) markFailed() {
 	atomic.StoreInt32(&f.hasFailed, 1)
+	metrics.FollowerFailed(f.followerId)
 }
 
 func (f *follower) failed() bool {
@@ -109,6 +112,8 @@ func (db *DB) processFollowers() {
 	newlyJoinedStreams := make(map[string]bool)
 	onFollowerJoined := func(f *follower) {
 		nextFollowerID++
+		f.followerId = nextFollowerID
+		metrics.FollowerJoined(nextFollowerID, f.PartitionNumber)
 		log.Debugf("Follower joined: %d -> %d", nextFollowerID, f.PartitionNumber)
 		followers[nextFollowerID] = f
 
@@ -296,7 +301,9 @@ func (db *DB) processFollowers() {
 			stats = make([]int, db.opts.NumPartitions)
 
 			for _, f := range followers {
-				log.Debugf("Queued for follower %d: %v", f.PartitionNumber, humanize.Comma(int64(len(f.entries))))
+				queued := int64(len(f.entries))
+				metrics.QueuedForFollower(f.followerId, int(queued))
+				log.Debugf("Queued for follower %d: %v", f.PartitionNumber, humanize.Comma(queued))
 			}
 		}
 	}

metrics/metrics.go

@@ -0,0 +1,138 @@
+package metrics
+
+import (
+	"sort"
+	"sync"
+	"time"
+
+	"github.com/getlantern/wal"
+)
+
+var (
+	leaderStats    *LeaderStats
+	followerStats  map[int]*FollowerStats
+	partitionStats map[int]*PartitionStats
+
+	mx sync.RWMutex
+)
+
+func init() {
+	reset()
+}
+
+func reset() {
+	leaderStats = &LeaderStats{}
+	followerStats = make(map[int]*FollowerStats, 0)
+	partitionStats = make(map[int]*PartitionStats, 0)
+}
+
+// Stats are the overall stats
+type Stats struct {
+	Leader     *LeaderStats
+	Followers  sortedFollowerStats
+	Partitions sortedPartitionStats
+}
+
+// LeaderStats provides stats for the cluster leader
+type LeaderStats struct {
+	ConnectedFollowers  int
+	CurrentlyReadingWAL time.Time
+}
+
+// FollowerStats provides stats for a single follower
+type FollowerStats struct {
+	followerId int
+	Partition  int
+	Queued     int
+}
+
+// PartitionStats provides stats for a single partition
+type PartitionStats struct {
+	Partition    int
+	NumFollowers int
+}
+
+type sortedFollowerStats []*FollowerStats
+
+func (s sortedFollowerStats) Len() int      { return len(s) }
+func (s sortedFollowerStats) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
+func (s sortedFollowerStats) Less(i, j int) bool {
+	if s[i].Partition < s[j].Partition {
+		return true
+	}
+	return s[i].followerId < s[j].followerId
+}
+
+type sortedPartitionStats []*PartitionStats
+
+func (s sortedPartitionStats) Len() int      { return len(s) }
+func (s sortedPartitionStats) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
+func (s sortedPartitionStats) Less(i, j int) bool {
+	return s[i].Partition < s[j].Partition
+}
+
+func CurrentlyReadingWAL(offset wal.Offset) {
+	ts := offset.TS()
+	mx.Lock()
+	leaderStats.CurrentlyReadingWAL = ts
+	mx.Unlock()
+}
+
+// FollowerJoined records the fact that a follower joined the leader
+func FollowerJoined(followerID int, partition int) {
+	mx.Lock()
+	defer mx.Unlock()
+	leaderStats.ConnectedFollowers++
+	followerStats[followerID] = &FollowerStats{
+		followerId: followerID,
+		Partition:  partition,
+		Queued:     0,
+	}
+	ps := partitionStats[partition]
+	if ps == nil {
+		ps = &PartitionStats{Partition: partition}
+		partitionStats[partition] = ps
+	}
+	ps.NumFollowers++
+}
+
+// FollowerFailed records the fact that a follower failed (which is analogous to leaving)
+func FollowerFailed(followerID int) {
+	mx.Lock()
+	defer mx.Unlock()
+	// Only delete once
+	fs, found := followerStats[followerID]
+	if found {
+		leaderStats.ConnectedFollowers--
+		delete(followerStats, followerID)
+		partitionStats[fs.Partition].NumFollowers--
+	}
+}
+
+// QueuedForFollower records how many measurements are queued for a given Follower
+func QueuedForFollower(followerID int, queued int) {
+	mx.Lock()
+	defer mx.Unlock()
+	followerStats[followerID].Queued = queued
+}
+
+func GetStats() *Stats {
+	mx.RLock()
+	s := &Stats{
+		Leader:     leaderStats,
+		Followers:  make(sortedFollowerStats, 0, len(followerStats)),
+		Partitions: make(sortedPartitionStats, 0, len(partitionStats)),
+	}
+
+	for _, fs := range followerStats {
+		s.Followers = append(s.Followers, fs)
+	}
+	for _, ps := range partitionStats {
+		s.Partitions = append(s.Partitions, ps)
+	}
+	mx.RUnlock()
+
+	sort.Sort(s.Followers)
+	sort.Sort(s.Partitions)
+	return s
+}

metrics/metrics_test.go

@@ -0,0 +1,62 @@
+package metrics
+
+import (
+	"testing"
+	"time"
+
+	"github.com/getlantern/wal"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestMetrics(t *testing.T) {
+	reset()
+
+	ts := time.Now()
+	FollowerJoined(1, 1)
+	FollowerJoined(2, 1)
+	FollowerJoined(3, 2)
+	FollowerJoined(4, 2)
+	CurrentlyReadingWAL(wal.NewOffsetForTS(ts))
+	QueuedForFollower(1, 11)
+	QueuedForFollower(2, 22)
+	QueuedForFollower(3, 33)
+	QueuedForFollower(4, 44)
+
+	s := GetStats()
+	assert.Equal(t, 4, s.Leader.ConnectedFollowers)
+	assert.Equal(t, ts.Format(time.RFC3339), s.Leader.CurrentlyReadingWAL.Format(time.RFC3339))
+
+	assert.Equal(t, 1, s.Followers[0].Partition)
+	assert.Equal(t, 11, s.Followers[0].Queued)
+	assert.Equal(t, 1, s.Followers[1].Partition)
+	assert.Equal(t, 22, s.Followers[1].Queued)
+	assert.Equal(t, 2, s.Followers[2].Partition)
+	assert.Equal(t, 33, s.Followers[2].Queued)
+	assert.Equal(t, 2, s.Followers[3].Partition)
+	assert.Equal(t, 44, s.Followers[3].Queued)
+
+	assert.Equal(t, 1, s.Partitions[0].Partition)
+	assert.Equal(t, 2, s.Partitions[0].NumFollowers)
+	assert.Equal(t, 2, s.Partitions[1].Partition)
+	assert.Equal(t, 2, s.Partitions[1].NumFollowers)
+
+	// Fail a couple of followers. Fail each twice to make sure we don't double-
+	// subtract.
+	FollowerFailed(2)
+	FollowerFailed(2)
+	FollowerFailed(3)
+	FollowerFailed(3)
+
+	s = GetStats()
+	assert.Equal(t, 2, s.Leader.ConnectedFollowers)
+	assert.Equal(t, 1, s.Followers[0].Partition)
+	assert.Equal(t, 11, s.Followers[0].Queued)
+	assert.Equal(t, 2, s.Followers[1].Partition)
+	assert.Equal(t, 44, s.Followers[1].Queued)
+
+	assert.Equal(t, 1, s.Partitions[0].Partition)
+	assert.Equal(t, 1, s.Partitions[0].NumFollowers)
+	assert.Equal(t, 2, s.Partitions[1].Partition)
+	assert.Equal(t, 1, s.Partitions[1].NumFollowers)
+}

web/handler.go

@@ -117,6 +117,7 @@ func Configure(db *zenodb.DB, router *mux.Router, opts *Opts) error {
 	router.PathPrefix("/cached/{permalink}").HandlerFunc(h.cachedQuery)
 	router.PathPrefix("/favicon").Handler(http.NotFoundHandler())
 	router.PathPrefix("/report/{permalink}").HandlerFunc(h.index)
+	router.PathPrefix("/metrics").HandlerFunc(h.metrics)
 	router.PathPrefix("/").HandlerFunc(h.index)
 
 	return nil

web/metrics.go

@@ -0,0 +1,17 @@
+package web
+
+import (
+	"encoding/json"
+	"net/http"
+
+	"github.com/getlantern/zenodb/metrics"
+)
+
+func (h *handler) metrics(resp http.ResponseWriter, req *http.Request) {
+	if !h.authenticate(resp, req) {
+		resp.WriteHeader(http.StatusForbidden)
+		return
+	}
+
+	json.NewEncoder(resp).Encode(metrics.GetStats())
+}