feat(redis): collect master slave link metrics (#851)

netdata · Sep 23, 2022 · 2fb5298 · 2fb5298
1 parent 770bc95
commit 2fb5298
Show file tree

Hide file tree

Showing 4 changed files with 48 additions and 5 deletions.
diff --git a/modules/redis/README.md b/modules/redis/README.md
@@ -44,6 +44,8 @@ It collects information and statistics about the server executing the following
 ### Replication
 
 - Connected replicas in `replicas`
+- Time elapsed since the last interaction with master in `seconds`
+- Time elapsed since the link between master and slave is down in `seconds`
 
 ### Persistence
 

diff --git a/modules/redis/charts.go b/modules/redis/charts.go
@@ -19,6 +19,8 @@ const (
 	prioNet
 
 	prioConnectedReplicas
+	prioMasterLastIOSinceTime
+	prioMasterLinkDownSinceTime
 
 	prioPersistenceRDBChanges
 	prioPersistenceRDBBgSaveNow
@@ -329,6 +331,28 @@ var (
 			{ID: "connected_slaves", Name: "connected"},
 		},
 	}
+	masterLastIOSinceTimeChart = module.Chart{
+		ID:       "master_last_io_since_time",
+		Title:    "Time elapsed since the last interaction with master",
+		Units:    "seconds",
+		Fam:      "replication",
+		Ctx:      "redis.master_last_io_since_time",
+		Priority: prioMasterLastIOSinceTime,
+		Dims: module.Dims{
+			{ID: "master_last_io_seconds_ago", Name: "time"},
+		},
+	}
+	masterLinkDownSinceTimeChart = module.Chart{
+		ID:       "master_link_down_since_stime",
+		Title:    "Time elapsed since the link between master and slave is down",
+		Units:    "seconds",
+		Fam:      "replication",
+		Ctx:      "redis.master_link_down_since_time",
+		Priority: prioMasterLinkDownSinceTime,
+		Dims: module.Dims{
+			{ID: "master_link_down_since_seconds", Name: "time"},
+		},
+	}
 )
 
 var (

diff --git a/modules/redis/collect_info.go b/modules/redis/collect_info.go
@@ -88,6 +88,12 @@ func (r *Redis) collectInfo(mx map[string]int64, info string) {
 	if has(mx, "keyspace_hits", "keyspace_misses") {
 		mx["keyspace_hit_rate"] = int64(calcKeyspaceHitRate(mx) * precision)
 	}
+	if has(mx, "master_last_io_seconds_ago") {
+		r.addReplSlaveChartsOnce.Do(r.addReplSlaveCharts)
+		if !has(mx, "master_link_down_since_seconds") {
+			mx["master_link_down_since_seconds"] = 0
+		}
+	}
 }
 
 var reKeyspaceValue = regexp.MustCompile(`^keys=(\d+),expires=(\d+)`)
@@ -220,6 +226,15 @@ func (r *Redis) addAOFCharts() {
 	}
 }
 
+func (r *Redis) addReplSlaveCharts() {
+	if err := r.Charts().Add(masterLastIOSinceTimeChart.Copy()); err != nil {
+		r.Warningf("error on adding '%s' chart", masterLastIOSinceTimeChart.ID)
+	}
+	if err := r.Charts().Add(masterLinkDownSinceTimeChart.Copy()); err != nil {
+		r.Warningf("error on adding '%s' chart", masterLinkDownSinceTimeChart.ID)
+	}
+}
+
 func has(m map[string]int64, key string, keys ...string) bool {
 	switch _, ok := m[key]; len(keys) {
 	case 0:

diff --git a/modules/redis/redis.go b/modules/redis/redis.go
@@ -30,10 +30,11 @@ func New() *Redis {
 			PingSamples: 5,
 		},
 
-		addAOFChartsOnce:  &sync.Once{},
-		pingSummary:       metrics.NewSummary(),
-		collectedCommands: make(map[string]bool),
-		collectedDbs:      make(map[string]bool),
+		addAOFChartsOnce:       &sync.Once{},
+		addReplSlaveChartsOnce: &sync.Once{},
+		pingSummary:            metrics.NewSummary(),
+		collectedCommands:      make(map[string]bool),
+		collectedDbs:           make(map[string]bool),
 	}
 }
 
@@ -56,7 +57,8 @@ type (
 		server  string
 		version *semver.Version
 
-		addAOFChartsOnce *sync.Once
+		addAOFChartsOnce       *sync.Once
+		addReplSlaveChartsOnce *sync.Once
 
 		pingSummary metrics.Summary