Add separate ro and rw slave counters

3scale-ops · Oct 4, 2023 · 9e4509f · 9e4509f
1 parent 7ec9bcc
commit 9e4509f
Show file tree

Hide file tree

Showing 2 changed files with 33 additions and 17 deletions.
diff --git a/pkg/redis/metrics/cluster_status_metrics.go b/pkg/redis/metrics/cluster_status_metrics.go
@@ -3,6 +3,7 @@ package metrics
 import (
 	"context"
 
+	"github.com/3scale/saas-operator/pkg/redis/client"
 	"github.com/3scale/saas-operator/pkg/redis/sharded"
 	"github.com/prometheus/client_golang/prometheus"
 	"sigs.k8s.io/controller-runtime/pkg/metrics"
@@ -13,14 +14,30 @@ var (
 		prometheus.GaugeOpts{
 			Name:      "server_info",
 			Namespace: "saas_redis_cluster_status",
-			Help:      `"redis server info"`,
+			Help:      "redis cluster member info",
 		},
 		[]string{"resource", "shard", "redis_server_host", "redis_server_alias", "role", "read_only"})
+	roSlaveCount = prometheus.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Name:      "ro_slave_count",
+			Namespace: "saas_redis_cluster_status",
+			Help:      "read-only slave count",
+		},
+		[]string{"resource", "shard"},
+	)
+	rwSlaveCount = prometheus.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Name:      "rw_slave_count",
+			Namespace: "saas_redis_cluster_status",
+			Help:      "read-write slave count",
+		},
+		[]string{"resource", "shard"},
+	)
 )
 
 func init() {
 	// Register custom metrics with the global prometheus registry
-	metrics.Registry.MustRegister(serverInfo)
+	metrics.Registry.MustRegister(serverInfo, roSlaveCount, rwSlaveCount)
 }
 
 func FromShardedCluster(ctx context.Context, cluster *sharded.Cluster, refresh bool, resource string) error {
@@ -33,6 +50,8 @@ func FromShardedCluster(ctx context.Context, cluster *sharded.Cluster, refresh b
 	}
 
 	for _, shard := range cluster.Shards {
+		roslave := 0
+		rwslave := 0
 
 		for _, server := range shard.Servers {
 			ro, ok := server.Config["slave-read-only"]
@@ -43,7 +62,18 @@ func FromShardedCluster(ctx context.Context, cluster *sharded.Cluster, refresh b
 				"redis_server_host": server.ID(), "redis_server_alias": server.GetAlias(),
 				"role": string(server.Role), "read_only": ro,
 			}).Set(float64(1))
+
+			if server.Role == client.Slave {
+				if ro == "yes" {
+					roslave++
+				} else {
+					rwslave++
+				}
+			}
 		}
+
+		roSlaveCount.With(prometheus.Labels{"resource": resource, "shard": shard.Name}).Set(float64(roslave))
+		rwSlaveCount.With(prometheus.Labels{"resource": resource, "shard": shard.Name}).Set(float64(rwslave))
 	}
 
 	return nil

diff --git a/pkg/redis/metrics/sentinel_metrics.go b/pkg/redis/metrics/sentinel_metrics.go
@@ -40,14 +40,6 @@ var (
 		},
 		[]string{"sentinel", "shard", "redis_server", "role"},
 	)
-	numSlaves = prometheus.NewGaugeVec(
-		prometheus.GaugeOpts{
-			Name:      "num_slaves",
-			Namespace: "saas_redis_sentinel",
-			Help:      `"sentinel master <name> num-slaves"`,
-		},
-		[]string{"sentinel", "shard", "redis_server", "role"},
-	)
 	numOtherSentinels = prometheus.NewGaugeVec(
 		prometheus.GaugeOpts{
 			Name:      "num_other_sentinels",
@@ -79,7 +71,7 @@ var (
 func init() {
 	// Register custom metrics with the global prometheus registry
 	metrics.Registry.MustRegister(
-		linkPendingCommands, lastOkPingReply, roleReportedTime, numSlaves,
+		linkPendingCommands, lastOkPingReply, roleReportedTime,
 		numOtherSentinels, masterLinkDownTime, slaveReplOffset,
 	)
 }
@@ -172,7 +164,6 @@ func (smg *SentinelMetricsGatherer) Stop() {
 	linkPendingCommands.Reset()
 	lastOkPingReply.Reset()
 	roleReportedTime.Reset()
-	numSlaves.Reset()
 	numOtherSentinels.Reset()
 	masterLinkDownTime.Reset()
 	slaveReplOffset.Reset()
@@ -199,10 +190,6 @@ func (smg *SentinelMetricsGatherer) gatherMetrics(ctx context.Context) error {
 			"redis_server": fmt.Sprintf("%s:%d", master.IP, master.Port), "role": master.RoleReported,
 		}).Set(float64(master.RoleReportedTime))
 
-		numSlaves.With(prometheus.Labels{"sentinel": smg.sentinelURI, "shard": master.Name,
-			"redis_server": fmt.Sprintf("%s:%d", master.IP, master.Port), "role": master.RoleReported,
-		}).Set(float64(master.NumSlaves))
-
 		numOtherSentinels.With(prometheus.Labels{"sentinel": smg.sentinelURI, "shard": master.Name,
 			"redis_server": fmt.Sprintf("%s:%d", master.IP, master.Port), "role": master.RoleReported,
 		}).Set(float64(master.NumOtherSentinels))
@@ -259,7 +246,6 @@ func cleanupMetrics(labels prometheus.Labels) {
 	linkPendingCommands.Delete(labels)
 	lastOkPingReply.Delete(labels)
 	roleReportedTime.Delete(labels)
-	numSlaves.Delete(labels)
 	numOtherSentinels.Delete(labels)
 	masterLinkDownTime.Delete(labels)
 	slaveReplOffset.Delete(labels)