From 043489ed970f4a6e9c14afbda760c885a90fc6c3 Mon Sep 17 00:00:00 2001
From: Frederik Thuysbaert <fretb@users.noreply.github.com>
Date: Mon, 6 May 2024 11:52:24 +0200
Subject: [PATCH] feat(inputs.redis): Add latency percentiles metric (#15293)

---
 plugins/inputs/redis/README.md     | 17 ++++++++++-
 plugins/inputs/redis/redis.go      | 47 +++++++++++++++++++++++++++++-
 plugins/inputs/redis/redis_test.go | 20 +++++++++++++
 3 files changed, 82 insertions(+), 2 deletions(-)

diff --git a/plugins/inputs/redis/README.md b/plugins/inputs/redis/README.md
index 35e8b3528e07e..358a2d0e847f8 100644
--- a/plugins/inputs/redis/README.md
+++ b/plugins/inputs/redis/README.md
@@ -157,6 +157,12 @@ and the elapsed time since the last rdb save (rdb\_last\_save\_time\_elapsed).
   - usec(int, mircoseconds)
   - usec_per_call(float, microseconds)
 
+- redis_latency_percentiles_usec
+  - fields:
+    - p50(float, microseconds)
+    - p99(float, microseconds)
+    - p99.9(float, microseconds)
+
 - redis_replication
   - tags:
     - replication_role
@@ -184,7 +190,10 @@ and the elapsed time since the last rdb save (rdb\_last\_save\_time\_elapsed).
 - The redis_keyspace measurement has an additional database tag:
   - database
 
-- The redis_cmdstat measurement has an additional tag:
+- The redis_cmdstat measurement has an additional command tag:
+  - command
+
+- The redis_latency_percentiles_usec measurement has an additional command tag:
   - command
 
 ## Example Output
@@ -228,6 +237,12 @@ redis_command:
 redis_cmdstat,command=publish,host=host,port=6379,replication_role=master,server=localhost calls=569514i,failed_calls=0i,rejected_calls=0i,usec=9916334i,usec_per_call=17.41 1559227136000000000
 ```
 
+redis_latency_percentiles_usec:
+
+```text
+redis_latency_percentiles_usec,command=zadd,host=host,port=6379,replication_role=master,server=localhost p50=9.023,p99=28.031,p99.9=43.007 1559227136000000000
+```
+
 redis_error:
 
 ```text
diff --git a/plugins/inputs/redis/redis.go b/plugins/inputs/redis/redis.go
index b19cdd8a611a3..960495631318b 100644
--- a/plugins/inputs/redis/redis.go
+++ b/plugins/inputs/redis/redis.go
@@ -415,6 +415,11 @@ func gatherInfoOutput(
 				gatherCommandstateLine(name, kline, acc, tags)
 				continue
 			}
+			if section == "Latencystats" {
+				kline := strings.TrimSpace(parts[1])
+				gatherLatencystatsLine(name, kline, acc, tags)
+				continue
+			}
 			if section == "Replication" && replicationSlaveMetricPrefix.MatchString(name) {
 				kline := strings.TrimSpace(parts[1])
 				gatherReplicationLine(name, kline, acc, tags)
@@ -515,7 +520,7 @@ func gatherKeyspaceLine(
 //
 //	cmdstat_publish:calls=33791,usec=208789,usec_per_call=6.18
 //
-// Tag: cmdstat=publish; Fields: calls=33791i,usec=208789i,usec_per_call=6.18
+// Tag: command=publish; Fields: calls=33791i,usec=208789i,usec_per_call=6.18
 func gatherCommandstateLine(
 	name string,
 	line string,
@@ -557,6 +562,46 @@ func gatherCommandstateLine(
 	acc.AddFields("redis_cmdstat", fields, tags)
 }
 
+// Parse the special latency_percentiles_usec lines.
+// Example:
+//
+//	latency_percentiles_usec_zadd:p50=9.023,p99=28.031,p99.9=43.007
+//
+// Tag: command=zadd; Fields: p50=9.023,p99=28.031,p99.9=43.007
+func gatherLatencystatsLine(
+	name string,
+	line string,
+	acc telegraf.Accumulator,
+	globalTags map[string]string,
+) {
+	if !strings.HasPrefix(name, "latency_percentiles_usec") {
+		return
+	}
+
+	fields := make(map[string]interface{})
+	tags := make(map[string]string)
+	for k, v := range globalTags {
+		tags[k] = v
+	}
+	tags["command"] = strings.TrimPrefix(name, "latency_percentiles_usec_")
+	parts := strings.Split(line, ",")
+	for _, part := range parts {
+		kv := strings.Split(part, "=")
+		if len(kv) != 2 {
+			continue
+		}
+
+		switch kv[0] {
+		case "p50", "p99", "p99.9":
+			fval, err := strconv.ParseFloat(kv[1], 64)
+			if err == nil {
+				fields[kv[0]] = fval
+			}
+		}
+	}
+	acc.AddFields("redis_latency_percentiles_usec", fields, tags)
+}
+
 // Parse the special Replication line
 // Example:
 //
diff --git a/plugins/inputs/redis/redis_test.go b/plugins/inputs/redis/redis_test.go
index e5e588fd76e23..7e381412ffd16 100644
--- a/plugins/inputs/redis/redis_test.go
+++ b/plugins/inputs/redis/redis_test.go
@@ -258,6 +258,22 @@ func TestRedis_ParseMetrics(t *testing.T) {
 	}
 	acc.AssertContainsTaggedFields(t, "redis_cmdstat", cmdstatPublishFields, cmdstatPublishTags)
 
+	latencyZaddTags := map[string]string{"host": "redis.net", "replication_role": "master", "command": "zadd"}
+	latencyZaddFields := map[string]interface{}{
+		"p50":   float64(9.023),
+		"p99":   float64(28.031),
+		"p99.9": float64(43.007),
+	}
+	acc.AssertContainsTaggedFields(t, "redis_latency_percentiles_usec", latencyZaddFields, latencyZaddTags)
+
+	latencyHgetallTags := map[string]string{"host": "redis.net", "replication_role": "master", "command": "hgetall"}
+	latencyHgetallFields := map[string]interface{}{
+		"p50":   float64(11.007),
+		"p99":   float64(34.047),
+		"p99.9": float64(66.047),
+	}
+	acc.AssertContainsTaggedFields(t, "redis_latency_percentiles_usec", latencyHgetallFields, latencyHgetallTags)
+
 	replicationTags := map[string]string{
 		"host":             "redis.net",
 		"replication_role": "slave",
@@ -541,6 +557,10 @@ errorstat_NOSCRIPT:count=4
 errorstat_WRONGPASS:count=2
 errorstat_WRONGTYPE:count=30
 
+# Latencystats
+latency_percentiles_usec_zadd:p50=9.023,p99=28.031,p99.9=43.007
+latency_percentiles_usec_hgetall:p50=11.007,p99=34.047,p99.9=66.047
+
 # Keyspace
 db0:keys=2,expires=0,avg_ttl=0