feat(inputs.couchbase): Add failover metrics (#13825)

plugins/inputs/couchbase/README.md

@@ -43,10 +43,14 @@ See the [CONFIGURATION.md][CONFIGURATION.md] for more details.
   ## Whether to collect cluster-wide bucket statistics
   ## It is recommended to disable this in favor of node_stats
   ## to get a better view of the cluster.
-  cluster_bucket_stats = true
+  # cluster_bucket_stats = true
 
   ## Whether to collect bucket stats for each individual node
-  node_bucket_stats = false
+  # node_bucket_stats = false
+
+  ## List of additional stats to collect, choose from:
+  ##  * autofailover
+  # additional_stats = []
 ```
 
 ## Metrics
@@ -65,6 +69,21 @@ Fields:
 - memory_free (unit: bytes, example: 23181365248.0)
 - memory_total (unit: bytes, example: 64424656896.0)
 
+### couchbase_autofailover
+
+Tags:
+
+- cluster: sanitized string from `servers` configuration field
+  e.g.: `http://user:password@couchbase-0.example.com:8091/endpoint` becomes
+  `http://couchbase-0.example.com:8091/endpoint`
+
+Fields:
+
+- count (unit: int, example: 1)
+- enabled (unit: bool, example: true)
+- max_count (unit: int, example: 2)
+- timeout (unit: int, example: 72)
+
 ### couchbase_bucket and couchbase_node_bucket
 
 Tags:

plugins/inputs/couchbase/couchbase.go

@@ -4,6 +4,7 @@ package couchbase
 import (
 	_ "embed"
 	"encoding/json"
+	"fmt"
 	"net/http"
 	"regexp"
 	"sync"
@@ -13,6 +14,7 @@ import (
 
 	"github.com/influxdata/telegraf"
 	"github.com/influxdata/telegraf/filter"
+	"github.com/influxdata/telegraf/internal/choice"
 	"github.com/influxdata/telegraf/plugins/common/tls"
 	"github.com/influxdata/telegraf/plugins/inputs"
 )
@@ -21,12 +23,11 @@ import (
 var sampleConfig string
 
 type Couchbase struct {
-	Servers []string
+	Servers             []string `toml:"servers"`
-
 	BucketStatsIncluded []string `toml:"bucket_stats_included"`
-
 	ClusterBucketStats  bool     `toml:"cluster_bucket_stats"`
 	NodeBucketStats     bool     `toml:"node_bucket_stats"`
+	AdditionalStats     []string `toml:"additional_stats"`
 
 	bucketInclude filter.Filter
 	client        *http.Client
@@ -34,6 +35,13 @@ type Couchbase struct {
 	tls.ClientConfig
 }
 
+type autoFailover struct {
+	Count    int  `json:"count"`
+	Enabled  bool `json:"enabled"`
+	MaxCount int  `json:"maxCount"`
+	Timeout  int  `json:"timeout"`
+}
+
 var regexpURI = regexp.MustCompile(`(\S+://)?(\S+\:\S+@)`)
 
 func (*Couchbase) SampleConfig() string {
@@ -87,9 +95,8 @@ func (cb *Couchbase) gatherServer(acc telegraf.Accumulator, addr string) error {
 		acc.AddFields("couchbase_node", fields, tags)
 	}
 
-	for name, bucket := range pool.BucketMap {
 	cluster := regexpURI.ReplaceAllString(addr, "${1}")
-
+	for name, bucket := range pool.BucketMap {
 		if cb.ClusterBucketStats {
 			fields := cb.basicBucketStats(bucket.BasicStats)
 			tags := map[string]string{"cluster": cluster, "bucket": name}
@@ -117,9 +124,49 @@ func (cb *Couchbase) gatherServer(acc telegraf.Accumulator, addr string) error {
 		}
 	}
 
+	if choice.Contains("autofailover", cb.AdditionalStats) {
+		tags := map[string]string{"cluster": cluster}
+		fields, err := cb.gatherAutoFailoverStats(addr)
+		if err != nil {
+			return fmt.Errorf("unable to collect autofailover settings: %w", err)
+		}
+
+		acc.AddFields("couchbase_autofailover", fields, tags)
+	}
+
 	return nil
 }
 
+func (cb *Couchbase) gatherAutoFailoverStats(server string) (map[string]any, error) {
+	var fields map[string]any
+
+	url := server + "/settings/autoFailover"
+	req, err := http.NewRequest("GET", url, nil)
+	if err != nil {
+		return fields, err
+	}
+
+	r, err := cb.client.Do(req)
+	if err != nil {
+		return fields, err
+	}
+	defer r.Body.Close()
+
+	var stats autoFailover
+	if err := json.NewDecoder(r.Body).Decode(&stats); err != nil {
+		return fields, err
+	}
+
+	fields = map[string]any{
+		"count":     stats.Count,
+		"enabled":   stats.Enabled,
+		"max_count": stats.MaxCount,
+		"timeout":   stats.Timeout,
+	}
+
+	return fields, nil
+}
+
 // basicBucketStats gets the basic bucket statistics
 func (cb *Couchbase) basicBucketStats(basicStats map[string]interface{}) map[string]interface{} {
 	fields := make(map[string]interface{})

plugins/inputs/couchbase/couchbase_test.go

@@ -171,6 +171,50 @@ func TestGatherNodeOnly(t *testing.T) {
 	acc.AssertDoesNotContainMeasurement(t, "couchbase_bucket")
 }
 
+func TestGatherFailover(t *testing.T) {
+	faker := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/pools":
+			_, _ = w.Write(readJSON(t, "testdata/pools_response.json"))
+		case "/pools/default":
+			_, _ = w.Write(readJSON(t, "testdata/pools_default_response.json"))
+		case "/pools/default/buckets":
+			_, _ = w.Write(readJSON(t, "testdata/bucket_response.json"))
+		case "/settings/autoFailover":
+			_, _ = w.Write(readJSON(t, "testdata/settings_autofailover.json"))
+		default:
+			w.WriteHeader(http.StatusNotFound)
+		}
+	}))
+
+	cb := Couchbase{
+		Servers:            []string{faker.URL},
+		ClusterBucketStats: false,
+		NodeBucketStats:    false,
+		AdditionalStats:    []string{"autofailover"},
+	}
+	require.NoError(t, cb.Init())
+
+	var acc testutil.Accumulator
+	require.NoError(t, cb.gatherServer(&acc, faker.URL))
+	require.Equal(t, 0, len(acc.Errors))
+	require.Equal(t, 8, len(acc.Metrics))
+
+	var metric *testutil.Metric
+	for _, m := range acc.Metrics {
+		if m.Measurement == "couchbase_autofailover" {
+			metric = m
+			break
+		}
+	}
+
+	require.NotNil(t, metric)
+	require.Equal(t, 1, metric.Fields["count"])
+	require.Equal(t, true, metric.Fields["enabled"])
+	require.Equal(t, 2, metric.Fields["max_count"])
+	require.Equal(t, 72, metric.Fields["timeout"])
+}
+
 func readJSON(t *testing.T, jsonFilePath string) []byte {
 	data, err := os.ReadFile(jsonFilePath)
 	require.NoErrorf(t, err, "could not read from data file %s", jsonFilePath)

plugins/inputs/couchbase/sample.conf

@@ -25,7 +25,11 @@
   ## Whether to collect cluster-wide bucket statistics
   ## It is recommended to disable this in favor of node_stats
   ## to get a better view of the cluster.
-  cluster_bucket_stats = true
+  # cluster_bucket_stats = true
 
   ## Whether to collect bucket stats for each individual node
-  node_bucket_stats = false
+  # node_bucket_stats = false
+
+  ## List of additional stats to collect, choose from:
+  ##  * autofailover
+  # additional_stats = []

plugins/inputs/couchbase/testdata/settings_autofailover.json

@@ -0,0 +1,11 @@
+{
+    "enabled": true,
+    "timeout": 72,
+    "count": 1,
+    "failoverOnDataDiskIssues": {
+      "enabled": true,
+      "timePeriod": 89
+    },
+    "maxCount": 2,
+    "canAbortRebalance": true
+}