stats: export vttablet_tablet_type to prometheus

changelog/19.0/19.0.0/summary.md

@@ -6,6 +6,8 @@
   - **[Deprecations and Deletions](#deprecations-and-deletions)**
   - **[Docker](#docker)**
     - [New MySQL Image](#mysql-image)
+  - **[New stats](#new-stats)**
+    - [VTTablet tablet type for Prometheus](#vttablet-tablet-type-for-prometheus)
 
 ## <a id="major-changes"/>Major Changes
 
@@ -21,3 +23,21 @@ In `v19.0` the Vitess team is shipping a new image: `vitess/mysql`.
 This lightweight image is a replacement of `vitess/lite` to only run `mysqld`.
 
 Several tags are available to let you choose what version of MySQL you want to use: `vitess/mysql:8.0.30`, `vitess/mysql:8.0.34`.
+
+### <a id="new-stats"/> New stats
+
+
+#### <a id="vttablet-tablet-type-for-prometheus"/> VTTablet tablet type for Prometheus
+
+VTTablet publishes the `TabletType` status which reports the current type of the tablet. For example, here's what it looks like in the local cluster example after setting up the initial cluster:
+
+```
+"TabletType": "primary"
+```
+
+Prior to v19, this data was not available via the Prometheus backend. In v19, this is also published as a Prometheus gauge with a `tablet_type` label and a value of `1.0`. For example:
+
+```
+$ curl -s localhost:15100/metrics | grep tablet_type{
+vttablet_tablet_type{tablet_type="primary"} 1
+```

go/stats/export.go

@@ -139,6 +139,47 @@ func pushOne(name string, v Variable) error {
 	return backend.PushOne(name, v)
 }
 
+// StringValueWithLabel is string publisher.
+//
+// The valueLabel is intended for use by backends which must export a stats in
+// numeric form, and is not used by expvars.
+//
+// Backends which need to provide a numeric value can set a constant value of 1
+// (or whatever is appropriate for the backend).
+type StringValueWithLabel struct {
+	*StringValue
+	help  string
+	label string
+}
+
+// Help returns the descriptive help message.
+func (s StringValueWithLabel) Help() string {
+	return s.help
+}
+
+// Label returns the value label.
+func (s StringValueWithLabel) Label() string {
+	return s.label
+}
+
+// NewStringValueWithLabel creates a new StringFuncWithValueLabel.
+func NewStringValueWithLabel(name, help string, label, initialValue string) *StringValueWithLabel {
+	s := NewStringValue("" /* omit name to avoid publish conflicts */)
+	s.Set(initialValue)
+
+	t := &StringValueWithLabel{
+		StringValue: s,
+		help:        help,
+		label:       label,
+	}
+
+	if name != "" {
+		publish(name, t)
+	}
+
+	return t
+}
+
 // StringMapFuncWithMultiLabels is a multidimensional string map publisher.
 //
 // Map keys are compound names made with joining multiple strings with '.',
@@ -270,36 +311,36 @@ func (f FloatFunc) String() string {
 	return strconv.FormatFloat(f(), 'g', -1, 64)
 }
 
-// String is expvar.String+Get+hook
-type String struct {
+// StringValue is expvar.StringValue+Get+hook
+type StringValue struct {
 	mu sync.Mutex
 	s  string
 }
 
-// NewString returns a new String
-func NewString(name string) *String {
-	v := new(String)
+// NewStringValue returns a new String
+func NewStringValue(name string) *StringValue {
+	v := new(StringValue)
 	publish(name, v)
 	return v
 }
 
 // Set sets the value
-func (v *String) Set(value string) {
+func (v *StringValue) Set(value string) {
 	v.mu.Lock()
 	v.s = value
 	v.mu.Unlock()
 }
 
 // Get returns the value
-func (v *String) Get() string {
+func (v *StringValue) Get() string {
 	v.mu.Lock()
 	s := v.s
 	v.mu.Unlock()
 	return s
 }
 
 // String is the implementation of expvar.var
-func (v *String) String() string {
+func (v *StringValue) String() string {
 	return strconv.Quote(v.Get())
 }
 

go/stats/export_test.go

@@ -44,13 +44,13 @@ func TestNoHook(t *testing.T) {
 
 func TestString(t *testing.T) {
 	var gotname string
-	var gotv *String
+	var gotv *StringValue
 	clearStats()
 	Register(func(name string, v expvar.Var) {
 		gotname = name
-		gotv = v.(*String)
+		gotv = v.(*StringValue)
 	})
-	v := NewString("String")
+	v := NewStringValue("String")
 	if gotname != "String" {
 		t.Errorf("want String, got %s", gotname)
 	}
@@ -189,3 +189,18 @@ func TestStringMapWithMultiLabels(t *testing.T) {
 
 	require.Equal(t, c.ValueLabel(), "ccc")
 }
+
+func TestStringValueWithLabel(t *testing.T) {
+	clearStats()
+	c := NewStringValueWithLabel("stringValue1", "help", "aaa", "ccc")
+
+	s := c.String()
+	require.Equal(t, "\"ccc\"", s)
+
+	label := c.Label()
+	require.Equal(t, "aaa", label)
+
+	c.Set("ddd")
+	s = c.String()
+	require.Equal(t, "\"ddd\"", s)
+}

go/stats/prometheusbackend/collectors.go

@@ -431,3 +431,36 @@ func (c *stringMapFuncWithMultiLabelsCollector) Collect(ch chan<- prometheus.Met
 		}
 	}
 }
+
+type stringValueWithLabelCollector struct {
+	svl  *stats.StringValueWithLabel
+	desc *prometheus.Desc
+}
+
+func newStringValueWithLabelCollector(svl *stats.StringValueWithLabel, name string) {
+	c := &stringValueWithLabelCollector{
+		svl: svl,
+		desc: prometheus.NewDesc(
+			name,
+			svl.Help(),
+			labelsToSnake([]string{svl.Label()}),
+			nil),
+	}
+
+	prometheus.MustRegister(c)
+}
+
+// Describe implements Collector.
+func (c *stringValueWithLabelCollector) Describe(ch chan<- *prometheus.Desc) {
+	ch <- c.desc
+}
+
+// Collect implements Collector.
+func (c *stringValueWithLabelCollector) Collect(ch chan<- prometheus.Metric) {
+	metric, err := prometheus.NewConstMetric(c.desc, prometheus.GaugeValue, 1.0, c.svl.Get())
+	if err != nil {
+		log.Errorf("Error adding metric: %s", c.desc)
+	} else {
+		ch <- metric
+	}
+}

go/stats/prometheusbackend/prometheusbackend.go

@@ -87,7 +87,9 @@ func (be PromBackend) publishPrometheusMetric(name string, v expvar.Var) {
 		newHistogramCollector(st, be.buildPromName(name))
 	case *stats.StringMapFuncWithMultiLabels:
 		newStringMapFuncWithMultiLabelsCollector(st, be.buildPromName(name))
-	case *stats.String, stats.StringFunc, stats.StringMapFunc, *stats.Rates, *stats.RatesFunc:
+	case *stats.StringValueWithLabel:
+		newStringValueWithLabelCollector(st, be.buildPromName(name))
+	case *stats.StringValue, stats.StringFunc, stats.StringMapFunc, *stats.Rates, *stats.RatesFunc:
 		// Silently ignore these types since they don't make sense to
 		// export to Prometheus' data model.
 	default:

go/stats/prometheusbackend/prometheusbackend_test.go

@@ -258,6 +258,16 @@ func TestPrometheusStringMapFuncWithMultiLabels(t *testing.T) {
 	checkHandlerForMetricWithMultiLabels(t, name, allLabels, []string{"bar", "baz", "world"}, 1)
 }
 
+func TestPrometheusStringValueWithLabel(t *testing.T) {
+	name := "blah_stringvaluewithlabel"
+	label := "label"
+	value := "value"
+
+	stats.NewStringValueWithLabel(name, "help", label, value)
+
+	checkHandlerForMetricWithMultiLabels(t, name, []string{label}, []string{value}, 1)
+}
+
 func checkHandlerForMetricWithMultiLabels(t *testing.T, metric string, labels []string, labelValues []string, value int64) {
 	response := testMetricsHandler(t)
 

go/stats/statsd/statsd.go

@@ -202,7 +202,7 @@ func (sb StatsBackend) addExpVar(kv expvar.KeyValue) {
 				}
 			}
 		}
-	case *stats.String:
+	case *stats.StringValue:
 		if k == "BuildGitRev" {
 			buildGitRecOnce.Do(func() {
 				checksum := crc32.ChecksumIEEE([]byte(v.Get()))

go/vt/grpccommon/options.go

@@ -54,5 +54,5 @@ func MaxMessageSize() int {
 }
 
 func init() {
-	stats.NewString("GrpcVersion").Set(grpc.Version)
+	stats.NewStringValue("GrpcVersion").Set(grpc.Version)
 }

go/vt/servenv/buildinfo.go

@@ -118,15 +118,15 @@ func init() {
 		goArch:             runtime.GOARCH,
 		version:            versionName,
 	}
-	stats.NewString("BuildHost").Set(AppVersion.buildHost)
-	stats.NewString("BuildUser").Set(AppVersion.buildUser)
+	stats.NewStringValue("BuildHost").Set(AppVersion.buildHost)
+	stats.NewStringValue("BuildUser").Set(AppVersion.buildUser)
 	stats.NewGauge("BuildTimestamp", "build timestamp").Set(AppVersion.buildTime)
-	stats.NewString("BuildGitRev").Set(AppVersion.buildGitRev)
-	stats.NewString("BuildGitBranch").Set(AppVersion.buildGitBranch)
+	stats.NewStringValue("BuildGitRev").Set(AppVersion.buildGitRev)
+	stats.NewStringValue("BuildGitBranch").Set(AppVersion.buildGitBranch)
 	stats.NewGauge("BuildNumber", "build number").Set(AppVersion.jenkinsBuildNumber)
-	stats.NewString("GoVersion").Set(AppVersion.goVersion)
-	stats.NewString("GoOS").Set(AppVersion.goOS)
-	stats.NewString("GoArch").Set(AppVersion.goArch)
+	stats.NewStringValue("GoVersion").Set(AppVersion.goVersion)
+	stats.NewStringValue("GoOS").Set(AppVersion.goOS)
+	stats.NewStringValue("GoArch").Set(AppVersion.goArch)
 
 	buildLabels := []string{"BuildHost", "BuildUser", "BuildTimestamp", "BuildGitRev", "BuildGitBranch", "BuildNumber"}
 	buildValues := []string{

go/vt/servenv/exporter_test.go

@@ -624,7 +624,7 @@ func TestHistogram(t *testing.T) {
 
 func TestPublish(t *testing.T) {
 	ebd := NewExporter("", "")
-	s := stats.NewString("")
+	s := stats.NewStringValue("")
 	ebd.Publish("gpub", s)
 	s.Set("1")
 	assert.Equal(t, `"1"`, expvar.Get("gpub").String())

go/vt/vttablet/tabletmanager/restore.go

@@ -59,8 +59,8 @@ var (
 	restoreConcurrency     = 4
 	waitForBackupInterval  time.Duration
 
-	statsRestoreBackupTime     *stats.String
-	statsRestoreBackupPosition *stats.String
+	statsRestoreBackupTime     *stats.StringValue
+	statsRestoreBackupPosition *stats.StringValue
 )
 
 func registerRestoreFlags(fs *pflag.FlagSet) {
@@ -116,8 +116,8 @@ func init() {
 	servenv.OnParseFor("vtcombo", registerPointInTimeRestoreFlags)
 	servenv.OnParseFor("vttablet", registerPointInTimeRestoreFlags)
 
-	statsRestoreBackupTime = stats.NewString("RestoredBackupTime")
-	statsRestoreBackupPosition = stats.NewString("RestorePosition")
+	statsRestoreBackupTime = stats.NewStringValue("RestoredBackupTime")
+	statsRestoreBackupPosition = stats.NewStringValue("RestorePosition")
 }
 
 // RestoreData is the main entry point for backup restore.

go/vt/vttablet/tabletmanager/tm_init.go

@@ -103,7 +103,7 @@ func registerInitFlags(fs *pflag.FlagSet) {
 
 var (
 	// statsTabletType is set to expose the current tablet type.
-	statsTabletType *stats.String
+	statsTabletType *stats.StringValueWithLabel
 
 	// statsTabletTypeCount exposes the current tablet type as a label,
 	// with the value counting the occurrences of the respective tablet type.
@@ -116,11 +116,11 @@ var (
 	// statsIsInSrvKeyspace is set to 1 (true), 0 (false) whether the tablet is in the serving keyspace
 	statsIsInSrvKeyspace *stats.Gauge
 
-	statsKeyspace      = stats.NewString("TabletKeyspace")
-	statsShard         = stats.NewString("TabletShard")
-	statsKeyRangeStart = stats.NewString("TabletKeyRangeStart")
-	statsKeyRangeEnd   = stats.NewString("TabletKeyRangeEnd")
-	statsAlias         = stats.NewString("TabletAlias")
+	statsKeyspace      = stats.NewStringValue("TabletKeyspace")
+	statsShard         = stats.NewStringValue("TabletShard")
+	statsKeyRangeStart = stats.NewStringValue("TabletKeyRangeStart")
+	statsKeyRangeEnd   = stats.NewStringValue("TabletKeyRangeEnd")
+	statsAlias         = stats.NewStringValue("TabletAlias")
 
 	// The following variables can be changed to speed up tests.
 	mysqlPortRetryInterval       = 1 * time.Second
@@ -131,7 +131,7 @@ func init() {
 	servenv.OnParseFor("vtcombo", registerInitFlags)
 	servenv.OnParseFor("vttablet", registerInitFlags)
 
-	statsTabletType = stats.NewString("TabletType")
+	statsTabletType = stats.NewStringValueWithLabel("TabletType", "Current tablet type", "TabletType", "")
 	statsTabletTypeCount = stats.NewCountersWithSingleLabel("TabletTypeCount", "Number of times the tablet changed to the labeled type", "type")
 	statsBackupIsRunning = stats.NewGaugesWithMultiLabels("BackupIsRunning", "Whether a backup is running", []string{"mode"})
 	statsIsInSrvKeyspace = stats.NewGauge("IsInSrvKeyspace", "Whether the vttablet is in the serving keyspace (1 = true / 0 = false)")