stellar · paulbellamy · May 18, 2021 · May 18, 2021 · May 18, 2021 · May 18, 2021
diff --git a/services/horizon/internal/db2/history/orderbook.go b/services/horizon/internal/db2/history/orderbook.go
@@ -6,6 +6,7 @@ import (
 	"math/big"
 
 	"github.com/stellar/go/amount"
+	"github.com/stellar/go/support/db"
 	"github.com/stellar/go/support/errors"
 	"github.com/stellar/go/xdr"
 )
@@ -101,6 +102,8 @@ func (q *Q) GetOrderBookSummary(ctx context.Context, sellingAsset, buyingAsset x
 			LIMIT $3
 		)
 	`
+	// Add explicit query type for prometheus metrics, since we use raw sql.
+	ctx = context.WithValue(ctx, &db.QueryTypeContextKey, db.SelectQueryType)
 	err = q.SelectRaw(ctx, &levels, selectPriceLevels, selling, buying, maxPriceLevels)
 	if err != nil {
 		return result, errors.Wrap(err, "cannot select price levels")

diff --git a/services/horizon/internal/db2/history/trade.go b/services/horizon/internal/db2/history/trade.go
@@ -8,6 +8,7 @@ import (
 	sq "github.com/Masterminds/squirrel"
 
 	"github.com/stellar/go/services/horizon/internal/db2"
+	"github.com/stellar/go/support/db"
 	"github.com/stellar/go/support/errors"
 	"github.com/stellar/go/xdr"
 )
@@ -195,6 +196,8 @@ func (q *TradesQ) Select(ctx context.Context, dest interface{}) error {
 		return errors.New("TradesQ.Page call is required before calling Select")
 	}
 
+	// Add explicit query type for prometheus metrics, since we use raw sql.
+	ctx = context.WithValue(ctx, &db.QueryTypeContextKey, db.SelectQueryType)
 	if q.rawSQL != "" {
 		q.Err = q.parent.SelectRaw(ctx, dest, q.rawSQL, q.rawArgs...)
 	} else {

diff --git a/services/horizon/internal/init.go b/services/horizon/internal/init.go
@@ -18,15 +18,15 @@ import (
 	"github.com/stellar/go/support/log"
 )
 
-func mustNewDBSession(subsystem db.Subsystem, databaseURL string, maxIdle, maxOpen int, registry *prometheus.Registry) db.SessionInterface {
+func mustNewDBSession(subservice db.Subservice, databaseURL string, maxIdle, maxOpen int, registry *prometheus.Registry) db.SessionInterface {
 	session, err := db.Open("postgres", databaseURL)
 	if err != nil {
 		log.Fatalf("cannot open Horizon DB: %v", err)
 	}
 
 	session.DB.SetMaxIdleConns(maxIdle)
 	session.DB.SetMaxOpenConns(maxOpen)
-	return db.RegisterMetrics(session, "horizon", subsystem, registry)
+	return db.RegisterMetrics(session, "horizon", subservice, registry)
 }
 
 func mustInitHorizonDB(app *App) {
@@ -44,7 +44,7 @@ func mustInitHorizonDB(app *App) {
 	}
 
 	app.historyQ = &history.Q{mustNewDBSession(
-		db.HistorySubsystem,
+		db.HistorySubservice,
 		app.config.DatabaseURL,
 		maxIdle,
 		maxOpen,
@@ -57,12 +57,12 @@ func initIngester(app *App) {
 	var coreSession db.SessionInterface
 	if !app.config.EnableCaptiveCoreIngestion {
 		coreSession = mustNewDBSession(
-			db.CoreSubsystem, app.config.StellarCoreDatabaseURL, ingest.MaxDBConnections, ingest.MaxDBConnections, app.prometheusRegistry)
+			db.CoreSubservice, app.config.StellarCoreDatabaseURL, ingest.MaxDBConnections, ingest.MaxDBConnections, app.prometheusRegistry)
 	}
 	app.ingester, err = ingest.NewSystem(ingest.Config{
 		CoreSession: coreSession,
 		HistorySession: mustNewDBSession(
-			db.IngestSubsystem, app.config.DatabaseURL, ingest.MaxDBConnections, ingest.MaxDBConnections, app.prometheusRegistry,
+			db.IngestSubservice, app.config.DatabaseURL, ingest.MaxDBConnections, ingest.MaxDBConnections, app.prometheusRegistry,
 		),
 		NetworkPassphrase: app.config.NetworkPassphrase,
 		// TODO:

diff --git a/support/db/metrics.go b/support/db/metrics.go
@@ -16,11 +16,11 @@ type CtxKey string
 var RouteContextKey = CtxKey("route")
 var QueryTypeContextKey = CtxKey("query_type")
 
-type Subsystem string
+type Subservice string
 
-var CoreSubsystem = Subsystem("core")
-var HistorySubsystem = Subsystem("history")
-var IngestSubsystem = Subsystem("ingest")
+var CoreSubservice = Subservice("core")
+var HistorySubservice = Subservice("history")
+var IngestSubservice = Subservice("ingest")
 
 type QueryType string
 
@@ -55,43 +55,53 @@ type SessionWithMetrics struct {
 	maxLifetimeClosedCounter prometheus.CounterFunc
 }
 
-func RegisterMetrics(base *Session, namespace string, sub Subsystem, registry *prometheus.Registry) SessionInterface {
-	subsystem := fmt.Sprintf("db_%s", sub)
+func RegisterMetrics(base *Session, namespace string, sub Subservice, registry *prometheus.Registry) SessionInterface {
 	s := &SessionWithMetrics{
 		SessionInterface: base,
 		registry:         registry,
 	}
 
 	s.queryCounter = prometheus.NewCounterVec(
-		prometheus.CounterOpts{Namespace: namespace, Subsystem: subsystem, Name: "query_total"},
+		prometheus.CounterOpts{
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "query_total",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
+		},
 		[]string{"query_type", "error", "route"},
 	)
 	registry.MustRegister(s.queryCounter)
 
 	s.queryDurationSummary = prometheus.NewSummaryVec(
 		prometheus.SummaryOpts{
-			Namespace: namespace, Subsystem: subsystem,
-			Name: "query_duration_seconds",
+			Namespace: namespace, Subsystem: "db",
+			Name:        "query_duration_seconds",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
 		},
 		[]string{"query_type", "error", "route"},
 	)
 	registry.MustRegister(s.queryDurationSummary)
 
 	// txnCounter: prometheus.NewCounter(
-	// 	prometheus.CounterOpts{Namespace: namespace, Subsystem: subsystem, Name: "transaction_total"},
+	// 	prometheus.CounterOpts{Namespace: namespace, Subsystem: "db", Name: "transaction_total"},
 	// ),
 	// registry.MustRegister(s.txnCounter)
 	// txnDuration: prometheus.NewHistogram(
 	// 	prometheus.HistogramOpts{
-	// 		Namespace: namespace, Subsystem: subsystem,
+	// 		Namespace: namespace, Subsystem: "db",
 	// 		Name:    "transaction_duration_seconds",
 	//		Buckets: prometheus.ExponentialBuckets(0.1, 3, 5),
 	// 	},
 	// ),
 	// registry.MustRegister(s.txnDuration)
 
 	s.maxOpenConnectionsGauge = prometheus.NewGaugeFunc(
-		prometheus.GaugeOpts{Namespace: namespace, Subsystem: subsystem, Name: "max_open_connections"},
+		prometheus.GaugeOpts{
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "max_open_connections",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
+		},
 		func() float64 {
 			// Right now MaxOpenConnections in Horizon is static however it's possible that
 			// it will change one day. In such case, using GaugeFunc is very cheap and will
@@ -102,23 +112,38 @@ func RegisterMetrics(base *Session, namespace string, sub Subsystem, registry *p
 	registry.MustRegister(s.maxOpenConnectionsGauge)
 
 	s.openConnectionsGauge = prometheus.NewGaugeFunc(
-		prometheus.GaugeOpts{Namespace: namespace, Subsystem: subsystem, Name: "open_connections"},
+		prometheus.GaugeOpts{
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "open_connections",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
+		},
 		func() float64 {
 			return float64(base.DB.Stats().OpenConnections)
 		},
 	)
 	registry.MustRegister(s.openConnectionsGauge)
 
 	s.inUseConnectionsGauge = prometheus.NewGaugeFunc(
-		prometheus.GaugeOpts{Namespace: namespace, Subsystem: subsystem, Name: "in_use_connections"},
+		prometheus.GaugeOpts{
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "in_use_connections",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
+		},
 		func() float64 {
 			return float64(base.DB.Stats().InUse)
 		},
 	)
 	registry.MustRegister(s.inUseConnectionsGauge)
 
 	s.idleConnectionsGauge = prometheus.NewGaugeFunc(
-		prometheus.GaugeOpts{Namespace: namespace, Subsystem: subsystem, Name: "idle_connections"},
+		prometheus.GaugeOpts{
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "idle_connections",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
+		},
 		func() float64 {
 			return float64(base.DB.Stats().Idle)
 		},
@@ -127,8 +152,11 @@ func RegisterMetrics(base *Session, namespace string, sub Subsystem, registry *p
 
 	s.waitCountCounter = prometheus.NewCounterFunc(
 		prometheus.CounterOpts{
-			Namespace: namespace, Subsystem: subsystem, Name: "wait_count_total",
-			Help: "total number of number of connections waited for",
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "wait_count_total",
+			Help:        "total number of number of connections waited for",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
 		},
 		func() float64 {
 			return float64(base.DB.Stats().WaitCount)
@@ -138,8 +166,11 @@ func RegisterMetrics(base *Session, namespace string, sub Subsystem, registry *p
 
 	s.waitDurationCounter = prometheus.NewCounterFunc(
 		prometheus.CounterOpts{
-			Namespace: namespace, Subsystem: subsystem, Name: "wait_duration_seconds_total",
-			Help: "total time blocked waiting for a new connection",
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "wait_duration_seconds_total",
+			Help:        "total time blocked waiting for a new connection",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
 		},
 		func() float64 {
 			return base.DB.Stats().WaitDuration.Seconds()
@@ -149,8 +180,11 @@ func RegisterMetrics(base *Session, namespace string, sub Subsystem, registry *p
 
 	s.maxIdleClosedCounter = prometheus.NewCounterFunc(
 		prometheus.CounterOpts{
-			Namespace: namespace, Subsystem: subsystem, Name: "max_idle_closed_total",
-			Help: "total number of number of connections closed due to SetMaxIdleConns",
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "max_idle_closed_total",
+			Help:        "total number of number of connections closed due to SetMaxIdleConns",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
 		},
 		func() float64 {
 			return float64(base.DB.Stats().MaxIdleClosed)
@@ -160,8 +194,11 @@ func RegisterMetrics(base *Session, namespace string, sub Subsystem, registry *p
 
 	s.maxIdleTimeClosedCounter = prometheus.NewCounterFunc(
 		prometheus.CounterOpts{
-			Namespace: namespace, Subsystem: subsystem, Name: "max_idle_time_closed_total",
-			Help: "total number of number of connections closed due to SetConnMaxIdleTime",
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "max_idle_time_closed_total",
+			Help:        "total number of number of connections closed due to SetConnMaxIdleTime",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
 		},
 		func() float64 {
 			return float64(base.DB.Stats().MaxIdleTimeClosed)
@@ -171,8 +208,11 @@ func RegisterMetrics(base *Session, namespace string, sub Subsystem, registry *p
 
 	s.maxLifetimeClosedCounter = prometheus.NewCounterFunc(
 		prometheus.CounterOpts{
-			Namespace: namespace, Subsystem: subsystem, Name: "max_lifetime_closed_total",
-			Help: "total number of number of connections closed due to SetConnMaxLifetime",
+			Namespace:   namespace,
+			Subsystem:   "db",
+			Name:        "max_lifetime_closed_total",
+			Help:        "total number of number of connections closed due to SetConnMaxLifetime",
+			ConstLabels: prometheus.Labels{"subservice": string(sub)},
 		},
 		func() float64 {
 			return float64(base.DB.Stats().MaxLifetimeClosed)