tsdb: add tenant-level store metrics to tsdb

pkg/kv/kvbase/BUILD.bazel

@@ -3,7 +3,10 @@ load("@io_bazel_rules_go//go:def.bzl", "go_library")
 
 go_library(
     name = "kvbase",
-    srcs = ["constants.go"],
+    srcs = [
+        "constants.go",
+        "metrics.go",
+    ],
     importpath = "github.com/cockroachdb/cockroach/pkg/kv/kvbase",
     visibility = ["//visibility:public"],
 )

pkg/kv/kvbase/metrics.go

@@ -0,0 +1,17 @@
+// Copyright 2023 The Cockroach Authors.
+//
+// Use of this software is governed by the Business Source License
+// included in the file licenses/BSL.txt.
+//
+// As of the Change Date specified in that file, in accordance with
+// the Business Source License, use of this software will be governed
+// by the Apache License, Version 2.0, included in the file
+// licenses/APL.txt.
+
+package kvbase
+
+// TenantsStorageMetricsSet is the set of all metric names contained
+// within TenantsStorageMetrics, recorded at the individual tenant level.
+//
+// Made available in kvbase to help avoid import cycles.
+var TenantsStorageMetricsSet map[string]struct{}

pkg/kv/kvserver/metrics.go

@@ -18,6 +18,7 @@ import (
 	"time"
 	"unsafe"
 
+	"github.com/cockroachdb/cockroach/pkg/kv/kvbase"
 	"github.com/cockroachdb/cockroach/pkg/kv/kvserver/allocator/allocatorimpl"
 	"github.com/cockroachdb/cockroach/pkg/kv/kvserver/batcheval/result"
 	"github.com/cockroachdb/cockroach/pkg/kv/kvserver/kvserverpb"
@@ -37,6 +38,12 @@ import (
 	"go.etcd.io/raft/v3/raftpb"
 )
 
+func init() {
+	// Inject the TenantStorageMetricsSet available in the kvbase pkg to
+	// avoid import cycles.
+	kvbase.TenantsStorageMetricsSet = tenantsStorageMetricsSet()
+}
+
 var (
 	// Replica metrics.
 	metaReplicaCount = metric.Metadata{
@@ -2204,6 +2211,8 @@ func (ref *tenantMetricsRef) assert(ctx context.Context) {
 // call acquire and release to properly reference count the metrics for
 // individual tenants.
 type TenantsStorageMetrics struct {
+	// NB: If adding more metrics to this struct, be sure to
+	// also update tenantsStorageMetricsSet().
 	LiveBytes      *aggmetric.AggGauge
 	KeyBytes       *aggmetric.AggGauge
 	ValBytes       *aggmetric.AggGauge
@@ -2232,6 +2241,33 @@ type TenantsStorageMetrics struct {
 	tenants syncutil.IntMap // map[int64(roachpb.TenantID)]*tenantStorageMetrics
 }
 
+// tenantsStorageMetricsSet returns the set of all metric names contained
+// within TenantsStorageMetrics.
+//
+// see kvbase.TenantsStorageMetricsSet for public access. Assigned in init().
+func tenantsStorageMetricsSet() map[string]struct{} {
+	return map[string]struct{}{
+		metaLiveBytes.Name:      {},
+		metaKeyBytes.Name:       {},
+		metaValBytes.Name:       {},
+		metaRangeKeyBytes.Name:  {},
+		metaRangeValBytes.Name:  {},
+		metaTotalBytes.Name:     {},
+		metaIntentBytes.Name:    {},
+		metaLiveCount.Name:      {},
+		metaKeyCount.Name:       {},
+		metaValCount.Name:       {},
+		metaRangeKeyCount.Name:  {},
+		metaRangeValCount.Name:  {},
+		metaIntentCount.Name:    {},
+		metaIntentAge.Name:      {},
+		metaGcBytesAge.Name:     {},
+		metaSysBytes.Name:       {},
+		metaSysCount.Name:       {},
+		metaAbortSpanBytes.Name: {},
+	}
+}
+
 var _ metric.Struct = (*TenantsStorageMetrics)(nil)
 
 // MetricStruct makes TenantsStorageMetrics a metric.Struct.

pkg/server/status/BUILD.bazel

@@ -44,8 +44,10 @@ go_library(
         "//pkg/build",
         "//pkg/keys",
         "//pkg/kv",
+        "//pkg/kv/kvbase",
         "//pkg/kv/kvpb",
         "//pkg/kv/kvserver/liveness",
+        "//pkg/multitenant",
         "//pkg/roachpb",
         "//pkg/rpc",
         "//pkg/server/status/statuspb",
@@ -69,6 +71,7 @@ go_library(
         "@com_github_cockroachdb_redact//:redact",
         "@com_github_dustin_go_humanize//:go-humanize",
         "@com_github_elastic_gosigar//:gosigar",
+        "@com_github_prometheus_client_model//go",
         "@com_github_shirou_gopsutil_v3//cpu",
         "@com_github_shirou_gopsutil_v3//net",
     ] + select({
@@ -134,6 +137,7 @@ go_test(
     deps = [
         "//pkg/base",
         "//pkg/build",
+        "//pkg/multitenant",
         "//pkg/roachpb",
         "//pkg/security/securityassets",
         "//pkg/security/securitytest",
@@ -143,6 +147,7 @@ go_test(
         "//pkg/sql/sem/catconstants",
         "//pkg/testutils/serverutils",
         "//pkg/ts/tspb",
+        "//pkg/ts/tsutil",
         "//pkg/util/leaktest",
         "//pkg/util/log",
         "//pkg/util/log/eventpb",
@@ -151,6 +156,7 @@ go_test(
         "//pkg/util/system",
         "//pkg/util/timeutil",
         "@com_github_kr_pretty//:pretty",
+        "@com_github_prometheus_client_model//go",
         "@com_github_shirou_gopsutil_v3//net",
         "@com_github_stretchr_testify//require",
     ],

pkg/server/status/recorder.go

@@ -27,8 +27,10 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/build"
 	"github.com/cockroachdb/cockroach/pkg/keys"
 	"github.com/cockroachdb/cockroach/pkg/kv"
+	"github.com/cockroachdb/cockroach/pkg/kv/kvbase"
 	"github.com/cockroachdb/cockroach/pkg/kv/kvpb"
 	"github.com/cockroachdb/cockroach/pkg/kv/kvserver/liveness"
+	"github.com/cockroachdb/cockroach/pkg/multitenant"
 	"github.com/cockroachdb/cockroach/pkg/roachpb"
 	"github.com/cockroachdb/cockroach/pkg/rpc"
 	"github.com/cockroachdb/cockroach/pkg/server/status/statuspb"
@@ -49,6 +51,7 @@ import (
 	"github.com/cockroachdb/redact"
 	"github.com/dustin/go-humanize"
 	"github.com/elastic/gosigar"
+	prometheusgo "github.com/prometheus/client_model/go"
 )
 
 const (
@@ -385,6 +388,7 @@ func (mr *MetricsRecorder) GetTimeSeriesData() []tspb.TimeSeriesData {
 	}
 
 	// Record time series from store-level registries.
+	tIDLabel := multitenant.TenantIDLabel
 	for storeID, r := range mr.mu.storeRegistries {
 		storeRecorder := registryRecorder{
 			registry:       r,
@@ -393,6 +397,21 @@ func (mr *MetricsRecorder) GetTimeSeriesData() []tspb.TimeSeriesData {
 			timestampNanos: now.UnixNano(),
 		}
 		storeRecorder.record(&data)
+
+		// Now record secondary tenant store metrics, if any exist in the process.
+		for tenantID := range mr.mu.tenantRegistries {
+			tenantStoreRecorder := registryRecorder{
+				registry:       r,
+				format:         storeTimeSeriesPrefix,
+				source:         tsutil.MakeTenantSource(storeID.String(), tenantID.String()),
+				timestampNanos: now.UnixNano(),
+			}
+			tenantID := tenantID.String()
+			tenantStoreRecorder.recordChild(&data, kvbase.TenantsStorageMetricsSet, &prometheusgo.LabelPair{
+				Name:  &tIDLabel,
+				Value: &tenantID,
+			})
+		}
 	}
 	atomic.CompareAndSwapInt64(&mr.lastDataCount, lastDataCount, int64(len(data)))
 	return data
@@ -659,6 +678,66 @@ func (rr registryRecorder) record(dest *[]tspb.TimeSeriesData) {
 	})
 }
 
+// recordChild filters the metrics in the registry down to those provided in
+// the metricsFilter argument, and iterates through any child metrics that
+// may exist on said metric. Child metrics whose label sets contains a match
+// against the childLabelFilter are recorded into the provided dest slice of
+// type tspb.TimeSeriesData.
+//
+// NB: Only available for Counter and Gauge metrics.
+func (rr registryRecorder) recordChild(
+	dest *[]tspb.TimeSeriesData,
+	metricsFilter map[string]struct{},
+	childLabelFilter *prometheusgo.LabelPair,
+) {
+	labels := rr.registry.GetLabels()
+	rr.registry.Select(metricsFilter, func(name string, v interface{}) {
+		prom, ok := v.(metric.PrometheusExportable)
+		if !ok {
+			return
+		}
+		promIter, ok := v.(metric.PrometheusIterable)
+		if !ok {
+			return
+		}
+		m := prom.ToPrometheusMetric()
+		m.Label = append(labels, prom.GetLabels()...)
+
+		processChildMetric := func(metric *prometheusgo.Metric) {
+			found := false
+			for _, label := range metric.Label {
+				if label.GetName() == childLabelFilter.GetName() &&
+					label.GetValue() == childLabelFilter.GetValue() {
+					found = true
+					break
+				}
+			}
+			if !found {
+				return
+			}
+			var value float64
+			if metric.Gauge != nil {
+				value = *metric.Gauge.Value
+			} else if metric.Counter != nil {
+				value = *metric.Counter.Value
+			} else {
+				return
+			}
+			*dest = append(*dest, tspb.TimeSeriesData{
+				Name:   fmt.Sprintf(rr.format, prom.GetName()),
+				Source: rr.source,
+				Datapoints: []tspb.TimeSeriesDatapoint{
+					{
+						TimestampNanos: rr.timestampNanos,
+						Value:          value,
+					},
+				},
+			})
+		}
+		promIter.Each(m.Label, processChildMetric)
+	})
+}
+
 // GetTotalMemory returns either the total system memory (in bytes) or if
 // possible the cgroups available memory.
 func GetTotalMemory(ctx context.Context) (int64, error) {