Merge #66640

66640: sql: introduce crdb_internal.index_usage_stats virtual table r=Azhng a=Azhng

This commit introduce crdb_internal.index_usage_stats virtual
table that is backed by new clusterindexusagestats package. This
new package implements a variant of the indexusagestats interface
and serves the data by issuing cluster RPC fanout.

Release note (sql change): introduce crdb_internal.index_usage_statistics
virtual table to surface index usage statistics.
sql.metrics.index_usage_stats.enabled cluster setting can be used to
turn on/off the subsystem. It is default to true.
sql.metrics.index_usage_stats.reset_interval can change the reset
interval of the collected statistics. It is default to 1 hour.

Addresses #64740

Followup to #66639

Co-authored-by: Azhng <[email protected]>

pkg/ccl/logictestccl/testdata/logic_test/multi_region_zone_configs

@@ -1725,5 +1725,5 @@ let $null_table_id
 SELECT table_id FROM crdb_internal.tables WHERE schema_name = 'crdb_internal' AND name = 'tables'
 
 # Validate that builtin errors if called on a table id
-query error pq: crdb_internal\.reset_multi_region_zone_configs_for_database\(\): database "\[4294967249\]" does not exist
+query error pq: crdb_internal\.reset_multi_region_zone_configs_for_database\(\): database "\[4294967248\]" does not exist
 SELECT crdb_internal.reset_multi_region_zone_configs_for_database($null_table_id)

pkg/cli/testdata/zip/partial1

@@ -30,6 +30,7 @@ debug zip --concurrency=1 --cpu-profile-duration=0s /dev/null
 [cluster] retrieving SQL data for crdb_internal.partitions... writing output: debug/crdb_internal.partitions.txt... done
 [cluster] retrieving SQL data for crdb_internal.zones... writing output: debug/crdb_internal.zones.txt... done
 [cluster] retrieving SQL data for crdb_internal.invalid_objects... writing output: debug/crdb_internal.invalid_objects.txt... done
+[cluster] retrieving SQL data for crdb_internal.index_usage_statistics... writing output: debug/crdb_internal.index_usage_statistics.txt... done
 [cluster] requesting nodes... received response... converting to JSON... writing binary output: debug/nodes.json... done
 [cluster] requesting liveness... received response... converting to JSON... writing binary output: debug/liveness.json... done
 [node 1] node status... converting to JSON... writing binary output: debug/nodes/1/status.json... done

pkg/cli/testdata/zip/partial1_excluded

@@ -30,6 +30,7 @@ debug zip /dev/null --concurrency=1 --exclude-nodes=2 --cpu-profile-duration=0
 [cluster] retrieving SQL data for crdb_internal.partitions... writing output: debug/crdb_internal.partitions.txt... done
 [cluster] retrieving SQL data for crdb_internal.zones... writing output: debug/crdb_internal.zones.txt... done
 [cluster] retrieving SQL data for crdb_internal.invalid_objects... writing output: debug/crdb_internal.invalid_objects.txt... done
+[cluster] retrieving SQL data for crdb_internal.index_usage_statistics... writing output: debug/crdb_internal.index_usage_statistics.txt... done
 [cluster] requesting nodes... received response... converting to JSON... writing binary output: debug/nodes.json... done
 [cluster] requesting liveness... received response... converting to JSON... writing binary output: debug/liveness.json... done
 [node 1] node status... converting to JSON... writing binary output: debug/nodes/1/status.json... done

pkg/cli/testdata/zip/partial2

@@ -30,6 +30,7 @@ debug zip --concurrency=1 --cpu-profile-duration=0 /dev/null
 [cluster] retrieving SQL data for crdb_internal.partitions... writing output: debug/crdb_internal.partitions.txt... done
 [cluster] retrieving SQL data for crdb_internal.zones... writing output: debug/crdb_internal.zones.txt... done
 [cluster] retrieving SQL data for crdb_internal.invalid_objects... writing output: debug/crdb_internal.invalid_objects.txt... done
+[cluster] retrieving SQL data for crdb_internal.index_usage_statistics... writing output: debug/crdb_internal.index_usage_statistics.txt... done
 [cluster] requesting nodes... received response... converting to JSON... writing binary output: debug/nodes.json... done
 [cluster] requesting liveness... received response... converting to JSON... writing binary output: debug/liveness.json... done
 [node 1] node status... converting to JSON... writing binary output: debug/nodes/1/status.json... done

pkg/cli/testdata/zip/testzip

@@ -30,6 +30,7 @@ debug zip --concurrency=1 --cpu-profile-duration=1s /dev/null
 [cluster] retrieving SQL data for crdb_internal.partitions... writing output: debug/crdb_internal.partitions.txt... done
 [cluster] retrieving SQL data for crdb_internal.zones... writing output: debug/crdb_internal.zones.txt... done
 [cluster] retrieving SQL data for crdb_internal.invalid_objects... writing output: debug/crdb_internal.invalid_objects.txt... done
+[cluster] retrieving SQL data for crdb_internal.index_usage_statistics... writing output: debug/crdb_internal.index_usage_statistics.txt... done
 [cluster] requesting nodes... received response... converting to JSON... writing binary output: debug/nodes.json... done
 [cluster] requesting liveness... received response... converting to JSON... writing binary output: debug/liveness.json... done
 [cluster] requesting CPU profiles

pkg/cli/testdata/zip/testzip_concurrent

@@ -264,6 +264,9 @@ zip
 [cluster] retrieving SQL data for crdb_internal.default_privileges...
 [cluster] retrieving SQL data for crdb_internal.default_privileges: done
 [cluster] retrieving SQL data for crdb_internal.default_privileges: writing output: debug/crdb_internal.default_privileges.txt...
+[cluster] retrieving SQL data for crdb_internal.index_usage_statistics...
+[cluster] retrieving SQL data for crdb_internal.index_usage_statistics: done
+[cluster] retrieving SQL data for crdb_internal.index_usage_statistics: writing output: debug/crdb_internal.index_usage_statistics.txt...
 [cluster] retrieving SQL data for crdb_internal.invalid_objects...
 [cluster] retrieving SQL data for crdb_internal.invalid_objects: done
 [cluster] retrieving SQL data for crdb_internal.invalid_objects: writing output: debug/crdb_internal.invalid_objects.txt...

pkg/cli/zip_cluster_wide.go

@@ -97,6 +97,7 @@ var debugZipTablesPerCluster = []string{
 	"crdb_internal.partitions",
 	"crdb_internal.zones",
 	"crdb_internal.invalid_objects",
+	"crdb_internal.index_usage_statistics",
 }
 
 // collectClusterData runs the data collection that only needs to

pkg/server/index_usage_stats_test.go

@@ -220,24 +220,24 @@ func TestStatusAPIIndexUsage(t *testing.T) {
 	}, /* expectedKeys */ 4 /* expectedEventCnt*/, 5*time.Second /* timeout */)
 
 	// First node should have nothing.
-	stats := firstLocalStatsReader.Get(indexKeyA)
+	stats := firstLocalStatsReader.Get(indexKeyA.TableID, indexKeyA.IndexID)
 	require.Equal(t, roachpb.IndexUsageStatistics{}, stats, "expecting empty stats on node 1, but found %v", stats)
 
-	stats = firstLocalStatsReader.Get(indexKeyB)
+	stats = firstLocalStatsReader.Get(indexKeyB.TableID, indexKeyB.IndexID)
 	require.Equal(t, roachpb.IndexUsageStatistics{}, stats, "expecting empty stats on node 1, but found %v", stats)
 
 	// Third node should have nothing.
-	stats = thirdLocalStatsReader.Get(indexKeyA)
+	stats = thirdLocalStatsReader.Get(indexKeyA.TableID, indexKeyA.IndexID)
 	require.Equal(t, roachpb.IndexUsageStatistics{}, stats, "expecting empty stats on node 3, but found %v", stats)
 
-	stats = thirdLocalStatsReader.Get(indexKeyB)
+	stats = thirdLocalStatsReader.Get(indexKeyB.TableID, indexKeyB.IndexID)
 	require.Equal(t, roachpb.IndexUsageStatistics{}, stats, "expecting empty stats on node 1, but found %v", stats)
 
 	// Second server should have nonempty local storage.
-	stats = secondLocalStatsReader.Get(indexKeyA)
+	stats = secondLocalStatsReader.Get(indexKeyA.TableID, indexKeyA.IndexID)
 	compareStatsHelper(t, expectedStatsIndexA, stats, time.Minute)
 
-	stats = secondLocalStatsReader.Get(indexKeyB)
+	stats = secondLocalStatsReader.Get(indexKeyB.TableID, indexKeyB.IndexID)
 	compareStatsHelper(t, expectedStatsIndexB, stats, time.Minute)
 
 	// Test cluster-wide RPC.

pkg/sql/catalog/catconstants/constants.go

@@ -60,6 +60,7 @@ const (
 	CrdbInternalGossipLivenessTableID
 	CrdbInternalGossipNetworkTableID
 	CrdbInternalIndexColumnsTableID
+	CrdbInternalIndexUsageStatisticsTableID
 	CrdbInternalInflightTraceSpanTableID
 	CrdbInternalJobsTableID
 	CrdbInternalKVNodeStatusTableID

pkg/sql/crdb_internal.go

@@ -46,6 +46,7 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/sql/catalog/descpb"
 	"github.com/cockroachdb/cockroach/pkg/sql/catalog/schemaexpr"
 	"github.com/cockroachdb/cockroach/pkg/sql/catalog/tabledesc"
+	"github.com/cockroachdb/cockroach/pkg/sql/idxusage"
 	"github.com/cockroachdb/cockroach/pkg/sql/pgwire/pgcode"
 	"github.com/cockroachdb/cockroach/pkg/sql/pgwire/pgerror"
 	"github.com/cockroachdb/cockroach/pkg/sql/privilege"
@@ -110,6 +111,7 @@ var crdbInternal = virtualSchema{
 		catconstants.CrdbInternalGossipLivenessTableID:            crdbInternalGossipLivenessTable,
 		catconstants.CrdbInternalGossipNetworkTableID:             crdbInternalGossipNetworkTable,
 		catconstants.CrdbInternalIndexColumnsTableID:              crdbInternalIndexColumnsTable,
+		catconstants.CrdbInternalIndexUsageStatisticsTableID:      crdbInternalIndexUsageStatistics,
 		catconstants.CrdbInternalInflightTraceSpanTableID:         crdbInternalInflightTraceSpanTable,
 		catconstants.CrdbInternalJobsTableID:                      crdbInternalJobsTable,
 		catconstants.CrdbInternalKVNodeStatusTableID:              crdbInternalKVNodeStatusTable,
@@ -4883,3 +4885,57 @@ CREATE TABLE crdb_internal.default_privileges (
 			})
 	},
 }
+
+var crdbInternalIndexUsageStatistics = virtualSchemaTable{
+	comment: `cluster-wide index usage statistics (in-memory, not durable).` +
+		`Querying this table is an expensive operation since it creates a` +
+		`cluster-wide RPC fanout.`,
+	schema: `
+CREATE TABLE crdb_internal.index_usage_statistics (
+  table_id        INT NOT NULL,
+  index_id        INT NOT NULL,
+  total_reads     INT NOT NULL,
+  last_read       TIMESTAMPTZ
+);`,
+	generator: func(ctx context.Context, p *planner, dbContext catalog.DatabaseDescriptor, stopper *stop.Stopper) (virtualTableGenerator, cleanupFunc, error) {
+		// Perform RPC Fanout.
+		stats, err :=
+			p.extendedEvalCtx.SQLStatusServer.IndexUsageStatistics(ctx, &serverpb.IndexUsageStatisticsRequest{})
+		if err != nil {
+			return nil, nil, err
+		}
+		indexStats := idxusage.NewLocalIndexUsageStatsFromExistingStats(&idxusage.Config{}, stats.Statistics)
+
+		row := make(tree.Datums, 4 /* number of columns for this virtual table */)
+		worker := func(pusher rowPusher) error {
+			return forEachTableDescAll(ctx, p, dbContext, hideVirtual,
+				func(db catalog.DatabaseDescriptor, _ string, table catalog.TableDescriptor) error {
+					tableID := table.GetID()
+					return catalog.ForEachIndex(table, catalog.IndexOpts{}, func(idx catalog.Index) error {
+						indexID := idx.GetID()
+						stats := indexStats.Get(roachpb.TableID(tableID), roachpb.IndexID(indexID))
+
+						lastScanTs := tree.DNull
+						if !stats.LastRead.IsZero() {
+							lastScanTs, err = tree.MakeDTimestampTZ(stats.LastRead, time.Nanosecond)
+							if err != nil {
+								return err
+							}
+						}
+
+						row = row[:0]
+
+						row = append(row,
+							tree.NewDInt(tree.DInt(tableID)),              // tableID
+							tree.NewDInt(tree.DInt(indexID)),              // indexID
+							tree.NewDInt(tree.DInt(stats.TotalReadCount)), // total_reads
+							lastScanTs, // last_scan
+						)
+
+						return pusher.pushRow(row...)
+					})
+				})
+		}
+		return setupGenerator(ctx, worker, stopper)
+	},
+}

pkg/sql/idxusage/local_idx_usage_stats.go

@@ -130,6 +130,20 @@ func NewLocalIndexUsageStats(cfg *Config) *LocalIndexUsageStats {
 	return is
 }
 
+// NewLocalIndexUsageStatsFromExistingStats returns a new instance of
+// LocalIndexUsageStats that is populated using given
+// []roachpb.CollectedIndexUsageStatistics. This constructor can be used to
+// quickly aggregate the index usage statistics received from the RPC fanout
+// and it is more efficient than the regular insert path because it performs
+// insert without taking the RWMutex lock.
+func NewLocalIndexUsageStatsFromExistingStats(
+	cfg *Config, stats []roachpb.CollectedIndexUsageStatistics,
+) *LocalIndexUsageStats {
+	s := NewLocalIndexUsageStats(cfg)
+	s.batchInsertUnsafe(stats)
+	return s
+}
+
 // Start starts the background goroutine that is responsible for collecting
 // index usage statistics.
 func (s *LocalIndexUsageStats) Start(ctx context.Context, stopper *stop.Stopper) {
@@ -159,11 +173,13 @@ func (s *LocalIndexUsageStats) record(ctx context.Context, payload indexUse) {
 }
 
 // Get returns the index usage statistics for a given key.
-func (s *LocalIndexUsageStats) Get(key roachpb.IndexUsageKey) roachpb.IndexUsageStatistics {
+func (s *LocalIndexUsageStats) Get(
+	tableID roachpb.TableID, indexID roachpb.IndexID,
+) roachpb.IndexUsageStatistics {
 	s.mu.RLock()
 	defer s.mu.RUnlock()
 
-	table, ok := s.mu.usageStats[key.TableID]
+	table, ok := s.mu.usageStats[tableID]
 	if !ok {
 		// We return a copy of the empty stats.
 		emptyStats := emptyIndexUsageStats
@@ -173,7 +189,7 @@ func (s *LocalIndexUsageStats) Get(key roachpb.IndexUsageKey) roachpb.IndexUsage
 	table.RLock()
 	defer table.RUnlock()
 
-	indexStats, ok := table.stats[key.IndexID]
+	indexStats, ok := table.stats[indexID]
 	if !ok {
 		emptyStats := emptyIndexUsageStats
 		return emptyStats
@@ -209,7 +225,7 @@ func (s *LocalIndexUsageStats) ForEach(options IteratorOptions, visitor StatsVis
 	s.mu.RUnlock()
 
 	for _, tableID := range tableIDLists {
-		tableIdxStats := s.getStatsForTableID(tableID, false /* createIfNotExists */)
+		tableIdxStats := s.getStatsForTableID(tableID, false /* createIfNotExists */, false /* unsafe */)
 
 		// This means the data s being cleared before we can fetch it. It's not an
 		// error, so we simply just skip over it.
@@ -231,6 +247,20 @@ func (s *LocalIndexUsageStats) ForEach(options IteratorOptions, visitor StatsVis
 	return nil
 }
 
+// batchInsertUnsafe inserts otherStats into s without taking on write lock.
+// This should only be called during initialization when we can be sure there's
+// no other users of s. This avoids the locking overhead when it's not
+// necessary.
+func (s *LocalIndexUsageStats) batchInsertUnsafe(
+	otherStats []roachpb.CollectedIndexUsageStatistics,
+) {
+	for _, newStats := range otherStats {
+		tableIndexStats := s.getStatsForTableID(newStats.Key.TableID, true /* createIfNotExists */, true /* unsafe */)
+		stats := tableIndexStats.getStatsForIndexID(newStats.Key.IndexID, true /* createIfNotExists */, true /* unsafe */)
+		stats.Add(&newStats.Stats)
+	}
+}
+
 func (s *LocalIndexUsageStats) clear() {
 	s.mu.Lock()
 	defer s.mu.Unlock()
@@ -241,8 +271,8 @@ func (s *LocalIndexUsageStats) clear() {
 }
 
 func (s *LocalIndexUsageStats) insertIndexUsage(idxUse *indexUse) {
-	tableStats := s.getStatsForTableID(idxUse.key.TableID, true /* createIfNotExists */)
-	indexStats := tableStats.getStatsForIndexID(idxUse.key.IndexID, true /* createIfNotExists */)
+	tableStats := s.getStatsForTableID(idxUse.key.TableID, true /* createIfNotExists */, false /* unsafe */)
+	indexStats := tableStats.getStatsForIndexID(idxUse.key.IndexID, true /* createIfNotExists */, false /* unsafe */)
 	indexStats.Lock()
 	defer indexStats.Unlock()
 	switch idxUse.usageTyp {
@@ -259,15 +289,21 @@ func (s *LocalIndexUsageStats) insertIndexUsage(idxUse *indexUse) {
 	}
 }
 
+// getStatsForTableID returns the tableIndexStats for the given roachpb.TableID.
+// If unsafe is set to true, then the lookup is performed without locking to the
+// internal RWMutex lock. This can be used when LocalIndexUsageStats is not
+// being concurrently accessed.
 func (s *LocalIndexUsageStats) getStatsForTableID(
-	id roachpb.TableID, createIfNotExists bool,
+	id roachpb.TableID, createIfNotExists bool, unsafe bool,
 ) *tableIndexStats {
-	if createIfNotExists {
-		s.mu.Lock()
-		defer s.mu.Unlock()
-	} else {
-		s.mu.RLock()
-		defer s.mu.RUnlock()
+	if !unsafe {
+		if createIfNotExists {
+			s.mu.Lock()
+			defer s.mu.Unlock()
+		} else {
+			s.mu.RLock()
+			defer s.mu.RUnlock()
+		}
 	}
 
 	if tableIndexStats, ok := s.mu.usageStats[id]; ok {
@@ -286,15 +322,22 @@ func (s *LocalIndexUsageStats) getStatsForTableID(
 	return nil
 }
 
+// getStatsForIndexID returns the indexStats for the given roachpb.IndexID.
+// If unsafe is set to true, then the lookup is performed without locking to the
+// internal RWMutex lock. This can be used when tableIndexStats is not being
+// concurrently accessed.
 func (t *tableIndexStats) getStatsForIndexID(
-	id roachpb.IndexID, createIfNotExists bool,
+	id roachpb.IndexID, createIfNotExists bool, unsafe bool,
 ) *indexStats {
-	if createIfNotExists {
-		t.Lock()
-		defer t.Unlock()
-	} else {
-		t.RLock()
-		defer t.RUnlock()
+	if !unsafe {
+		if createIfNotExists {
+			t.Lock()
+			defer t.Unlock()
+		} else {
+			t.RLock()
+			defer t.RUnlock()
+		}
+
 	}
 
 	if stats, ok := t.stats[id]; ok {
@@ -329,7 +372,7 @@ func (t *tableIndexStats) iterateIndexStats(
 	}
 
 	for _, indexID := range indexIDs {
-		indexStats := t.getStatsForIndexID(indexID, false /* createIfNotExists */)
+		indexStats := t.getStatsForIndexID(indexID, false /* createIfNotExists */, false /* unsafe */)
 
 		// This means the data is being cleared  before we can fetch it. It's not an
 		// error, so we simply just skip over it.

pkg/sql/idxusage/local_index_usage_stats_test.go

@@ -149,7 +149,7 @@ func TestIndexUsageStatisticsSubsystem(t *testing.T) {
 	t.Run("point lookup", func(t *testing.T) {
 		actualEntryCount := 0
 		for _, index := range indices {
-			stats := localIndexUsage.Get(index)
+			stats := localIndexUsage.Get(index.TableID, index.IndexID)
 			require.NotNil(t, stats)
 
 			actualEntryCount++

pkg/sql/logictest/testdata/logic_test/crdb_internal

@@ -39,6 +39,7 @@ crdb_internal  gossip_liveness              table  NULL  NULL  NULL
 crdb_internal  gossip_network               table  NULL  NULL  NULL
 crdb_internal  gossip_nodes                 table  NULL  NULL  NULL
 crdb_internal  index_columns                table  NULL  NULL  NULL
+crdb_internal  index_usage_statistics       table  NULL  NULL  NULL
 crdb_internal  interleaved                  table  NULL  NULL  NULL
 crdb_internal  invalid_objects              table  NULL  NULL  NULL
 crdb_internal  jobs                         table  NULL  NULL  NULL

pkg/sql/logictest/testdata/logic_test/crdb_internal_tenant

@@ -52,6 +52,7 @@ crdb_internal  gossip_liveness              table  NULL  NULL  NULL
 crdb_internal  gossip_network               table  NULL  NULL  NULL
 crdb_internal  gossip_nodes                 table  NULL  NULL  NULL
 crdb_internal  index_columns                table  NULL  NULL  NULL
+crdb_internal  index_usage_statistics       table  NULL  NULL  NULL
 crdb_internal  interleaved                  table  NULL  NULL  NULL
 crdb_internal  invalid_objects              table  NULL  NULL  NULL
 crdb_internal  jobs                         table  NULL  NULL  NULL

pkg/sql/logictest/testdata/logic_test/create_statements

@@ -427,6 +427,17 @@ CREATE TABLE crdb_internal.index_columns (
    column_direction STRING NULL,
    implicit BOOL NULL
 )  {}  {}
+CREATE TABLE crdb_internal.index_usage_statistics (
+   table_id INT8 NOT NULL,
+   index_id INT8 NOT NULL,
+   total_reads INT8 NOT NULL,
+   last_read TIMESTAMPTZ NULL
+)  CREATE TABLE crdb_internal.index_usage_statistics (
+   table_id INT8 NOT NULL,
+   index_id INT8 NOT NULL,
+   total_reads INT8 NOT NULL,
+   last_read TIMESTAMPTZ NULL
+)  {}  {}
 CREATE TABLE crdb_internal.interleaved (
    database_name STRING NOT NULL,
    schema_name STRING NOT NULL,

pkg/sql/logictest/testdata/logic_test/grant_table

@@ -51,6 +51,7 @@ test           crdb_internal       gossip_liveness                        public
 test           crdb_internal       gossip_network                         public   SELECT
 test           crdb_internal       gossip_nodes                           public   SELECT
 test           crdb_internal       index_columns                          public   SELECT
+test           crdb_internal       index_usage_statistics                 public   SELECT
 test           crdb_internal       interleaved                            public   SELECT
 test           crdb_internal       invalid_objects                        public   SELECT
 test           crdb_internal       jobs                                   public   SELECT