pkg/sql/sqlstats: counter metric for flushed fingerprint counts

Addresses: cockroachdb#119779

The count of unique fingeprints flushed to `system.statement_statistics`
and `system.transaction_statistics` is the core component that
determines data cardinality within the SQL stats subsystem. Today, we
don't have good metrics around this source of cardinality. As we aim to
reduce cardinality by improving our fingerprinting algorithms, creating
a metric to count the number of unique statement and transaction
fingerprints included in each flush of the in-memory SQL stats will be a
helpful measurement to benchmark cardinality reduction.

This patch adds a new metric to track the # of unique fingerprints (stmt
and txn) included in each flush.

Release note (ops change): A new counter metric,
`sql.stats.flush.fingerprint.count`, has been introduced. The metric
tracks the number of unique statement and transaction fingerprints
included in the SQL Stats flush.

docs/generated/metrics/metrics.html

@@ -1452,6 +1452,7 @@
 <tr><td>APPLICATION</td><td>sql.stats.flush.done_signals.ignored</td><td>Number of times the SQL Stats activity update job ignored the signal sent to it indicating a flush has completed</td><td>flush done signals ignored</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.flush.duration</td><td>Time took to in nanoseconds to complete SQL Stats flush</td><td>SQL Stats Flush</td><td>HISTOGRAM</td><td>NANOSECONDS</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.flush.error</td><td>Number of errors encountered when flushing SQL Stats</td><td>SQL Stats Flush</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.stats.flush.fingerprint.count</td><td>The number of unique statement and transaction fingerprints included in the SQL Stats flush</td><td>statement &amp; transaction fingerprints</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.mem.current</td><td>Current memory usage for fingerprint storage</td><td>Memory</td><td>GAUGE</td><td>BYTES</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.mem.max</td><td>Memory usage for fingerprint storage</td><td>Memory</td><td>HISTOGRAM</td><td>BYTES</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.reported.mem.current</td><td>Current memory usage for reported fingerprint storage</td><td>Memory</td><td>GAUGE</td><td>BYTES</td><td>AVG</td><td>NONE</td></tr>

pkg/sql/conn_executor.go

@@ -487,6 +487,7 @@ func NewServer(cfg *ExecutorConfig, pool *mon.BytesMonitor) *Server {
 		Knobs:                   cfg.SQLStatsTestingKnobs,
 		FlushCounter:            serverMetrics.StatsMetrics.SQLStatsFlushStarted,
 		FlushDoneSignalsIgnored: serverMetrics.StatsMetrics.SQLStatsFlushDoneSignalsIgnored,
+		FlushedFingerprintCount: serverMetrics.StatsMetrics.SQLStatsFlushFingerprintCount,
 		FailureCounter:          serverMetrics.StatsMetrics.SQLStatsFlushFailure,
 		FlushDuration:           serverMetrics.StatsMetrics.SQLStatsFlushDuration,
 	}, memSQLStats)
@@ -587,6 +588,7 @@ func makeServerMetrics(cfg *ExecutorConfig) ServerMetrics {
 			DiscardedStatsCount:                 metric.NewCounter(MetaDiscardedSQLStats),
 			SQLStatsFlushStarted:                metric.NewCounter(MetaSQLStatsFlushStarted),
 			SQLStatsFlushDoneSignalsIgnored:     metric.NewCounter(MetaSQLStatsFlushDoneSignalsIgnored),
+			SQLStatsFlushFingerprintCount:       metric.NewCounter(MetaSQLStatsFlushFingerprintCount),
 
 			SQLStatsFlushFailure: metric.NewCounter(MetaSQLStatsFlushFailure),
 			SQLStatsFlushDuration: metric.NewHistogram(metric.HistogramOptions{

pkg/sql/exec_util.go

@@ -1135,6 +1135,12 @@ var (
 		Measurement: "SQL Stats Flush",
 		Unit:        metric.Unit_COUNT,
 	}
+	MetaSQLStatsFlushFingerprintCount = metric.Metadata{
+		Name:        "sql.stats.flush.fingerprint.count",
+		Help:        "The number of unique statement and transaction fingerprints included in the SQL Stats flush",
+		Measurement: "statement & transaction fingerprints",
+		Unit:        metric.Unit_COUNT,
+	}
 	MetaSQLStatsFlushDoneSignalsIgnored = metric.Metadata{
 		Name: "sql.stats.flush.done_signals.ignored",
 		Help: "Number of times the SQL Stats activity update job ignored the signal sent to it indicating " +

pkg/sql/executor_statement_metrics.go

@@ -77,6 +77,7 @@ type StatsMetrics struct {
 
 	SQLStatsFlushStarted            *metric.Counter
 	SQLStatsFlushDoneSignalsIgnored *metric.Counter
+	SQLStatsFlushFingerprintCount   *metric.Counter
 	SQLStatsFlushFailure            *metric.Counter
 	SQLStatsFlushDuration           metric.IHistogram
 	SQLStatsRemovedRows             *metric.Counter

pkg/sql/sqlstats/persistedsqlstats/flush.go

@@ -68,8 +68,10 @@ func (s *PersistedSQLStats) Flush(ctx context.Context) {
 		return
 	}
 
+	fingerprintCount := s.SQLStats.GetTotalFingerprintCount()
+	s.cfg.FlushedFingerprintCount.Inc(fingerprintCount)
 	log.Infof(ctx, "flushing %d stmt/txn fingerprints (%d bytes) after %s",
-		s.SQLStats.GetTotalFingerprintCount(), s.SQLStats.GetTotalFingerprintBytes(), timeutil.Since(s.lastFlushStarted))
+		fingerprintCount, s.SQLStats.GetTotalFingerprintBytes(), timeutil.Since(s.lastFlushStarted))
 	s.lastFlushStarted = now
 
 	aggregatedTs := s.ComputeAggregatedTs()

pkg/sql/sqlstats/persistedsqlstats/provider.go

@@ -50,6 +50,7 @@ type Config struct {
 	FlushDuration           metric.IHistogram
 	FlushDoneSignalsIgnored *metric.Counter
 	FailureCounter          *metric.Counter
+	FlushedFingerprintCount *metric.Counter
 
 	// Testing knobs.
 	Knobs *sqlstats.TestingKnobs