Merge pull request #156598 from andy-kimball/backport24.3-156595

release-24.3: sql: add read/write metrics

Author: andy-kimball

docs/generated/metrics/metrics.html

@@ -1779,6 +1779,14 @@
 <tr><td>APPLICATION</td><td>sql.statements.active.internal</td><td>Number of currently active user SQL statements (internal queries)</td><td>SQL Internal Statements</td><td>GAUGE</td><td>COUNT</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>APPLICATION</td><td>sql.statements.auto_retry.count</td><td>Number of SQL statement automatic retries</td><td>SQL Statements</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
 <tr><td>APPLICATION</td><td>sql.statements.auto_retry.count.internal</td><td>Number of SQL statement automatic retries (internal queries)</td><td>SQL Internal Statements</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.bytes_read.count</td><td>Number of bytes read by SQL statements</td><td>Bytes</td><td>COUNTER</td><td>BYTES</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.bytes_read.count.internal</td><td>Number of bytes read by SQL statements (internal queries)</td><td>SQL Internal Statements</td><td>COUNTER</td><td>BYTES</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.index_bytes_written.count</td><td>Number of primary and secondary index bytes modified by SQL statements</td><td>Bytes</td><td>COUNTER</td><td>BYTES</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.index_bytes_written.count.internal</td><td>Number of primary and secondary index bytes modified by SQL statements (internal queries)</td><td>SQL Internal Statements</td><td>COUNTER</td><td>BYTES</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.index_rows_written.count</td><td>Number of primary and secondary index rows modified by SQL statements</td><td>Rows</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.index_rows_written.count.internal</td><td>Number of primary and secondary index rows modified by SQL statements (internal queries)</td><td>SQL Internal Statements</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.rows_read.count</td><td>Number of rows read by SQL statements</td><td>Rows</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
+<tr><td>APPLICATION</td><td>sql.statements.rows_read.count.internal</td><td>Number of rows read by SQL statements (internal queries)</td><td>SQL Internal Statements</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.activity.update.latency</td><td>The latency of updates made by the SQL activity updater job. Includes failed update attempts</td><td>Nanoseconds</td><td>HISTOGRAM</td><td>NANOSECONDS</td><td>AVG</td><td>NONE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.activity.updates.failed</td><td>Number of update attempts made by the SQL activity updater job that failed with errors</td><td>failed updates</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>
 <tr><td>APPLICATION</td><td>sql.stats.activity.updates.successful</td><td>Number of successful updates made by the SQL activity updater job</td><td>successful updates</td><td>COUNTER</td><td>COUNT</td><td>AVG</td><td>NON_NEGATIVE_DERIVATIVE</td></tr>

pkg/roachprod/opentelemetry/cockroachdb_metrics.go

@@ -1797,6 +1797,14 @@ var cockroachdbMetrics = map[string]string{
 	"sql_service_latency_sum":                                     "sql.service.latency.sum",
 	"sql_statements_active":                                       "sql.statements.active",
 	"sql_statements_active_internal":                              "sql.statements.active.internal",
+	"sql_statements_rows_read_count":                              "sql.statements.rows_read.count",
+	"sql_statements_rows_read_count_internal":                     "sql.statements.rows_read.count.internal",
+	"sql_statements_bytes_read_count":                             "sql.statements.bytes_read.count",
+	"sql_statements_bytes_read_count_internal":                    "sql.statements.bytes_read.count.internal",
+	"sql_statements_index_rows_written_count":                     "sql.statements.index_rows_written.count",
+	"sql_statements_index_rows_written_count_internal":            "sql.statements.index_rows_written.count.internal",
+	"sql_statements_index_bytes_written_count":                    "sql.statements.index_bytes_written.count",
+	"sql_statements_index_bytes_written_count_internal":           "sql.statements.index_bytes_written.count.internal",
 	"sql_stats_activity_update_latency":                           "sql.stats.activity.update.latency",
 	"sql_stats_activity_update_latency_bucket":                    "sql.stats.activity.update.latency.bucket",
 	"sql_stats_activity_update_latency_count":                     "sql.stats.activity.update.latency.count",

pkg/sql/BUILD.bazel

@@ -930,6 +930,7 @@ go_test(
         "//pkg/util/log/logconfig",
         "//pkg/util/log/logpb",
         "//pkg/util/log/logtestutils",
+        "//pkg/util/metamorphic",
         "//pkg/util/metric",
         "//pkg/util/mon",
         "//pkg/util/protoutil",

pkg/sql/conn_executor.go

@@ -551,6 +551,10 @@ func makeMetrics(internal bool) Metrics {
 			FullTableOrIndexScanRejectedCount: metric.NewCounter(getMetricMeta(MetaFullTableOrIndexScanRejected, internal)),
 			TxnRetryCount:                     metric.NewCounter(getMetricMeta(MetaTxnRetry, internal)),
 			StatementRetryCount:               metric.NewCounter(getMetricMeta(MetaStatementRetry, internal)),
+			StatementRowsRead:                 metric.NewCounter(getMetricMeta(MetaStatementRowsRead, internal)),
+			StatementBytesRead:                metric.NewCounter(getMetricMeta(MetaStatementBytesRead, internal)),
+			StatementIndexRowsWritten:         metric.NewCounter(getMetricMeta(MetaStatementIndexRowsWritten, internal)),
+			StatementIndexBytesWritten:        metric.NewCounter(getMetricMeta(MetaStatementIndexBytesWritten, internal)),
 		},
 		StartedStatementCounters:  makeStartedStatementCounters(internal),
 		ExecutedStatementCounters: makeExecutedStatementCounters(internal),

pkg/sql/conn_executor_exec.go

@@ -2405,12 +2405,23 @@ func (ex *connExecutor) makeExecPlan(
 
 // topLevelQueryStats returns some basic statistics about the run of the query.
 type topLevelQueryStats struct {
-	// bytesRead is the number of bytes read from disk.
-	bytesRead int64
-	// rowsRead is the number of rows read from disk.
+	// rowsRead is the number of rows read from primary and secondary indexes.
 	rowsRead int64
-	// rowsWritten is the number of rows written.
+	// bytesRead is the number of bytes read from primary and secondary indexes.
+	bytesRead int64
+	// rowsWritten is the number of rows written to the primary index. It does not
+	// include rows written to secondary indexes.
+	// NB: There is an asymmetry between rowsRead and rowsWritten - rowsRead
+	// includes rows read from secondary indexes, while rowsWritten does not
+	// include rows written to secondary indexes. This matches the behavior of
+	// EXPLAIN ANALYZE and SQL "rows affected".
 	rowsWritten int64
+	// indexRowsWritten is the number of rows written to primary and secondary
+	// indexes. It is always >= rowsWritten.
+	indexRowsWritten int64
+	// indexBytesWritten is the number of bytes written to primary and secondary
+	// indexes.
+	indexBytesWritten int64
 	// networkEgressEstimate is an estimate for the number of bytes sent to the
 	// client. It is used for estimating the number of RUs consumed by a query.
 	networkEgressEstimate int64
@@ -2424,6 +2435,8 @@ func (s *topLevelQueryStats) add(other *topLevelQueryStats) {
 	s.bytesRead += other.bytesRead
 	s.rowsRead += other.rowsRead
 	s.rowsWritten += other.rowsWritten
+	s.indexBytesWritten += other.indexBytesWritten
+	s.indexRowsWritten += other.indexRowsWritten
 	s.networkEgressEstimate += other.networkEgressEstimate
 	s.clientTime += other.clientTime
 }

pkg/sql/delete.go

@@ -232,6 +232,15 @@ func (d *deleteNode) rowsWritten() int64 {
 	return d.run.td.rowsWritten
 }
 
+func (d *deleteNode) indexRowsWritten() int64 {
+	return d.run.td.indexRowsWritten
+}
+
+func (d *deleteNode) indexBytesWritten() int64 {
+	// No bytes counted as written for a deletion.
+	return 0
+}
+
 func (d *deleteNode) enableAutoCommit() {
 	d.run.td.enableAutoCommit()
 }

pkg/sql/delete_range.go

@@ -77,6 +77,17 @@ func (d *deleteRangeNode) rowsWritten() int64 {
 	return int64(d.rowCount)
 }
 
+func (d *deleteRangeNode) indexRowsWritten() int64 {
+	// Same as rowsWritten, because deleteRangeNode only applies to primary index
+	// rows (it is not used if there's a secondary index on the table).
+	return int64(d.rowCount)
+}
+
+func (d *deleteRangeNode) indexBytesWritten() int64 {
+	// No bytes counted as written for a deletion.
+	return 0
+}
+
 // startExec implements the planNode interface.
 func (d *deleteRangeNode) startExec(params runParams) error {
 	if err := params.p.cancelChecker.Check(); err != nil {

pkg/sql/distsql_running.go

@@ -1451,6 +1451,8 @@ func (r *DistSQLReceiver) pushMeta(meta *execinfrapb.ProducerMetadata) execinfra
 		r.stats.bytesRead += meta.Metrics.BytesRead
 		r.stats.rowsRead += meta.Metrics.RowsRead
 		r.stats.rowsWritten += meta.Metrics.RowsWritten
+		r.stats.indexRowsWritten += meta.Metrics.IndexRowsWritten
+		r.stats.indexBytesWritten += meta.Metrics.IndexBytesWritten
 		if r.progressAtomic != nil && r.expectedRowsRead != 0 {
 			progress := float64(r.stats.rowsRead) / float64(r.expectedRowsRead)
 			atomic.StoreUint64(r.progressAtomic, math.Float64bits(progress))

pkg/sql/exec_util.go

@@ -1204,6 +1204,30 @@ var (
 		Measurement: "SQL Statements",
 		Unit:        metric.Unit_COUNT,
 	}
+	MetaStatementRowsRead = metric.Metadata{
+		Name:        "sql.statements.rows_read.count",
+		Help:        "Number of rows read by SQL statements",
+		Measurement: "Rows",
+		Unit:        metric.Unit_COUNT,
+	}
+	MetaStatementBytesRead = metric.Metadata{
+		Name:        "sql.statements.bytes_read.count",
+		Help:        "Number of bytes read by SQL statements",
+		Measurement: "Bytes",
+		Unit:        metric.Unit_BYTES,
+	}
+	MetaStatementIndexRowsWritten = metric.Metadata{
+		Name:        "sql.statements.index_rows_written.count",
+		Help:        "Number of primary and secondary index rows modified by SQL statements",
+		Measurement: "Rows",
+		Unit:        metric.Unit_COUNT,
+	}
+	MetaStatementIndexBytesWritten = metric.Metadata{
+		Name:        "sql.statements.index_bytes_written.count",
+		Help:        "Number of primary and secondary index bytes modified by SQL statements",
+		Measurement: "Bytes",
+		Unit:        metric.Unit_BYTES,
+	}
 )
 
 func getMetricMeta(meta metric.Metadata, internal bool) metric.Metadata {

pkg/sql/execinfrapb/data.proto

@@ -334,12 +334,23 @@ message RemoteProducerMetadata {
   }
   // Metrics are unconditionally emitted by table readers.
   message Metrics {
-    // Total number of bytes read while executing a statement.
+    // BytesRead is the number of primary and secondary index bytes read while
+    // executing a statement.
     optional int64 bytes_read = 1 [(gogoproto.nullable) = false];
-    // Total number of rows read while executing a statement.
+    // RowsRead is the number of primary and secondary index rows read while
+    // executing a statement.
     optional int64 rows_read = 2 [(gogoproto.nullable) = false];
-    // Total number of rows modified while executing a statement.
+    // RowsWritten is the number of primary index rows modified while executing
+    // a statement. It does not include secondary index rows.
     optional int64 rows_written = 3 [(gogoproto.nullable) = false];
+    // IndexRowsWritten is the number of primary and secondary index rows
+    // modified while executing a statement. It is always >= RowsWritten.
+    optional int64 index_rows_written = 5 [(gogoproto.nullable) = false];
+    // IndexBytesWritten is the number of primary and secondary index bytes
+    // modified while executing a statement.
+    optional int64 index_bytes_written = 6 [(gogoproto.nullable) = false];
+    reserved 4;
+    // NEXT ID: 7
   }
   oneof value {
     RangeInfos range_info = 1;