sql: add session setting for scan costing methodology

In this PR we add a session setting to gate usage of an upcoming feature
in which the optimizer will use `AvgSize`, the average column size, from
table stats to cost scans and index joins. When enabled, the optimizer
will revert to the old default method of costing scans, where each
column is treated as the same size. By default, this setting will be
off.

Informs: cockroachdb#72332

Release note: None

docs/generated/settings/settings-for-tenants.txt

@@ -85,6 +85,7 @@ sql.cross_db_fks.enabled	boolean	false	if true, creating foreign key references
 sql.cross_db_sequence_owners.enabled	boolean	false	if true, creating sequences owned by tables from other databases is allowed
 sql.cross_db_sequence_references.enabled	boolean	false	if true, sequences referenced by tables from other databases are allowed
 sql.cross_db_views.enabled	boolean	false	if true, creating views that refer to other databases is allowed
+sql.defaults.cost_scans_with_default_col_size.enabled	boolean	false	setting to true uses the same size for all columns to compute scan cost
 sql.defaults.datestyle	enumeration	iso, mdy	default value for DateStyle session setting [iso, mdy = 0, iso, dmy = 1, iso, ymd = 2]
 sql.defaults.datestyle.enabled	boolean	false	default value for datestyle_enabled session setting
 sql.defaults.default_hash_sharded_index_bucket_count	integer	8	used as bucket count if bucket count is not specified in hash sharded index definition

docs/generated/settings/settings.html

@@ -92,6 +92,7 @@
 <tr><td><code>sql.cross_db_sequence_owners.enabled</code></td><td>boolean</td><td><code>false</code></td><td>if true, creating sequences owned by tables from other databases is allowed</td></tr>
 <tr><td><code>sql.cross_db_sequence_references.enabled</code></td><td>boolean</td><td><code>false</code></td><td>if true, sequences referenced by tables from other databases are allowed</td></tr>
 <tr><td><code>sql.cross_db_views.enabled</code></td><td>boolean</td><td><code>false</code></td><td>if true, creating views that refer to other databases is allowed</td></tr>
+<tr><td><code>sql.defaults.cost_scans_with_default_col_size.enabled</code></td><td>boolean</td><td><code>false</code></td><td>setting to true uses the same size for all columns to compute scan cost</td></tr>
 <tr><td><code>sql.defaults.datestyle</code></td><td>enumeration</td><td><code>iso, mdy</code></td><td>default value for DateStyle session setting [iso, mdy = 0, iso, dmy = 1, iso, ymd = 2]</td></tr>
 <tr><td><code>sql.defaults.datestyle.enabled</code></td><td>boolean</td><td><code>false</code></td><td>default value for datestyle_enabled session setting</td></tr>
 <tr><td><code>sql.defaults.default_hash_sharded_index_bucket_count</code></td><td>integer</td><td><code>8</code></td><td>used as bucket count if bucket count is not specified in hash sharded index definition</td></tr>

pkg/sql/exec_util.go

@@ -707,6 +707,13 @@ var largeFullScanRows = settings.RegisterFloatSetting(
 	1000.0,
 ).WithPublic()
 
+var costScansWithDefaultColSize = settings.RegisterBoolSetting(
+	settings.TenantWritable,
+	`sql.defaults.cost_scans_with_default_col_size.enabled`,
+	"setting to true uses the same size for all columns to compute scan cost",
+	false,
+).WithPublic()
+
 var errNoTransactionInProgress = errors.New("there is no transaction in progress")
 var errTransactionInProgress = errors.New("there is already a transaction in progress")
 
@@ -3076,6 +3083,12 @@ func (m *sessionDataMutator) SetParallelizeMultiKeyLookupJoinsEnabled(val bool)
 	m.data.ParallelizeMultiKeyLookupJoinsEnabled = val
 }
 
+// TODO(harding): Remove this when costing scans based on average column size
+// is fully supported.
+func (m *sessionDataMutator) SetCostScansWithDefaultColSize(val bool) {
+	m.data.CostScansWithDefaultColSize = val
+}
+
 // Utility functions related to scrubbing sensitive information on SQL Stats.
 
 // quantizeCounts ensures that the Count field in the

pkg/sql/explain_bundle.go

@@ -536,6 +536,7 @@ func (c *stmtEnvCollector) PrintSessionSettings(w io.Writer) error {
 		{sessionSetting: "datestyle_enabled", clusterSetting: dateStyleEnabled, convFunc: boolToOnOff},
 		{sessionSetting: "disallow_full_table_scans", clusterSetting: disallowFullTableScans, convFunc: boolToOnOff},
 		{sessionSetting: "large_full_scan_rows", clusterSetting: largeFullScanRows},
+		{sessionSetting: "cost_scans_with_default_col_size", clusterSetting: costScansWithDefaultColSize, convFunc: boolToOnOff},
 		{sessionSetting: "distsql", clusterSetting: DistSQLClusterExecMode, convFunc: distsqlConv},
 		{sessionSetting: "vectorize", clusterSetting: VectorizeClusterMode, convFunc: vectorizeConv},
 	}

pkg/sql/logictest/testdata/logic_test/information_schema

@@ -4645,6 +4645,7 @@ bytea_output                                          hex
 check_function_bodies                                 on
 client_encoding                                       UTF8
 client_min_messages                                   notice
+cost_scans_with_default_col_size                      off
 database                                              test
 datestyle                                             ISO, MDY
 datestyle_enabled                                     off

pkg/sql/logictest/testdata/logic_test/pg_catalog

@@ -4045,6 +4045,7 @@ bytea_output                                          hex                 NULL
 check_function_bodies                                 on                  NULL      NULL        NULL        string
 client_encoding                                       UTF8                NULL      NULL        NULL        string
 client_min_messages                                   notice              NULL      NULL        NULL        string
+cost_scans_with_default_col_size                      off                 NULL      NULL        NULL        string
 database                                              test                NULL      NULL        NULL        string
 datestyle                                             ISO, MDY            NULL      NULL        NULL        string
 datestyle_enabled                                     off                 NULL      NULL        NULL        string
@@ -4158,6 +4159,7 @@ bytea_output                                          hex                 NULL
 check_function_bodies                                 on                  NULL  user     NULL      on                  on
 client_encoding                                       UTF8                NULL  user     NULL      UTF8                UTF8
 client_min_messages                                   notice              NULL  user     NULL      notice              notice
+cost_scans_with_default_col_size                      off                 NULL  user     NULL      off                 off
 database                                              test                NULL  user     NULL      ·                   test
 datestyle                                             ISO, MDY            NULL  user     NULL      ISO, MDY            ISO, MDY
 datestyle_enabled                                     off                 NULL  user     NULL      off                 off
@@ -4265,6 +4267,7 @@ bytea_output                                          NULL    NULL     NULL
 check_function_bodies                                 NULL    NULL     NULL     NULL        NULL
 client_encoding                                       NULL    NULL     NULL     NULL        NULL
 client_min_messages                                   NULL    NULL     NULL     NULL        NULL
+cost_scans_with_default_col_size                      NULL    NULL     NULL     NULL        NULL
 crdb_version                                          NULL    NULL     NULL     NULL        NULL
 database                                              NULL    NULL     NULL     NULL        NULL
 datestyle                                             NULL    NULL     NULL     NULL        NULL

pkg/sql/logictest/testdata/logic_test/show_source

@@ -33,6 +33,7 @@ bytea_output                                          hex
 check_function_bodies                                 on
 client_encoding                                       UTF8
 client_min_messages                                   notice
+cost_scans_with_default_col_size                      off
 database                                              test
 datestyle                                             ISO, MDY
 datestyle_enabled                                     off

pkg/sql/opt/memo/memo.go

@@ -135,22 +135,23 @@ type Memo struct {
 	// planning. We need to cross-check these before reusing a cached memo.
 	// NOTE: If you add new fields here, be sure to add them to the relevant
 	//       fields in explain_bundle.go.
-	reorderJoinsLimit       int
-	zigzagJoinEnabled       bool
-	useHistograms           bool
-	useMultiColStats        bool
-	localityOptimizedSearch bool
-	safeUpdates             bool
-	preferLookupJoinsForFKs bool
-	saveTablesPrefix        string
-	dateStyleEnabled        bool
-	intervalStyleEnabled    bool
-	dateStyle               pgdate.DateStyle
-	intervalStyle           duration.IntervalStyle
-	propagateInputOrdering  bool
-	disallowFullTableScans  bool
-	largeFullScanRows       float64
-	nullOrderedLast         bool
+	reorderJoinsLimit           int
+	zigzagJoinEnabled           bool
+	useHistograms               bool
+	useMultiColStats            bool
+	localityOptimizedSearch     bool
+	safeUpdates                 bool
+	preferLookupJoinsForFKs     bool
+	saveTablesPrefix            string
+	dateStyleEnabled            bool
+	intervalStyleEnabled        bool
+	dateStyle                   pgdate.DateStyle
+	intervalStyle               duration.IntervalStyle
+	propagateInputOrdering      bool
+	disallowFullTableScans      bool
+	largeFullScanRows           float64
+	nullOrderedLast             bool
+	costScansWithDefaultColSize bool
 
 	// curRank is the highest currently in-use scalar expression rank.
 	curRank opt.ScalarRank
@@ -180,23 +181,24 @@ func (m *Memo) Init(evalCtx *tree.EvalContext) {
 	// This initialization pattern ensures that fields are not unwittingly
 	// reused. Field reuse must be explicit.
 	*m = Memo{
-		metadata:                m.metadata,
-		reorderJoinsLimit:       int(evalCtx.SessionData().ReorderJoinsLimit),
-		zigzagJoinEnabled:       evalCtx.SessionData().ZigzagJoinEnabled,
-		useHistograms:           evalCtx.SessionData().OptimizerUseHistograms,
-		useMultiColStats:        evalCtx.SessionData().OptimizerUseMultiColStats,
-		localityOptimizedSearch: evalCtx.SessionData().LocalityOptimizedSearch,
-		safeUpdates:             evalCtx.SessionData().SafeUpdates,
-		preferLookupJoinsForFKs: evalCtx.SessionData().PreferLookupJoinsForFKs,
-		saveTablesPrefix:        evalCtx.SessionData().SaveTablesPrefix,
-		intervalStyleEnabled:    evalCtx.SessionData().IntervalStyleEnabled,
-		dateStyleEnabled:        evalCtx.SessionData().DateStyleEnabled,
-		dateStyle:               evalCtx.SessionData().GetDateStyle(),
-		intervalStyle:           evalCtx.SessionData().GetIntervalStyle(),
-		propagateInputOrdering:  evalCtx.SessionData().PropagateInputOrdering,
-		disallowFullTableScans:  evalCtx.SessionData().DisallowFullTableScans,
-		largeFullScanRows:       evalCtx.SessionData().LargeFullScanRows,
-		nullOrderedLast:         evalCtx.SessionData().NullOrderedLast,
+		metadata:                    m.metadata,
+		reorderJoinsLimit:           int(evalCtx.SessionData().ReorderJoinsLimit),
+		zigzagJoinEnabled:           evalCtx.SessionData().ZigzagJoinEnabled,
+		useHistograms:               evalCtx.SessionData().OptimizerUseHistograms,
+		useMultiColStats:            evalCtx.SessionData().OptimizerUseMultiColStats,
+		localityOptimizedSearch:     evalCtx.SessionData().LocalityOptimizedSearch,
+		safeUpdates:                 evalCtx.SessionData().SafeUpdates,
+		preferLookupJoinsForFKs:     evalCtx.SessionData().PreferLookupJoinsForFKs,
+		saveTablesPrefix:            evalCtx.SessionData().SaveTablesPrefix,
+		intervalStyleEnabled:        evalCtx.SessionData().IntervalStyleEnabled,
+		dateStyleEnabled:            evalCtx.SessionData().DateStyleEnabled,
+		dateStyle:                   evalCtx.SessionData().GetDateStyle(),
+		intervalStyle:               evalCtx.SessionData().GetIntervalStyle(),
+		propagateInputOrdering:      evalCtx.SessionData().PropagateInputOrdering,
+		disallowFullTableScans:      evalCtx.SessionData().DisallowFullTableScans,
+		largeFullScanRows:           evalCtx.SessionData().LargeFullScanRows,
+		nullOrderedLast:             evalCtx.SessionData().NullOrderedLast,
+		costScansWithDefaultColSize: evalCtx.SessionData().CostScansWithDefaultColSize,
 	}
 	m.metadata.Init()
 	m.logPropsBuilder.init(evalCtx, m)
@@ -312,7 +314,8 @@ func (m *Memo) IsStale(
 		m.propagateInputOrdering != evalCtx.SessionData().PropagateInputOrdering ||
 		m.disallowFullTableScans != evalCtx.SessionData().DisallowFullTableScans ||
 		m.largeFullScanRows != evalCtx.SessionData().LargeFullScanRows ||
-		m.nullOrderedLast != evalCtx.SessionData().NullOrderedLast {
+		m.nullOrderedLast != evalCtx.SessionData().NullOrderedLast ||
+		m.costScansWithDefaultColSize != evalCtx.SessionData().CostScansWithDefaultColSize {
 		return true, nil
 	}
 

pkg/sql/opt/memo/memo_test.go

@@ -272,6 +272,12 @@ func TestMemoIsStale(t *testing.T) {
 	evalCtx.SessionData().NullOrderedLast = false
 	notStale()
 
+	// Stale enable cost scans with default column size.
+	evalCtx.SessionData().CostScansWithDefaultColSize = true
+	stale()
+	evalCtx.SessionData().CostScansWithDefaultColSize = false
+	notStale()
+
 	// Stale data sources and schema. Create new catalog so that data sources are
 	// recreated and can be modified independently.
 	catalog = testcat.New()

pkg/sql/sessiondatapb/local_only_session_data.proto

@@ -23,17 +23,17 @@ message LocalOnlySessionData {
   string save_tables_prefix = 1;
   // OptimizerFKCascadesLimit is the maximum number of cascading operations that
   // are run for a single query.
-  int64 optimizer_fk_cascades_limit = 2 [(gogoproto.customname)="OptimizerFKCascadesLimit"];
+  int64 optimizer_fk_cascades_limit = 2 [(gogoproto.customname) = "OptimizerFKCascadesLimit"];
   // StmtTimeout is the duration a query is permitted to run before it is
   // canceled by the session. If set to 0, there is no timeout.
-  int64 stmt_timeout = 3 [(gogoproto.casttype)="time.Duration"];
+  int64 stmt_timeout = 3 [(gogoproto.casttype) = "time.Duration"];
   // IdleInSessionTimeout is the duration a session is permitted to idle before
   // the session is canceled. If set to 0, there is no timeout.
-  int64 idle_in_session_timeout = 4 [(gogoproto.casttype)="time.Duration"];
+  int64 idle_in_session_timeout = 4 [(gogoproto.casttype) = "time.Duration"];
   // IdleInTransactionSessionTimeout is the duration a session is permitted to
   // idle in a transaction before the session is canceled.
   // If set to 0, there is no timeout.
-  int64 idle_in_transaction_session_timeout = 5 [(gogoproto.casttype)="time.Duration"];
+  int64 idle_in_transaction_session_timeout = 5 [(gogoproto.casttype) = "time.Duration"];
   // NoticeDisplaySeverity indicates the level of Severity to send notices for the given
   // session. This should ideally be of type pgnotice.DisplaySeverity, but cannot be done
   // due to a circular dependency.
@@ -76,7 +76,7 @@ message LocalOnlySessionData {
   bool safe_updates = 15;
   // PreferLookupJoinsForFKs causes foreign key operations to prefer lookup
   // joins.
-  bool prefer_lookup_joins_for_fks = 16 [(gogoproto.customname)="PreferLookupJoinsForFKs"];
+  bool prefer_lookup_joins_for_fks = 16 [(gogoproto.customname) = "PreferLookupJoinsForFKs"];
   // ZigzagJoinEnabled indicates whether the optimizer should try and plan a
   // zigzag join.
   bool zigzag_join_enabled = 17;
@@ -140,29 +140,29 @@ message LocalOnlySessionData {
   // ResultsBufferSize specifies the size at which the pgwire results buffer
   // will self-flush.
   int64 results_buffer_size = 38;
-	// PropagateInputOrdering indicates that when planning a subquery or CTE, the
-	// inner ordering should be propagated to the outer scope if the outer scope
-	// is unordered. PropagateInputOrdering is currently experimental.
-	bool propagate_input_ordering = 39;
+  // PropagateInputOrdering indicates that when planning a subquery or CTE, the
+  // inner ordering should be propagated to the outer scope if the outer scope
+  // is unordered. PropagateInputOrdering is currently experimental.
+  bool propagate_input_ordering = 39;
   // ExperimentalDistSQLPlanningMode indicates whether the experimental
   // DistSQL planning driven by the optimizer is enabled.
   int64 experimental_distsql_planning_mode = 40 [
-    (gogoproto.customname)="ExperimentalDistSQLPlanningMode",
-    (gogoproto.casttype)="ExperimentalDistSQLPlanningMode"
+    (gogoproto.customname) = "ExperimentalDistSQLPlanningMode",
+    (gogoproto.casttype) = "ExperimentalDistSQLPlanningMode"
   ];
   // DistSQLMode indicates whether to run queries using the distributed
   // execution engine.
   int64 dist_sql_mode = 41 [
-    (gogoproto.customname)="DistSQLMode",
-    (gogoproto.casttype)="DistSQLExecMode"
+    (gogoproto.customname) = "DistSQLMode",
+    (gogoproto.casttype) = "DistSQLExecMode"
   ];
   // SerialNormalizationMode indicates how to handle the SERIAL pseudo-type.
-  int64 serial_normalization_mode = 42 [(gogoproto.casttype)="SerialNormalizationMode"];
+  int64 serial_normalization_mode = 42 [(gogoproto.casttype) = "SerialNormalizationMode"];
   // NewSchemaChangerMode indicates whether to use the new schema changer.
-  int64 new_schema_changer_mode = 43 [(gogoproto.casttype)="NewSchemaChangerMode"];
+  int64 new_schema_changer_mode = 43 [(gogoproto.casttype) = "NewSchemaChangerMode"];
   // SequenceCache stores sequence values which have been cached using the
   // CACHE sequence option.
-  map<uint32, sessiondatapb.SequenceCacheEntry> sequence_cache = 44 [(gogoproto.casttype)="SequenceCache"];
+  map<uint32, sessiondatapb.SequenceCacheEntry> sequence_cache = 44 [(gogoproto.casttype) = "SequenceCache"];
   // PlacementEnabled indicates whether PLACEMENT can be used or not.
   bool placement_enabled = 45;
   // SessionUserProto is the name of the user which has performed a SET ROLE
@@ -222,6 +222,10 @@ message LocalOnlySessionData {
   // CheckFunctionBodies indicates whether functions are validated during
   // creation.
   bool check_function_bodies = 60;
+  // CostScansWithDefaultColSize is true when the optimizer should cost scans
+  // and joins using the same default number of bytes per column instead of
+  // column sizes from the AvgSize table statistic.
+  bool cost_scans_with_default_col_size = 61;
 
   ///////////////////////////////////////////////////////////////////////////
   // WARNING: consider whether a session parameter you're adding needs to  //

pkg/sql/sessiondatapb/session_data.proto

@@ -72,7 +72,7 @@ message SessionData {
   // attempting to acquire a lock on a key or while blocking on an existing
   // lock in order to perform a non-locking read on a key.
   google.protobuf.Duration lock_timeout = 15 [(gogoproto.nullable) = false,
-                                              (gogoproto.stdduration) = true];
+    (gogoproto.stdduration) = true];
   // Internal is true if this query came from InternalExecutor or an internal
   // planner.
   bool internal = 16;
@@ -103,7 +103,7 @@ message DataConversionConfig {
   // IntervalStyle indicates the style to parse and display intervals as.
   util.duration.IntervalStyle interval_style = 3;
   // DateStyle indicates the style to parse and display dates as.
-  util.timeutil.pgdate.DateStyle date_style = 4 [(gogoproto.nullable)=false];
+  util.timeutil.pgdate.DateStyle date_style = 4 [(gogoproto.nullable) = false];
 }
 
 // BytesEncodeFormat is the configuration for bytes to string conversions.

pkg/sql/vars.go

@@ -1889,6 +1889,25 @@ var varGen = map[string]sessionVar{
 			return rowexec.ParallelizeMultiKeyLookupJoinsEnabled.String(sv)
 		},
 	},
+
+	// TODO(harding): Remove this when costing scans based on average column size
+	// is fully supported.
+	// CockroachDB extension.
+	`cost_scans_with_default_col_size`: {
+		GetStringVal: makePostgresBoolGetStringValFn(`cost_scans_with_default_col_size`),
+		Set: func(_ context.Context, m sessionDataMutator, s string) error {
+			b, err := paramparse.ParseBoolVar(`cost_scans_with_default_col_size`, s)
+			if err != nil {
+				return err
+			}
+			m.SetCostScansWithDefaultColSize(b)
+			return nil
+		},
+		Get: func(evalCtx *extendedEvalContext) (string, error) {
+			return formatBoolAsPostgresSetting(evalCtx.SessionData().CostScansWithDefaultColSize), nil
+		},
+		GlobalDefault: globalFalse,
+	},
 }
 
 const compatErrMsg = "this parameter is currently recognized only for compatibility and has no effect in CockroachDB."