sessionctx: add tidb_analyze_column_options global variable (#54200)

ref #53567

Author: 0xPoe

docs/design/2024-05-23-predicate-columns.md

@@ -26,7 +26,7 @@
     - [Performance Tests](#performance-tests)
   - [Impacts \& Risks](#impacts--risks)
     - [If new predicate columns appear, they cannot be analyzed in time](#if-new-predicate-columns-appear-they-cannot-be-analyzed-in-time)
-    - [Use PREDICATE COLUMNS when your workload's query pattern is relatively stable](#use-predicate-columns-when-your-workloads-query-pattern-is--relatively-stable)
+    - [Use PREDICATE COLUMNS when your workload's query pattern is relatively stable](#use-predicate-columns-when-your-workloads-query-pattern-is-relatively-stable)
   - [Investigation \& Alternatives](#investigation--alternatives)
     - [CRDB](#crdb)
       - [Summary](#summary)
@@ -214,14 +214,14 @@ In the experimental implementation, we introduce a new global variable `tidb_ena
 
 But because we decided to track all columns by default, so it becomes unnecessary to use this variable. We will mark it deprecated and remove it in the future.
 
-In this feature, we introduce a new global variable `tidb_analyze_default_column_choice` to control whether to use predicate columns or all columns in the analyze process.
+In this feature, we introduce a new global variable `tidb_analyze_column_options` to control whether to use predicate columns or all columns in the analyze process.
 
 Users can set this variable to `ALL` or `PREDICATE` to analyze all columns or only predicate columns. The default value will be `PREDICATE` after this feature is fully implemented.
 
 ```sql
-SET GLOBAL tidb_analyze_default_column_choice = 'PREDICATE';
+SET GLOBAL tidb_analyze_column_options = 'PREDICATE';
 
-SET GLOBAL tidb_analyze_default_column_choice = 'ALL';
+SET GLOBAL tidb_analyze_column_options = 'ALL';
 ```
 
 | Value     | Description                                                    |

pkg/executor/set_test.go

@@ -681,6 +681,14 @@ func TestSetVar(t *testing.T) {
 	require.Error(t, tk.ExecToErr("set tidb_enable_column_tracking = 0"))
 	require.Error(t, tk.ExecToErr("set global tidb_enable_column_tracking = -1"))
 
+	// test for tidb_analyze_column_options
+	tk.MustQuery("select @@tidb_analyze_column_options").Check(testkit.Rows("ALL"))
+	tk.MustExec("set global tidb_analyze_column_options = 'PREDICATE'")
+	tk.MustQuery("select @@tidb_analyze_column_options").Check(testkit.Rows("PREDICATE"))
+	tk.MustExec("set global tidb_analyze_column_options = 'all'")
+	tk.MustQuery("select @@tidb_analyze_column_options").Check(testkit.Rows("ALL"))
+	require.Error(t, tk.ExecToErr("set global tidb_analyze_column_options = 'UNKNOWN'"))
+
 	// test for tidb_ignore_prepared_cache_close_stmt
 	tk.MustQuery("select @@global.tidb_ignore_prepared_cache_close_stmt").Check(testkit.Rows("0")) // default value is 0
 	tk.MustExec("set global tidb_ignore_prepared_cache_close_stmt=1")

pkg/sessionctx/variable/sysvar.go

@@ -34,6 +34,7 @@ import (
 	"github.com/pingcap/tidb/pkg/parser"
 	"github.com/pingcap/tidb/pkg/parser/ast"
 	"github.com/pingcap/tidb/pkg/parser/charset"
+	"github.com/pingcap/tidb/pkg/parser/model"
 	"github.com/pingcap/tidb/pkg/parser/mysql"
 	"github.com/pingcap/tidb/pkg/planner/util/fixcontrol"
 	"github.com/pingcap/tidb/pkg/privilege/privileges/ldap"
@@ -1010,7 +1011,33 @@ var defaultSysVars = []*SysVar{
 			RunAutoAnalyze.Store(TiDBOptOn(val))
 			return nil
 		},
-	}, {
+	},
+	{
+		Scope: ScopeGlobal,
+		Name:  TiDBAnalyzeColumnOptions,
+		Value: DefTiDBAnalyzeColumnOptions,
+		Type:  TypeStr,
+		GetGlobal: func(ctx context.Context, s *SessionVars) (string, error) {
+			return AnalyzeColumnOptions.Load(), nil
+		},
+		SetGlobal: func(_ context.Context, s *SessionVars, val string) error {
+			AnalyzeColumnOptions.Store(strings.ToUpper(val))
+			return nil
+		},
+		Validation: func(s *SessionVars, normalizedValue string, originalValue string, scope ScopeFlag) (string, error) {
+			choice := strings.ToUpper(normalizedValue)
+			if choice != model.AllColumns.String() && choice != model.PredicateColumns.String() {
+				return "", errors.Errorf(
+					"invalid value for %s, it should be either '%s' or '%s'",
+					TiDBAnalyzeColumnOptions,
+					model.AllColumns.String(),
+					model.PredicateColumns.String(),
+				)
+			}
+			return normalizedValue, nil
+		},
+	},
+	{
 		Scope: ScopeGlobal, Name: TiDBEnableAutoAnalyzePriorityQueue, Value: BoolToOnOff(DefTiDBEnableAutoAnalyzePriorityQueue), Type: TypeBool,
 		GetGlobal: func(_ context.Context, s *SessionVars) (string, error) {
 			return BoolToOnOff(EnableAutoAnalyzePriorityQueue.Load()), nil

pkg/sessionctx/variable/tidb_vars.go

@@ -978,6 +978,11 @@ const (
 	// TiDBEnableColumnTracking enables collecting predicate columns.
 	// DEPRECATED: This variable is deprecated, please do not use this variable.
 	TiDBEnableColumnTracking = "tidb_enable_column_tracking"
+	// TiDBAnalyzeColumnOptions specifies the default column selection strategy for both manual and automatic analyze operations.
+	// It accepts two values:
+	// `PREDICATE`: Analyze only the columns that are used in the predicates of the query.
+	// `ALL`: Analyze all columns in the table.
+	TiDBAnalyzeColumnOptions = "tidb_analyze_column_options"
 	// TiDBDisableColumnTrackingTime records the last time TiDBEnableColumnTracking is set off.
 	// It is used to invalidate the collected predicate columns after turning off TiDBEnableColumnTracking, which avoids physical deletion.
 	// It doesn't have cache in memory, and we directly get/set the variable value from/to mysql.tidb.
@@ -1373,6 +1378,7 @@ const (
 	DefTiDBMemQuotaAnalyze                         = -1
 	DefTiDBEnableAutoAnalyze                       = true
 	DefTiDBEnableAutoAnalyzePriorityQueue          = true
+	DefTiDBAnalyzeColumnOptions                    = "ALL"
 	DefTiDBMemOOMAction                            = "CANCEL"
 	DefTiDBMaxAutoAnalyzeTime                      = 12 * 60 * 60
 	DefTiDBEnablePrepPlanCache                     = true
@@ -1499,20 +1505,29 @@ const (
 
 // Process global variables.
 var (
-	ProcessGeneralLog                    = atomic.NewBool(false)
-	RunAutoAnalyze                       = atomic.NewBool(DefTiDBEnableAutoAnalyze)
-	EnableAutoAnalyzePriorityQueue       = atomic.NewBool(DefTiDBEnableAutoAnalyzePriorityQueue)
-	GlobalLogMaxDays                     = atomic.NewInt32(int32(config.GetGlobalConfig().Log.File.MaxDays))
-	QueryLogMaxLen                       = atomic.NewInt32(DefTiDBQueryLogMaxLen)
-	EnablePProfSQLCPU                    = atomic.NewBool(false)
-	EnableBatchDML                       = atomic.NewBool(false)
-	EnableTmpStorageOnOOM                = atomic.NewBool(DefTiDBEnableTmpStorageOnOOM)
-	ddlReorgWorkerCounter          int32 = DefTiDBDDLReorgWorkerCount
-	ddlReorgBatchSize              int32 = DefTiDBDDLReorgBatchSize
-	ddlFlashbackConcurrency        int32 = DefTiDBDDLFlashbackConcurrency
-	ddlErrorCountLimit             int64 = DefTiDBDDLErrorCountLimit
-	ddlReorgRowFormat              int64 = DefTiDBRowFormatV2
-	maxDeltaSchemaCount            int64 = DefTiDBMaxDeltaSchemaCount
+	ProcessGeneralLog              = atomic.NewBool(false)
+	RunAutoAnalyze                 = atomic.NewBool(DefTiDBEnableAutoAnalyze)
+	EnableAutoAnalyzePriorityQueue = atomic.NewBool(DefTiDBEnableAutoAnalyzePriorityQueue)
+	// AnalyzeColumnOptions is a global variable that indicates the default column choice for ANALYZE.
+	// The value of this variable is a string that can be one of the following values:
+	// "PREDICATE", "ALL".
+	// The behavior of the analyze operation depends on the value of `tidb_persist_analyze_options`:
+	// 1. If `tidb_persist_analyze_options` is enabled and the column choice from the analyze options record is set to `default`,
+	//    the value of `tidb_analyze_column_options` determines the behavior of the analyze operation.
+	// 2. If `tidb_persist_analyze_options` is disabled, `tidb_analyze_column_options` is used directly to decide
+	//    whether to analyze all columns or just the predicate columns.
+	AnalyzeColumnOptions          = atomic.NewString(DefTiDBAnalyzeColumnOptions)
+	GlobalLogMaxDays              = atomic.NewInt32(int32(config.GetGlobalConfig().Log.File.MaxDays))
+	QueryLogMaxLen                = atomic.NewInt32(DefTiDBQueryLogMaxLen)
+	EnablePProfSQLCPU             = atomic.NewBool(false)
+	EnableBatchDML                = atomic.NewBool(false)
+	EnableTmpStorageOnOOM         = atomic.NewBool(DefTiDBEnableTmpStorageOnOOM)
+	ddlReorgWorkerCounter   int32 = DefTiDBDDLReorgWorkerCount
+	ddlReorgBatchSize       int32 = DefTiDBDDLReorgBatchSize
+	ddlFlashbackConcurrency int32 = DefTiDBDDLFlashbackConcurrency
+	ddlErrorCountLimit      int64 = DefTiDBDDLErrorCountLimit
+	ddlReorgRowFormat       int64 = DefTiDBRowFormatV2
+	maxDeltaSchemaCount     int64 = DefTiDBMaxDeltaSchemaCount
 	// DDLSlowOprThreshold is the threshold for ddl slow operations, uint is millisecond.
 	DDLSlowOprThreshold                  = config.GetGlobalConfig().Instance.DDLSlowOprThreshold
 	ForcePriority                        = int32(DefTiDBForcePriority)