Files
tidb/ddl/column.go

1737 lines
64 KiB
Go

// Copyright 2015 PingCAP, Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// See the License for the specific language governing permissions and
// limitations under the License.
package ddl
import (
"bytes"
"fmt"
"math"
"math/bits"
"strings"
"sync/atomic"
"time"
"github.com/cznic/mathutil"
"github.com/pingcap/errors"
"github.com/pingcap/failpoint"
"github.com/pingcap/parser/ast"
"github.com/pingcap/parser/model"
"github.com/pingcap/parser/mysql"
"github.com/pingcap/parser/terror"
ddlutil "github.com/pingcap/tidb/ddl/util"
"github.com/pingcap/tidb/infoschema"
"github.com/pingcap/tidb/kv"
"github.com/pingcap/tidb/meta"
"github.com/pingcap/tidb/meta/autoid"
"github.com/pingcap/tidb/metrics"
"github.com/pingcap/tidb/sessionctx"
"github.com/pingcap/tidb/table"
"github.com/pingcap/tidb/tablecodec"
"github.com/pingcap/tidb/types"
"github.com/pingcap/tidb/types/json"
"github.com/pingcap/tidb/util"
"github.com/pingcap/tidb/util/logutil"
decoder "github.com/pingcap/tidb/util/rowDecoder"
"github.com/pingcap/tidb/util/sqlexec"
"github.com/pingcap/tidb/util/timeutil"
"github.com/prometheus/client_golang/prometheus"
"go.uber.org/zap"
)
// adjustColumnInfoInAddColumn is used to set the correct position of column info when adding column.
// 1. The added column was append at the end of tblInfo.Columns, due to ddl state was not public then.
// It should be moved to the correct position when the ddl state to be changed to public.
// 2. The offset of column should also to be set to the right value.
func adjustColumnInfoInAddColumn(tblInfo *model.TableInfo, offset int) {
oldCols := tblInfo.Columns
newCols := make([]*model.ColumnInfo, 0, len(oldCols))
newCols = append(newCols, oldCols[:offset]...)
newCols = append(newCols, oldCols[len(oldCols)-1])
newCols = append(newCols, oldCols[offset:len(oldCols)-1]...)
// Adjust column offset.
offsetChanged := make(map[int]int, len(newCols)-offset-1)
for i := offset + 1; i < len(newCols); i++ {
offsetChanged[newCols[i].Offset] = i
newCols[i].Offset = i
}
newCols[offset].Offset = offset
// Update index column offset info.
// TODO: There may be some corner cases for index column offsets, we may check this later.
for _, idx := range tblInfo.Indices {
for _, col := range idx.Columns {
newOffset, ok := offsetChanged[col.Offset]
if ok {
col.Offset = newOffset
}
}
}
tblInfo.Columns = newCols
}
// adjustColumnInfoInDropColumn is used to set the correct position of column info when dropping column.
// 1. The offset of column should to be set to the last of the columns.
// 2. The dropped column is moved to the end of tblInfo.Columns, due to it was not public any more.
func adjustColumnInfoInDropColumn(tblInfo *model.TableInfo, offset int) {
oldCols := tblInfo.Columns
// Adjust column offset.
offsetChanged := make(map[int]int, len(oldCols)-offset-1)
for i := offset + 1; i < len(oldCols); i++ {
offsetChanged[oldCols[i].Offset] = i - 1
oldCols[i].Offset = i - 1
}
oldCols[offset].Offset = len(oldCols) - 1
// For expression index, we drop hidden columns and index simultaneously.
// So we need to change the offset of expression index.
offsetChanged[offset] = len(oldCols) - 1
// Update index column offset info.
// TODO: There may be some corner cases for index column offsets, we may check this later.
for _, idx := range tblInfo.Indices {
for _, col := range idx.Columns {
newOffset, ok := offsetChanged[col.Offset]
if ok {
col.Offset = newOffset
}
}
}
newCols := make([]*model.ColumnInfo, 0, len(oldCols))
newCols = append(newCols, oldCols[:offset]...)
newCols = append(newCols, oldCols[offset+1:]...)
newCols = append(newCols, oldCols[offset])
tblInfo.Columns = newCols
}
func createColumnInfo(tblInfo *model.TableInfo, colInfo *model.ColumnInfo, pos *ast.ColumnPosition) (*model.ColumnInfo, *ast.ColumnPosition, int, error) {
// Check column name duplicate.
cols := tblInfo.Columns
offset := len(cols)
// Should initialize pos when it is nil.
if pos == nil {
pos = &ast.ColumnPosition{}
}
// Get column offset.
if pos.Tp == ast.ColumnPositionFirst {
offset = 0
} else if pos.Tp == ast.ColumnPositionAfter {
c := model.FindColumnInfo(cols, pos.RelativeColumn.Name.L)
if c == nil {
return nil, pos, 0, infoschema.ErrColumnNotExists.GenWithStackByArgs(pos.RelativeColumn, tblInfo.Name)
}
// Insert offset is after the mentioned column.
offset = c.Offset + 1
}
colInfo.ID = allocateColumnID(tblInfo)
colInfo.State = model.StateNone
// To support add column asynchronous, we should mark its offset as the last column.
// So that we can use origin column offset to get value from row.
colInfo.Offset = len(cols)
// Append the column info to the end of the tblInfo.Columns.
// It will reorder to the right offset in "Columns" when it state change to public.
tblInfo.Columns = append(cols, colInfo)
return colInfo, pos, offset, nil
}
func checkAddColumn(t *meta.Meta, job *model.Job) (*model.TableInfo, *model.ColumnInfo, *model.ColumnInfo, *ast.ColumnPosition, int, error) {
schemaID := job.SchemaID
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, schemaID)
if err != nil {
return nil, nil, nil, nil, 0, errors.Trace(err)
}
col := &model.ColumnInfo{}
pos := &ast.ColumnPosition{}
offset := 0
err = job.DecodeArgs(col, pos, &offset)
if err != nil {
job.State = model.JobStateCancelled
return nil, nil, nil, nil, 0, errors.Trace(err)
}
columnInfo := model.FindColumnInfo(tblInfo.Columns, col.Name.L)
if columnInfo != nil {
if columnInfo.State == model.StatePublic {
// We already have a column with the same column name.
job.State = model.JobStateCancelled
return nil, nil, nil, nil, 0, infoschema.ErrColumnExists.GenWithStackByArgs(col.Name)
}
}
return tblInfo, columnInfo, col, pos, offset, nil
}
func onAddColumn(d *ddlCtx, t *meta.Meta, job *model.Job) (ver int64, err error) {
// Handle the rolling back job.
if job.IsRollingback() {
ver, err = onDropColumn(t, job)
if err != nil {
return ver, errors.Trace(err)
}
return ver, nil
}
failpoint.Inject("errorBeforeDecodeArgs", func(val failpoint.Value) {
if val.(bool) {
failpoint.Return(ver, errors.New("occur an error before decode args"))
}
})
tblInfo, columnInfo, col, pos, offset, err := checkAddColumn(t, job)
if err != nil {
return ver, errors.Trace(err)
}
if columnInfo == nil {
columnInfo, _, offset, err = createColumnInfo(tblInfo, col, pos)
if err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
logutil.BgLogger().Info("[ddl] run add column job", zap.String("job", job.String()), zap.Reflect("columnInfo", *columnInfo), zap.Int("offset", offset))
// Set offset arg to job.
if offset != 0 {
job.Args = []interface{}{columnInfo, pos, offset}
}
if err = checkAddColumnTooManyColumns(len(tblInfo.Columns)); err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
}
originalState := columnInfo.State
switch columnInfo.State {
case model.StateNone:
// none -> delete only
job.SchemaState = model.StateDeleteOnly
columnInfo.State = model.StateDeleteOnly
ver, err = updateVersionAndTableInfoWithCheck(t, job, tblInfo, originalState != columnInfo.State)
case model.StateDeleteOnly:
// delete only -> write only
job.SchemaState = model.StateWriteOnly
columnInfo.State = model.StateWriteOnly
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != columnInfo.State)
case model.StateWriteOnly:
// write only -> reorganization
job.SchemaState = model.StateWriteReorganization
columnInfo.State = model.StateWriteReorganization
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != columnInfo.State)
case model.StateWriteReorganization:
// reorganization -> public
// Adjust table column offset.
adjustColumnInfoInAddColumn(tblInfo, offset)
columnInfo.State = model.StatePublic
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != columnInfo.State)
if err != nil {
return ver, errors.Trace(err)
}
// Finish this job.
job.FinishTableJob(model.JobStateDone, model.StatePublic, ver, tblInfo)
asyncNotifyEvent(d, &ddlutil.Event{Tp: model.ActionAddColumn, TableInfo: tblInfo, ColumnInfos: []*model.ColumnInfo{columnInfo}})
default:
err = ErrInvalidDDLState.GenWithStackByArgs("column", columnInfo.State)
}
return ver, errors.Trace(err)
}
func checkAddColumns(t *meta.Meta, job *model.Job) (*model.TableInfo, []*model.ColumnInfo, []*model.ColumnInfo, []*ast.ColumnPosition, []int, []bool, error) {
schemaID := job.SchemaID
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, schemaID)
if err != nil {
return nil, nil, nil, nil, nil, nil, errors.Trace(err)
}
columns := []*model.ColumnInfo{}
positions := []*ast.ColumnPosition{}
offsets := []int{}
ifNotExists := []bool{}
err = job.DecodeArgs(&columns, &positions, &offsets, &ifNotExists)
if err != nil {
job.State = model.JobStateCancelled
return nil, nil, nil, nil, nil, nil, errors.Trace(err)
}
columnInfos := make([]*model.ColumnInfo, 0, len(columns))
newColumns := make([]*model.ColumnInfo, 0, len(columns))
newPositions := make([]*ast.ColumnPosition, 0, len(columns))
newOffsets := make([]int, 0, len(columns))
newIfNotExists := make([]bool, 0, len(columns))
for i, col := range columns {
columnInfo := model.FindColumnInfo(tblInfo.Columns, col.Name.L)
if columnInfo != nil {
if columnInfo.State == model.StatePublic {
// We already have a column with the same column name.
if ifNotExists[i] {
// TODO: Should return a warning.
logutil.BgLogger().Warn("[ddl] check add columns, duplicate column", zap.Stringer("col", col.Name))
continue
}
job.State = model.JobStateCancelled
return nil, nil, nil, nil, nil, nil, infoschema.ErrColumnExists.GenWithStackByArgs(col.Name)
}
columnInfos = append(columnInfos, columnInfo)
}
newColumns = append(newColumns, columns[i])
newPositions = append(newPositions, positions[i])
newOffsets = append(newOffsets, offsets[i])
newIfNotExists = append(newIfNotExists, ifNotExists[i])
}
return tblInfo, columnInfos, newColumns, newPositions, newOffsets, newIfNotExists, nil
}
func setColumnsState(columnInfos []*model.ColumnInfo, state model.SchemaState) {
for i := range columnInfos {
columnInfos[i].State = state
}
}
func setIndicesState(indexInfos []*model.IndexInfo, state model.SchemaState) {
for _, indexInfo := range indexInfos {
indexInfo.State = state
}
}
func onAddColumns(d *ddlCtx, t *meta.Meta, job *model.Job) (ver int64, err error) {
// Handle the rolling back job.
if job.IsRollingback() {
ver, err = onDropColumns(t, job)
if err != nil {
return ver, errors.Trace(err)
}
return ver, nil
}
failpoint.Inject("errorBeforeDecodeArgs", func(val failpoint.Value) {
if val.(bool) {
failpoint.Return(ver, errors.New("occur an error before decode args"))
}
})
tblInfo, columnInfos, columns, positions, offsets, ifNotExists, err := checkAddColumns(t, job)
if err != nil {
return ver, errors.Trace(err)
}
if len(columnInfos) == 0 {
if len(columns) == 0 {
job.State = model.JobStateCancelled
return ver, nil
}
for i := range columns {
columnInfo, pos, offset, err := createColumnInfo(tblInfo, columns[i], positions[i])
if err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
logutil.BgLogger().Info("[ddl] run add columns job", zap.String("job", job.String()), zap.Reflect("columnInfo", *columnInfo), zap.Int("offset", offset))
positions[i] = pos
offsets[i] = offset
if err = checkAddColumnTooManyColumns(len(tblInfo.Columns)); err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
columnInfos = append(columnInfos, columnInfo)
}
// Set arg to job.
job.Args = []interface{}{columnInfos, positions, offsets, ifNotExists}
}
originalState := columnInfos[0].State
switch columnInfos[0].State {
case model.StateNone:
// none -> delete only
job.SchemaState = model.StateDeleteOnly
setColumnsState(columnInfos, model.StateDeleteOnly)
ver, err = updateVersionAndTableInfoWithCheck(t, job, tblInfo, originalState != columnInfos[0].State)
case model.StateDeleteOnly:
// delete only -> write only
job.SchemaState = model.StateWriteOnly
setColumnsState(columnInfos, model.StateWriteOnly)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != columnInfos[0].State)
case model.StateWriteOnly:
// write only -> reorganization
job.SchemaState = model.StateWriteReorganization
setColumnsState(columnInfos, model.StateWriteReorganization)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != columnInfos[0].State)
case model.StateWriteReorganization:
// reorganization -> public
// Adjust table column offsets.
oldCols := tblInfo.Columns[:len(tblInfo.Columns)-len(offsets)]
newCols := tblInfo.Columns[len(tblInfo.Columns)-len(offsets):]
tblInfo.Columns = oldCols
for i := range offsets {
// For multiple columns with after position, should adjust offsets.
// e.g. create table t(a int);
// alter table t add column b int after a, add column c int after a;
// alter table t add column a1 int after a, add column b1 int after b, add column c1 int after c;
// alter table t add column a1 int after a, add column b1 int first;
if positions[i].Tp == ast.ColumnPositionAfter {
for j := 0; j < i; j++ {
if (positions[j].Tp == ast.ColumnPositionAfter && offsets[j] < offsets[i]) || positions[j].Tp == ast.ColumnPositionFirst {
offsets[i]++
}
}
}
tblInfo.Columns = append(tblInfo.Columns, newCols[i])
adjustColumnInfoInAddColumn(tblInfo, offsets[i])
}
setColumnsState(columnInfos, model.StatePublic)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != columnInfos[0].State)
if err != nil {
return ver, errors.Trace(err)
}
// Finish this job.
job.FinishTableJob(model.JobStateDone, model.StatePublic, ver, tblInfo)
asyncNotifyEvent(d, &ddlutil.Event{Tp: model.ActionAddColumns, TableInfo: tblInfo, ColumnInfos: columnInfos})
default:
err = ErrInvalidDDLState.GenWithStackByArgs("column", columnInfos[0].State)
}
return ver, errors.Trace(err)
}
func onDropColumns(t *meta.Meta, job *model.Job) (ver int64, _ error) {
tblInfo, colInfos, delCount, idxInfos, err := checkDropColumns(t, job)
if err != nil {
return ver, errors.Trace(err)
}
if len(colInfos) == 0 {
job.State = model.JobStateCancelled
return ver, nil
}
originalState := colInfos[0].State
switch colInfos[0].State {
case model.StatePublic:
// public -> write only
job.SchemaState = model.StateWriteOnly
setColumnsState(colInfos, model.StateWriteOnly)
setIndicesState(idxInfos, model.StateWriteOnly)
for _, colInfo := range colInfos {
err = checkDropColumnForStatePublic(tblInfo, colInfo)
if err != nil {
return ver, errors.Trace(err)
}
}
ver, err = updateVersionAndTableInfoWithCheck(t, job, tblInfo, originalState != colInfos[0].State)
case model.StateWriteOnly:
// write only -> delete only
job.SchemaState = model.StateDeleteOnly
setColumnsState(colInfos, model.StateDeleteOnly)
setIndicesState(idxInfos, model.StateDeleteOnly)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != colInfos[0].State)
case model.StateDeleteOnly:
// delete only -> reorganization
job.SchemaState = model.StateDeleteReorganization
setColumnsState(colInfos, model.StateDeleteReorganization)
setIndicesState(idxInfos, model.StateDeleteReorganization)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != colInfos[0].State)
case model.StateDeleteReorganization:
// reorganization -> absent
// All reorganization jobs are done, drop this column.
if len(idxInfos) > 0 {
newIndices := make([]*model.IndexInfo, 0, len(tblInfo.Indices))
for _, idx := range tblInfo.Indices {
if !indexInfoContains(idx.ID, idxInfos) {
newIndices = append(newIndices, idx)
}
}
tblInfo.Indices = newIndices
}
indexIDs := indexInfosToIDList(idxInfos)
tblInfo.Columns = tblInfo.Columns[:len(tblInfo.Columns)-delCount]
setColumnsState(colInfos, model.StateNone)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != colInfos[0].State)
if err != nil {
return ver, errors.Trace(err)
}
// Finish this job.
if job.IsRollingback() {
job.FinishTableJob(model.JobStateRollbackDone, model.StateNone, ver, tblInfo)
} else {
job.FinishTableJob(model.JobStateDone, model.StateNone, ver, tblInfo)
job.Args = append(job.Args, indexIDs, getPartitionIDs(tblInfo))
}
default:
err = errInvalidDDLJob.GenWithStackByArgs("table", tblInfo.State)
}
return ver, errors.Trace(err)
}
func checkDropColumns(t *meta.Meta, job *model.Job) (*model.TableInfo, []*model.ColumnInfo, int, []*model.IndexInfo, error) {
schemaID := job.SchemaID
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, schemaID)
if err != nil {
return nil, nil, 0, nil, errors.Trace(err)
}
var colNames []model.CIStr
var ifExists []bool
err = job.DecodeArgs(&colNames, &ifExists)
if err != nil {
job.State = model.JobStateCancelled
return nil, nil, 0, nil, errors.Trace(err)
}
newColNames := make([]model.CIStr, 0, len(colNames))
colInfos := make([]*model.ColumnInfo, 0, len(colNames))
newIfExists := make([]bool, 0, len(colNames))
indexInfos := make([]*model.IndexInfo, 0)
for i, colName := range colNames {
colInfo := model.FindColumnInfo(tblInfo.Columns, colName.L)
if colInfo == nil || colInfo.Hidden {
if ifExists[i] {
// TODO: Should return a warning.
logutil.BgLogger().Warn(fmt.Sprintf("column %s doesn't exist", colName))
continue
}
job.State = model.JobStateCancelled
return nil, nil, 0, nil, ErrCantDropFieldOrKey.GenWithStack("column %s doesn't exist", colName)
}
if err = isDroppableColumn(tblInfo, colName); err != nil {
job.State = model.JobStateCancelled
return nil, nil, 0, nil, errors.Trace(err)
}
newColNames = append(newColNames, colName)
newIfExists = append(newIfExists, ifExists[i])
colInfos = append(colInfos, colInfo)
idxInfos := listIndicesWithColumn(colName.L, tblInfo.Indices)
indexInfos = append(indexInfos, idxInfos...)
}
job.Args = []interface{}{newColNames, newIfExists}
return tblInfo, colInfos, len(colInfos), indexInfos, nil
}
func checkDropColumnForStatePublic(tblInfo *model.TableInfo, colInfo *model.ColumnInfo) (err error) {
// Set this column's offset to the last and reset all following columns' offsets.
adjustColumnInfoInDropColumn(tblInfo, colInfo.Offset)
// When the dropping column has not-null flag and it hasn't the default value, we can backfill the column value like "add column".
// NOTE: If the state of StateWriteOnly can be rollbacked, we'd better reconsider the original default value.
// And we need consider the column without not-null flag.
if colInfo.GetOriginDefaultValue() == nil && mysql.HasNotNullFlag(colInfo.Flag) {
// If the column is timestamp default current_timestamp, and DDL owner is new version TiDB that set column.Version to 1,
// then old TiDB update record in the column write only stage will uses the wrong default value of the dropping column.
// Because new version of the column default value is UTC time, but old version TiDB will think the default value is the time in system timezone.
// But currently will be ok, because we can't cancel the drop column job when the job is running,
// so the column will be dropped succeed and client will never see the wrong default value of the dropped column.
// More info about this problem, see PR#9115.
originDefVal, err := generateOriginDefaultValue(colInfo)
if err != nil {
return err
}
return colInfo.SetOriginDefaultValue(originDefVal)
}
return nil
}
func onDropColumn(t *meta.Meta, job *model.Job) (ver int64, _ error) {
tblInfo, colInfo, idxInfos, err := checkDropColumn(t, job)
if err != nil {
return ver, errors.Trace(err)
}
originalState := colInfo.State
switch colInfo.State {
case model.StatePublic:
// public -> write only
job.SchemaState = model.StateWriteOnly
colInfo.State = model.StateWriteOnly
setIndicesState(idxInfos, model.StateWriteOnly)
err = checkDropColumnForStatePublic(tblInfo, colInfo)
if err != nil {
return ver, errors.Trace(err)
}
ver, err = updateVersionAndTableInfoWithCheck(t, job, tblInfo, originalState != colInfo.State)
case model.StateWriteOnly:
// write only -> delete only
job.SchemaState = model.StateDeleteOnly
colInfo.State = model.StateDeleteOnly
setIndicesState(idxInfos, model.StateDeleteOnly)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != colInfo.State)
case model.StateDeleteOnly:
// delete only -> reorganization
job.SchemaState = model.StateDeleteReorganization
colInfo.State = model.StateDeleteReorganization
setIndicesState(idxInfos, model.StateDeleteReorganization)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != colInfo.State)
case model.StateDeleteReorganization:
// reorganization -> absent
// All reorganization jobs are done, drop this column.
if len(idxInfos) > 0 {
newIndices := make([]*model.IndexInfo, 0, len(tblInfo.Indices))
for _, idx := range tblInfo.Indices {
if !indexInfoContains(idx.ID, idxInfos) {
newIndices = append(newIndices, idx)
}
}
tblInfo.Indices = newIndices
}
indexIDs := indexInfosToIDList(idxInfos)
tblInfo.Columns = tblInfo.Columns[:len(tblInfo.Columns)-1]
colInfo.State = model.StateNone
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != colInfo.State)
if err != nil {
return ver, errors.Trace(err)
}
// Finish this job.
if job.IsRollingback() {
job.FinishTableJob(model.JobStateRollbackDone, model.StateNone, ver, tblInfo)
} else {
// We should set related index IDs for job
job.FinishTableJob(model.JobStateDone, model.StateNone, ver, tblInfo)
job.Args = append(job.Args, indexIDs, getPartitionIDs(tblInfo))
}
default:
err = errInvalidDDLJob.GenWithStackByArgs("table", tblInfo.State)
}
return ver, errors.Trace(err)
}
func checkDropColumn(t *meta.Meta, job *model.Job) (*model.TableInfo, *model.ColumnInfo, []*model.IndexInfo, error) {
schemaID := job.SchemaID
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, schemaID)
if err != nil {
return nil, nil, nil, errors.Trace(err)
}
var colName model.CIStr
err = job.DecodeArgs(&colName)
if err != nil {
job.State = model.JobStateCancelled
return nil, nil, nil, errors.Trace(err)
}
colInfo := model.FindColumnInfo(tblInfo.Columns, colName.L)
if colInfo == nil || colInfo.Hidden {
job.State = model.JobStateCancelled
return nil, nil, nil, ErrCantDropFieldOrKey.GenWithStack("column %s doesn't exist", colName)
}
if err = isDroppableColumn(tblInfo, colName); err != nil {
job.State = model.JobStateCancelled
return nil, nil, nil, errors.Trace(err)
}
idxInfos := listIndicesWithColumn(colName.L, tblInfo.Indices)
if len(idxInfos) > 0 {
for _, idxInfo := range idxInfos {
err = checkDropIndexOnAutoIncrementColumn(tblInfo, idxInfo)
if err != nil {
job.State = model.JobStateCancelled
return nil, nil, nil, err
}
}
}
return tblInfo, colInfo, idxInfos, nil
}
func onSetDefaultValue(t *meta.Meta, job *model.Job) (ver int64, _ error) {
newCol := &model.ColumnInfo{}
err := job.DecodeArgs(newCol)
if err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
return updateColumnDefaultValue(t, job, newCol, &newCol.Name)
}
func needChangeColumnData(oldCol, newCol *model.ColumnInfo) bool {
toUnsigned := mysql.HasUnsignedFlag(newCol.Flag)
originUnsigned := mysql.HasUnsignedFlag(oldCol.Flag)
needTruncationOrToggleSign := func() bool {
return newCol.Flen > 0 && newCol.Flen < oldCol.Flen || toUnsigned != originUnsigned
}
// Deal with the same type.
if oldCol.Tp == newCol.Tp {
switch oldCol.Tp {
case mysql.TypeNewDecimal:
// Since type decimal will encode the precision, frac, negative(signed) and wordBuf into storage together, there is no short
// cut to eliminate data reorg change for column type change between decimal.
return oldCol.Flen != newCol.Flen || oldCol.Decimal != newCol.Decimal || toUnsigned != originUnsigned
case mysql.TypeEnum, mysql.TypeSet:
return isElemsChangedToModifyColumn(oldCol.Elems, newCol.Elems)
}
return needTruncationOrToggleSign()
}
// Deal with the different type.
switch oldCol.Tp {
case mysql.TypeVarchar, mysql.TypeString, mysql.TypeVarString, mysql.TypeBlob, mysql.TypeTinyBlob, mysql.TypeMediumBlob, mysql.TypeLongBlob:
switch newCol.Tp {
case mysql.TypeVarchar, mysql.TypeString, mysql.TypeVarString, mysql.TypeBlob, mysql.TypeTinyBlob, mysql.TypeMediumBlob, mysql.TypeLongBlob:
return needTruncationOrToggleSign()
}
case mysql.TypeTiny, mysql.TypeShort, mysql.TypeInt24, mysql.TypeLong, mysql.TypeLonglong:
switch newCol.Tp {
case mysql.TypeTiny, mysql.TypeShort, mysql.TypeInt24, mysql.TypeLong, mysql.TypeLonglong:
return needTruncationOrToggleSign()
}
}
return true
}
func isElemsChangedToModifyColumn(oldElems, newElems []string) bool {
if len(newElems) < len(oldElems) {
return true
}
for index, oldElem := range oldElems {
newElem := newElems[index]
if oldElem != newElem {
return true
}
}
return false
}
type modifyColumnJobParameter struct {
newCol *model.ColumnInfo
oldColName *model.CIStr
modifyColumnTp byte
updatedAutoRandomBits uint64
changingCol *model.ColumnInfo
changingIdxs []*model.IndexInfo
pos *ast.ColumnPosition
}
func getModifyColumnInfo(t *meta.Meta, job *model.Job) (*model.DBInfo, *model.TableInfo, *model.ColumnInfo, *modifyColumnJobParameter, error) {
jobParam := &modifyColumnJobParameter{pos: &ast.ColumnPosition{}}
err := job.DecodeArgs(&jobParam.newCol, &jobParam.oldColName, jobParam.pos, &jobParam.modifyColumnTp, &jobParam.updatedAutoRandomBits, &jobParam.changingCol, &jobParam.changingIdxs)
if err != nil {
job.State = model.JobStateCancelled
return nil, nil, nil, jobParam, errors.Trace(err)
}
dbInfo, err := checkSchemaExistAndCancelNotExistJob(t, job)
if err != nil {
return nil, nil, nil, jobParam, errors.Trace(err)
}
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, job.SchemaID)
if err != nil {
return nil, nil, nil, jobParam, errors.Trace(err)
}
oldCol := model.FindColumnInfo(tblInfo.Columns, jobParam.oldColName.L)
if oldCol == nil || oldCol.State != model.StatePublic {
job.State = model.JobStateCancelled
return nil, nil, nil, jobParam, errors.Trace(infoschema.ErrColumnNotExists.GenWithStackByArgs(*(jobParam.oldColName), tblInfo.Name))
}
return dbInfo, tblInfo, oldCol, jobParam, errors.Trace(err)
}
func (w *worker) onModifyColumn(d *ddlCtx, t *meta.Meta, job *model.Job) (ver int64, _ error) {
dbInfo, tblInfo, oldCol, jobParam, err := getModifyColumnInfo(t, job)
if err != nil {
return ver, err
}
if job.IsRollingback() {
// For those column-type-change jobs which don't reorg the data.
if !needChangeColumnData(oldCol, jobParam.newCol) {
return rollbackModifyColumnJob(t, tblInfo, job, oldCol, jobParam.modifyColumnTp)
}
// For those column-type-change jobs which reorg the data.
return rollbackModifyColumnJobWithData(t, tblInfo, job, oldCol, jobParam)
}
// If we want to rename the column name, we need to check whether it already exists.
if jobParam.newCol.Name.L != jobParam.oldColName.L {
c := model.FindColumnInfo(tblInfo.Columns, jobParam.newCol.Name.L)
if c != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(infoschema.ErrColumnExists.GenWithStackByArgs(jobParam.newCol.Name))
}
}
failpoint.Inject("uninitializedOffsetAndState", func(val failpoint.Value) {
if val.(bool) {
if jobParam.newCol.State != model.StatePublic {
failpoint.Return(ver, errors.New("the column state is wrong"))
}
}
})
if jobParam.updatedAutoRandomBits > 0 {
if err := checkAndApplyNewAutoRandomBits(job, t, tblInfo, jobParam.newCol, jobParam.oldColName, jobParam.updatedAutoRandomBits); err != nil {
return ver, errors.Trace(err)
}
}
if !needChangeColumnData(oldCol, jobParam.newCol) {
return w.doModifyColumn(t, job, dbInfo, tblInfo, jobParam.newCol, oldCol, jobParam.pos)
}
if jobParam.changingCol == nil {
changingColPos := &ast.ColumnPosition{Tp: ast.ColumnPositionNone}
newColName := model.NewCIStr(genChangingColumnUniqueName(tblInfo, oldCol))
if mysql.HasPriKeyFlag(oldCol.Flag) {
job.State = model.JobStateCancelled
msg := "tidb_enable_change_column_type is true and this column has primary key flag"
return ver, errUnsupportedModifyColumn.GenWithStackByArgs(msg)
}
// TODO: Check whether we need to check OriginDefaultValue.
jobParam.changingCol = jobParam.newCol.Clone()
jobParam.changingCol.Name = newColName
jobParam.changingCol.ChangeStateInfo = &model.ChangeStateInfo{DependencyColumnOffset: oldCol.Offset}
_, _, _, err := createColumnInfo(tblInfo, jobParam.changingCol, changingColPos)
if err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
idxInfos, offsets := findIndexesByColName(tblInfo.Indices, oldCol.Name.L)
jobParam.changingIdxs = make([]*model.IndexInfo, 0, len(idxInfos))
for i, idxInfo := range idxInfos {
newIdxInfo := idxInfo.Clone()
newIdxInfo.Name = model.NewCIStr(genChangingIndexUniqueName(tblInfo, idxInfo))
newIdxInfo.ID = allocateIndexID(tblInfo)
newIdxInfo.Columns[offsets[i]].Name = newColName
newIdxInfo.Columns[offsets[i]].Offset = jobParam.changingCol.Offset
jobParam.changingIdxs = append(jobParam.changingIdxs, newIdxInfo)
}
tblInfo.Indices = append(tblInfo.Indices, jobParam.changingIdxs...)
} else {
tblInfo.Columns[len(tblInfo.Columns)-1] = jobParam.changingCol
copy(tblInfo.Indices[len(tblInfo.Indices)-len(jobParam.changingIdxs):], jobParam.changingIdxs)
}
return w.doModifyColumnTypeWithData(d, t, job, dbInfo, tblInfo, jobParam.changingCol, oldCol, jobParam.newCol.Name, jobParam.pos, jobParam.changingIdxs)
}
// rollbackModifyColumnJobWithData is used to rollback modify-column job which need to reorg the data.
func rollbackModifyColumnJobWithData(t *meta.Meta, tblInfo *model.TableInfo, job *model.Job, oldCol *model.ColumnInfo, jobParam *modifyColumnJobParameter) (ver int64, err error) {
// If the not-null change is included, we should clean the flag info in oldCol.
if jobParam.modifyColumnTp == mysql.TypeNull {
// Reset NotNullFlag flag.
tblInfo.Columns[oldCol.Offset].Flag = oldCol.Flag &^ mysql.NotNullFlag
// Reset PreventNullInsertFlag flag.
tblInfo.Columns[oldCol.Offset].Flag = oldCol.Flag &^ mysql.PreventNullInsertFlag
}
if jobParam.changingCol != nil {
// changingCol isn't nil means the job has been in the mid state. These appended changingCol and changingIndex should
// be removed from the tableInfo as well.
tblInfo.Columns = tblInfo.Columns[:len(tblInfo.Columns)-1]
tblInfo.Indices = tblInfo.Indices[:len(tblInfo.Indices)-len(jobParam.changingIdxs)]
}
ver, err = updateVersionAndTableInfoWithCheck(t, job, tblInfo, true)
if err != nil {
return ver, errors.Trace(err)
}
job.FinishTableJob(model.JobStateRollbackDone, model.StateNone, ver, tblInfo)
// Refactor the job args to add the abandoned temporary index ids into delete range table.
idxIDs := make([]int64, 0, len(jobParam.changingIdxs))
for _, idx := range jobParam.changingIdxs {
idxIDs = append(idxIDs, idx.ID)
}
job.Args = []interface{}{idxIDs, getPartitionIDs(tblInfo)}
return ver, nil
}
func (w *worker) doModifyColumnTypeWithData(
d *ddlCtx, t *meta.Meta, job *model.Job,
dbInfo *model.DBInfo, tblInfo *model.TableInfo, changingCol, oldCol *model.ColumnInfo,
colName model.CIStr, pos *ast.ColumnPosition, changingIdxs []*model.IndexInfo) (ver int64, _ error) {
var err error
originalState := changingCol.State
switch changingCol.State {
case model.StateNone:
// Column from null to not null.
if !mysql.HasNotNullFlag(oldCol.Flag) && mysql.HasNotNullFlag(changingCol.Flag) {
// Introduce the `mysql.PreventNullInsertFlag` flag to prevent users from inserting or updating null values.
err := modifyColsFromNull2NotNull(w, dbInfo, tblInfo, []*model.ColumnInfo{oldCol}, oldCol.Name, oldCol.Tp != changingCol.Tp)
if err != nil {
if ErrWarnDataTruncated.Equal(err) || errInvalidUseOfNull.Equal(err) {
job.State = model.JobStateRollingback
}
return ver, err
}
}
// none -> delete only
updateChangingInfo(changingCol, changingIdxs, model.StateDeleteOnly)
failpoint.Inject("mockInsertValueAfterCheckNull", func(val failpoint.Value) {
if valStr, ok := val.(string); ok {
var ctx sessionctx.Context
ctx, err := w.sessPool.get()
if err != nil {
failpoint.Return(ver, err)
}
defer w.sessPool.put(ctx)
_, _, err = ctx.(sqlexec.RestrictedSQLExecutor).ExecRestrictedSQL(valStr)
if err != nil {
job.State = model.JobStateCancelled
failpoint.Return(ver, err)
}
}
})
ver, err = updateVersionAndTableInfoWithCheck(t, job, tblInfo, originalState != changingCol.State)
if err != nil {
return ver, errors.Trace(err)
}
// Make sure job args change after `updateVersionAndTableInfoWithCheck`, otherwise, the job args will
// be updated in `updateDDLJob` even if it meets an error in `updateVersionAndTableInfoWithCheck`.
job.SchemaState = model.StateDeleteOnly
job.Args = append(job.Args, changingCol, changingIdxs)
case model.StateDeleteOnly:
// Column from null to not null.
if !mysql.HasNotNullFlag(oldCol.Flag) && mysql.HasNotNullFlag(changingCol.Flag) {
// Introduce the `mysql.PreventNullInsertFlag` flag to prevent users from inserting or updating null values.
err := modifyColsFromNull2NotNull(w, dbInfo, tblInfo, []*model.ColumnInfo{oldCol}, oldCol.Name, oldCol.Tp != changingCol.Tp)
if err != nil {
if ErrWarnDataTruncated.Equal(err) || errInvalidUseOfNull.Equal(err) {
job.State = model.JobStateRollingback
}
return ver, err
}
}
// delete only -> write only
updateChangingInfo(changingCol, changingIdxs, model.StateWriteOnly)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != changingCol.State)
if err != nil {
return ver, errors.Trace(err)
}
job.SchemaState = model.StateWriteOnly
case model.StateWriteOnly:
// write only -> reorganization
updateChangingInfo(changingCol, changingIdxs, model.StateWriteReorganization)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != changingCol.State)
if err != nil {
return ver, errors.Trace(err)
}
job.SchemaState = model.StateWriteReorganization
case model.StateWriteReorganization:
tbl, err := getTable(d.store, dbInfo.ID, tblInfo)
if err != nil {
return ver, errors.Trace(err)
}
reorgInfo, err := getReorgInfo(d, t, job, tbl, BuildElements(changingCol, changingIdxs))
if err != nil || reorgInfo.first {
// If we run reorg firstly, we should update the job snapshot version
// and then run the reorg next time.
return ver, errors.Trace(err)
}
err = w.runReorgJob(t, reorgInfo, tbl.Meta(), d.lease, func() (addIndexErr error) {
defer util.Recover(metrics.LabelDDL, "onModifyColumn",
func() {
addIndexErr = errCancelledDDLJob.GenWithStack("modify table `%v` column `%v` panic", tblInfo.Name, oldCol.Name)
}, false)
return w.updateColumnAndIndexes(tbl, oldCol, changingCol, changingIdxs, reorgInfo)
})
if err != nil {
if errWaitReorgTimeout.Equal(err) {
// If timeout, we should return, check for the owner and re-wait job done.
return ver, nil
}
if needRollbackData(err) {
if err1 := t.RemoveDDLReorgHandle(job, reorgInfo.elements); err1 != nil {
logutil.BgLogger().Warn("[ddl] run modify column job failed, RemoveDDLReorgHandle failed, can't convert job to rollback",
zap.String("job", job.String()), zap.Error(err1))
return ver, errors.Trace(err)
}
logutil.BgLogger().Warn("[ddl] run modify column job failed, convert job to rollback", zap.String("job", job.String()), zap.Error(err))
// When encounter these error above, we change the job to rolling back job directly.
job.State = model.JobStateRollingback
return ver, errors.Trace(err)
}
// Clean up the channel of notifyCancelReorgJob. Make sure it can't affect other jobs.
w.reorgCtx.cleanNotifyReorgCancel()
return ver, errors.Trace(err)
}
// Clean up the channel of notifyCancelReorgJob. Make sure it can't affect other jobs.
w.reorgCtx.cleanNotifyReorgCancel()
// Remove the old column and indexes. Update the relative column name and index names.
oldIdxIDs := make([]int64, 0, len(changingIdxs))
tblInfo.Columns = tblInfo.Columns[:len(tblInfo.Columns)-1]
for _, cIdx := range changingIdxs {
idxName := getChangingIndexOriginName(cIdx)
for i, idx := range tblInfo.Indices {
if strings.EqualFold(idxName, idx.Name.O) {
cIdx.Name = model.NewCIStr(idxName)
tblInfo.Indices[i] = cIdx
oldIdxIDs = append(oldIdxIDs, idx.ID)
break
}
}
}
changingColumnUniqueName := changingCol.Name
changingCol.Name = colName
changingCol.ChangeStateInfo = nil
tblInfo.Indices = tblInfo.Indices[:len(tblInfo.Indices)-len(changingIdxs)]
// Adjust table column offset.
if err = adjustColumnInfoInModifyColumn(job, tblInfo, changingCol, oldCol, pos, changingColumnUniqueName.L); err != nil {
// TODO: Do rollback.
return ver, errors.Trace(err)
}
updateChangingInfo(changingCol, changingIdxs, model.StatePublic)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != changingCol.State)
if err != nil {
return ver, errors.Trace(err)
}
// Finish this job.
job.FinishTableJob(model.JobStateDone, model.StatePublic, ver, tblInfo)
// Refactor the job args to add the old index ids into delete range table.
job.Args = []interface{}{oldIdxIDs, getPartitionIDs(tblInfo)}
asyncNotifyEvent(d, &ddlutil.Event{Tp: model.ActionModifyColumn, TableInfo: tblInfo, ColumnInfos: []*model.ColumnInfo{changingCol}})
default:
err = ErrInvalidDDLState.GenWithStackByArgs("column", changingCol.State)
}
return ver, errors.Trace(err)
}
// needRollbackData indicates whether it needs to rollback data when specific error occurs.
func needRollbackData(err error) bool {
return kv.ErrKeyExists.Equal(err) || errCancelledDDLJob.Equal(err) || errCantDecodeRecord.Equal(err) ||
types.ErrOverflow.Equal(err) || types.ErrDataTooLong.Equal(err) || types.ErrTruncated.Equal(err) ||
json.ErrInvalidJSONText.Equal(err) || types.ErrBadNumber.Equal(err)
}
// BuildElements is exported for testing.
func BuildElements(changingCol *model.ColumnInfo, changingIdxs []*model.IndexInfo) []*meta.Element {
elements := make([]*meta.Element, 0, len(changingIdxs)+1)
elements = append(elements, &meta.Element{ID: changingCol.ID, TypeKey: meta.ColumnElementKey})
for _, idx := range changingIdxs {
elements = append(elements, &meta.Element{ID: idx.ID, TypeKey: meta.IndexElementKey})
}
return elements
}
func (w *worker) updatePhysicalTableRow(t table.PhysicalTable, oldColInfo, colInfo *model.ColumnInfo, reorgInfo *reorgInfo) error {
logutil.BgLogger().Info("[ddl] start to update table row", zap.String("job", reorgInfo.Job.String()), zap.String("reorgInfo", reorgInfo.String()))
return w.writePhysicalTableRecord(t.(table.PhysicalTable), typeUpdateColumnWorker, nil, oldColInfo, colInfo, reorgInfo)
}
// updateColumnAndIndexes handles the modify column reorganization state for a table.
func (w *worker) updateColumnAndIndexes(t table.Table, oldCol, col *model.ColumnInfo, idxes []*model.IndexInfo, reorgInfo *reorgInfo) error {
// TODO: Support partition tables.
if bytes.Equal(reorgInfo.currElement.TypeKey, meta.ColumnElementKey) {
err := w.updatePhysicalTableRow(t.(table.PhysicalTable), oldCol, col, reorgInfo)
if err != nil {
return errors.Trace(err)
}
}
// Get the original start handle and end handle.
currentVer, err := getValidCurrentVersion(reorgInfo.d.store)
if err != nil {
return errors.Trace(err)
}
originalStartHandle, originalEndHandle, err := getTableRange(reorgInfo.d, t.(table.PhysicalTable), currentVer.Ver, reorgInfo.Job.Priority)
if err != nil {
return errors.Trace(err)
}
startElementOffset := 0
startElementOffsetToResetHandle := -1
// This backfill job starts with backfilling index data, whose index ID is currElement.ID.
if bytes.Equal(reorgInfo.currElement.TypeKey, meta.IndexElementKey) {
for i, idx := range idxes {
if reorgInfo.currElement.ID == idx.ID {
startElementOffset = i
startElementOffsetToResetHandle = i
break
}
}
}
for i := startElementOffset; i < len(idxes); i++ {
// This backfill job has been exited during processing. At that time, the element is reorgInfo.elements[i+1] and handle range is [reorgInfo.StartHandle, reorgInfo.EndHandle].
// Then the handle range of the rest elements' is [originalStartHandle, originalEndHandle].
if i == startElementOffsetToResetHandle+1 {
reorgInfo.StartHandle, reorgInfo.EndHandle = originalStartHandle, originalEndHandle
}
reorgInfo.currElement = reorgInfo.elements[i+1]
// Write the reorg info to store so the whole reorganize process can recover from panic.
err := reorgInfo.UpdateReorgMeta(reorgInfo.StartHandle)
logutil.BgLogger().Info("[ddl] update column and indexes", zap.Int64("jobID", reorgInfo.Job.ID),
zap.ByteString("elementType", reorgInfo.currElement.TypeKey), zap.Int64("elementID", reorgInfo.currElement.ID),
zap.String("startHandle", toString(reorgInfo.StartHandle)), zap.String("endHandle", toString(reorgInfo.EndHandle)))
if err != nil {
return errors.Trace(err)
}
err = w.addTableIndex(t, idxes[i], reorgInfo)
if err != nil {
return errors.Trace(err)
}
}
return nil
}
type updateColumnWorker struct {
*backfillWorker
oldColInfo *model.ColumnInfo
newColInfo *model.ColumnInfo
metricCounter prometheus.Counter
// The following attributes are used to reduce memory allocation.
rowRecords []*rowRecord
rowDecoder *decoder.RowDecoder
rowMap map[int64]types.Datum
// For SQL Mode and warnings.
sqlMode mysql.SQLMode
}
func newUpdateColumnWorker(sessCtx sessionctx.Context, worker *worker, id int, t table.PhysicalTable, oldCol, newCol *model.ColumnInfo, decodeColMap map[int64]decoder.Column, sqlMode mysql.SQLMode) *updateColumnWorker {
rowDecoder := decoder.NewRowDecoder(t, t.WritableCols(), decodeColMap)
return &updateColumnWorker{
backfillWorker: newBackfillWorker(sessCtx, worker, id, t),
oldColInfo: oldCol,
newColInfo: newCol,
metricCounter: metrics.BackfillTotalCounter.WithLabelValues("update_col_speed"),
rowDecoder: rowDecoder,
rowMap: make(map[int64]types.Datum, len(decodeColMap)),
sqlMode: sqlMode,
}
}
func (w *updateColumnWorker) AddMetricInfo(cnt float64) {
w.metricCounter.Add(cnt)
}
type rowRecord struct {
key []byte // It's used to lock a record. Record it to reduce the encoding time.
vals []byte // It's the record.
warning *terror.Error // It's used to record the cast warning of a record.
}
// getNextHandle gets next handle of entry that we are going to process.
func (w *updateColumnWorker) getNextHandle(taskRange reorgBackfillTask, taskDone bool, lastAccessedHandle kv.Handle) (nextHandle kv.Handle) {
if !taskDone {
// The task is not done. So we need to pick the last processed entry's handle and add one.
return lastAccessedHandle.Next()
}
// The task is done. So we need to choose a handle outside this range.
// Some corner cases should be considered:
// - The end of task range is MaxInt64.
// - The end of the task is excluded in the range.
if (taskRange.endHandle.IsInt() && taskRange.endHandle.IntValue() == math.MaxInt64) || !taskRange.endIncluded {
return taskRange.endHandle
}
return taskRange.endHandle.Next()
}
func (w *updateColumnWorker) fetchRowColVals(txn kv.Transaction, taskRange reorgBackfillTask) ([]*rowRecord, kv.Handle, bool, error) {
w.rowRecords = w.rowRecords[:0]
startTime := time.Now()
// taskDone means that the added handle is out of taskRange.endHandle.
taskDone := false
var lastAccessedHandle kv.Handle
oprStartTime := startTime
err := iterateSnapshotRows(w.sessCtx.GetStore(), w.priority, w.table, txn.StartTS(), taskRange.startHandle, taskRange.endHandle, taskRange.endIncluded,
func(handle kv.Handle, recordKey kv.Key, rawRow []byte) (bool, error) {
oprEndTime := time.Now()
logSlowOperations(oprEndTime.Sub(oprStartTime), "iterateSnapshotRows in updateColumnWorker fetchRowColVals", 0)
oprStartTime = oprEndTime
if !taskRange.endIncluded {
taskDone = handle.Compare(taskRange.endHandle) >= 0
} else {
taskDone = handle.Compare(taskRange.endHandle) > 0
}
if taskDone || len(w.rowRecords) >= w.batchCnt {
return false, nil
}
if err1 := w.getRowRecord(handle, recordKey, rawRow); err1 != nil {
return false, errors.Trace(err1)
}
lastAccessedHandle = handle
if handle.Equal(taskRange.endHandle) {
// If taskRange.endIncluded == false, we will not reach here when handle == taskRange.endHandle.
taskDone = true
return false, nil
}
return true, nil
})
if len(w.rowRecords) == 0 {
taskDone = true
}
logutil.BgLogger().Debug("[ddl] txn fetches handle info", zap.Uint64("txnStartTS", txn.StartTS()), zap.String("taskRange", taskRange.String()), zap.Duration("takeTime", time.Since(startTime)))
return w.rowRecords, w.getNextHandle(taskRange, taskDone, lastAccessedHandle), taskDone, errors.Trace(err)
}
func (w *updateColumnWorker) getRowRecord(handle kv.Handle, recordKey []byte, rawRow []byte) error {
_, err := w.rowDecoder.DecodeAndEvalRowWithMap(w.sessCtx, handle, rawRow, time.UTC, timeutil.SystemLocation(), w.rowMap)
if err != nil {
return errors.Trace(errCantDecodeRecord.GenWithStackByArgs("column", err))
}
if _, ok := w.rowMap[w.newColInfo.ID]; ok {
// The column is already added by update or insert statement, skip it.
w.cleanRowMap()
return nil
}
var recordWarning *terror.Error
newColVal, err := table.CastValue(w.sessCtx, w.rowMap[w.oldColInfo.ID], w.newColInfo, false, false)
if err != nil {
err = w.reformatErrors(err)
if IsNormalWarning(err) || (!w.sqlMode.HasStrictMode() && IsStrictWarning(err)) {
// Keep the warnings.
recordWarning = errors.Cause(err).(*terror.Error)
} else {
return errors.Trace(err)
}
}
failpoint.Inject("MockReorgTimeoutInOneRegion", func(val failpoint.Value) {
if val.(bool) {
if handle.IntValue() == 3000 && atomic.CompareAndSwapInt32(&TestCheckReorgTimeout, 0, 1) {
failpoint.Return(errors.Trace(errWaitReorgTimeout))
}
}
})
w.rowMap[w.newColInfo.ID] = newColVal
newColumnIDs := make([]int64, 0, len(w.rowMap))
newRow := make([]types.Datum, 0, len(w.rowMap))
for colID, val := range w.rowMap {
newColumnIDs = append(newColumnIDs, colID)
newRow = append(newRow, val)
}
sctx, rd := w.sessCtx.GetSessionVars().StmtCtx, &w.sessCtx.GetSessionVars().RowEncoder
newRowVal, err := tablecodec.EncodeRow(sctx, newRow, newColumnIDs, nil, nil, rd)
if err != nil {
return errors.Trace(err)
}
w.rowRecords = append(w.rowRecords, &rowRecord{key: recordKey, vals: newRowVal, warning: recordWarning})
w.cleanRowMap()
return nil
}
// reformatErrors casted error because `convertTo` function couldn't package column name and datum value for some errors.
func (w *updateColumnWorker) reformatErrors(err error) error {
// Since row count is not precious in concurrent reorganization, here we substitute row count with datum value.
if types.ErrTruncated.Equal(err) {
err = types.ErrTruncated.GenWithStack("Data truncated for column '%s', value is '%s'", w.oldColInfo.Name, w.rowMap[w.oldColInfo.ID])
}
return err
}
// IsNormalWarning is used to check the normal warnings, for example data-truncated warnings.
// This kind of warning will be always thrown out regard less of what kind of the sql mode is.
func IsNormalWarning(err error) bool {
// TODO: there are more errors here can be identified as normal warnings.
if types.ErrTruncatedWrongVal.Equal(err) {
return true
}
return false
}
// IsStrictWarning is used to check whether the error can be transferred as a warning under a
// non-strict SQL Mode.
func IsStrictWarning(err error) bool {
// TODO: there are more errors here can be identified as warnings under non-strict SQL mode.
if types.ErrOverflow.Equal(err) || types.ErrTruncated.Equal(err) {
return true
}
return false
}
func (w *updateColumnWorker) cleanRowMap() {
for id := range w.rowMap {
delete(w.rowMap, id)
}
}
// BackfillDataInTxn will backfill the table record in a transaction, lock corresponding rowKey, if the value of rowKey is changed.
func (w *updateColumnWorker) BackfillDataInTxn(handleRange reorgBackfillTask) (taskCtx backfillTaskContext, errInTxn error) {
oprStartTime := time.Now()
errInTxn = kv.RunInNewTxn(w.sessCtx.GetStore(), true, func(txn kv.Transaction) error {
taskCtx.addedCount = 0
taskCtx.scanCount = 0
txn.SetOption(kv.Priority, w.priority)
rowRecords, nextHandle, taskDone, err := w.fetchRowColVals(txn, handleRange)
if err != nil {
return errors.Trace(err)
}
taskCtx.nextHandle = nextHandle
taskCtx.done = taskDone
warningsMap := make(map[errors.ErrorID]*terror.Error, len(rowRecords))
warningsCountMap := make(map[errors.ErrorID]int64, len(rowRecords))
for _, rowRecord := range rowRecords {
taskCtx.scanCount++
err = txn.Set(rowRecord.key, rowRecord.vals)
if err != nil {
return errors.Trace(err)
}
taskCtx.addedCount++
if rowRecord.warning != nil {
if _, ok := warningsCountMap[rowRecord.warning.ID()]; ok {
warningsCountMap[rowRecord.warning.ID()]++
} else {
warningsCountMap[rowRecord.warning.ID()] = 1
warningsMap[rowRecord.warning.ID()] = rowRecord.warning
}
}
}
// Collect the warnings.
taskCtx.warnings, taskCtx.warningsCount = warningsMap, warningsCountMap
return nil
})
logSlowOperations(time.Since(oprStartTime), "BackfillDataInTxn", 3000)
return
}
func updateChangingInfo(changingCol *model.ColumnInfo, changingIdxs []*model.IndexInfo, schemaState model.SchemaState) {
changingCol.State = schemaState
for _, idx := range changingIdxs {
idx.State = schemaState
}
}
// doModifyColumn updates the column information and reorders all columns. It does not support modifying column data.
func (w *worker) doModifyColumn(
t *meta.Meta, job *model.Job, dbInfo *model.DBInfo, tblInfo *model.TableInfo,
newCol, oldCol *model.ColumnInfo, pos *ast.ColumnPosition) (ver int64, _ error) {
// Column from null to not null.
if !mysql.HasNotNullFlag(oldCol.Flag) && mysql.HasNotNullFlag(newCol.Flag) {
noPreventNullFlag := !mysql.HasPreventNullInsertFlag(oldCol.Flag)
// Introduce the `mysql.PreventNullInsertFlag` flag to prevent users from inserting or updating null values.
err := modifyColsFromNull2NotNull(w, dbInfo, tblInfo, []*model.ColumnInfo{oldCol}, newCol.Name, oldCol.Tp != newCol.Tp)
if err != nil {
if ErrWarnDataTruncated.Equal(err) || errInvalidUseOfNull.Equal(err) {
job.State = model.JobStateRollingback
}
return ver, err
}
// The column should get into prevent null status first.
if noPreventNullFlag {
return updateVersionAndTableInfoWithCheck(t, job, tblInfo, true)
}
}
if err := adjustColumnInfoInModifyColumn(job, tblInfo, newCol, oldCol, pos, ""); err != nil {
return ver, errors.Trace(err)
}
ver, err := updateVersionAndTableInfoWithCheck(t, job, tblInfo, true)
if err != nil {
// Modified the type definition of 'null' to 'not null' before this, so rollBack the job when an error occurs.
job.State = model.JobStateRollingback
return ver, errors.Trace(err)
}
job.FinishTableJob(model.JobStateDone, model.StatePublic, ver, tblInfo)
// For those column-type-change type which doesn't need reorg data, we should also mock the job args for delete range.
job.Args = []interface{}{[]int64{}, []int64{}}
return ver, nil
}
func adjustColumnInfoInModifyColumn(
job *model.Job, tblInfo *model.TableInfo, newCol, oldCol *model.ColumnInfo, pos *ast.ColumnPosition, changingColUniqueLowerName string) error {
// We need the latest column's offset and state. This information can be obtained from the store.
newCol.Offset = oldCol.Offset
newCol.State = oldCol.State
// Calculate column's new position.
oldPos, newPos := oldCol.Offset, oldCol.Offset
if pos.Tp == ast.ColumnPositionAfter {
// TODO: The check of "RelativeColumn" can be checked in advance. When "EnableChangeColumnType" is true, unnecessary state changes can be reduced.
if oldCol.Name.L == pos.RelativeColumn.Name.L {
// `alter table tableName modify column b int after b` will return ver,ErrColumnNotExists.
// Modified the type definition of 'null' to 'not null' before this, so rollback the job when an error occurs.
job.State = model.JobStateRollingback
return infoschema.ErrColumnNotExists.GenWithStackByArgs(oldCol.Name, tblInfo.Name)
}
relative := model.FindColumnInfo(tblInfo.Columns, pos.RelativeColumn.Name.L)
if relative == nil || relative.State != model.StatePublic {
job.State = model.JobStateRollingback
return infoschema.ErrColumnNotExists.GenWithStackByArgs(pos.RelativeColumn, tblInfo.Name)
}
if relative.Offset < oldPos {
newPos = relative.Offset + 1
} else {
newPos = relative.Offset
}
} else if pos.Tp == ast.ColumnPositionFirst {
newPos = 0
}
columnChanged := make(map[string]*model.ColumnInfo)
columnChanged[oldCol.Name.L] = newCol
if newPos == oldPos {
tblInfo.Columns[newPos] = newCol
} else {
cols := tblInfo.Columns
// Reorder columns in place.
if newPos < oldPos {
// ******** +(new) ****** -(old) ********
// [newPos:old-1] should shift one step to the right.
copy(cols[newPos+1:], cols[newPos:oldPos])
} else {
// ******** -(old) ****** +(new) ********
// [old+1:newPos] should shift one step to the left.
copy(cols[oldPos:], cols[oldPos+1:newPos+1])
}
cols[newPos] = newCol
for i, col := range tblInfo.Columns {
if col.Offset != i {
columnChanged[col.Name.L] = col
col.Offset = i
}
}
}
// Change offset and name in indices.
for _, idx := range tblInfo.Indices {
for _, c := range idx.Columns {
cName := c.Name.L
// With the unique changing column name, the format is designed as `_Col$_xx_uniqueNum`.
// The suffix `_uniqueNum` will be trimmed when we get the origin changing column name.
// There is a possibility that some other column named as `xx_xx_xx`, and it will be get
// trimmed as `xx_xx`. So here we check here and only do the trim for the changing index column.
if len(changingColUniqueLowerName) != 0 && c.Name.L == changingColUniqueLowerName {
cName = strings.ToLower(getChangingColumnOriginName(c))
}
if newCol, ok := columnChanged[cName]; ok {
c.Name = newCol.Name
c.Offset = newCol.Offset
}
}
}
return nil
}
func checkAndApplyNewAutoRandomBits(job *model.Job, t *meta.Meta, tblInfo *model.TableInfo,
newCol *model.ColumnInfo, oldName *model.CIStr, newAutoRandBits uint64) error {
schemaID := job.SchemaID
newLayout := autoid.NewAutoRandomIDLayout(&newCol.FieldType, newAutoRandBits)
// GenAutoRandomID first to prevent concurrent update.
_, err := t.GenAutoRandomID(schemaID, tblInfo.ID, 1)
if err != nil {
return err
}
currentIncBitsVal, err := t.GetAutoRandomID(schemaID, tblInfo.ID)
if err != nil {
return err
}
// Find the max number of available shard bits by
// counting leading zeros in current inc part of auto_random ID.
availableBits := bits.LeadingZeros64(uint64(currentIncBitsVal))
isOccupyingIncBits := newLayout.TypeBitsLength-newLayout.IncrementalBits > uint64(availableBits)
if isOccupyingIncBits {
availableBits := mathutil.Min(autoid.MaxAutoRandomBits, availableBits)
errMsg := fmt.Sprintf(autoid.AutoRandomOverflowErrMsg, availableBits, newAutoRandBits, oldName.O)
job.State = model.JobStateCancelled
return ErrInvalidAutoRandom.GenWithStackByArgs(errMsg)
}
tblInfo.AutoRandomBits = newAutoRandBits
return nil
}
// checkForNullValue ensure there are no null values of the column of this table.
// `isDataTruncated` indicates whether the new field and the old field type are the same, in order to be compatible with mysql.
func checkForNullValue(ctx sessionctx.Context, isDataTruncated bool, schema, table, newCol model.CIStr, oldCols ...*model.ColumnInfo) error {
colsStr := ""
for i, col := range oldCols {
if i == 0 {
colsStr += "`" + col.Name.L + "` is null"
} else {
colsStr += " or `" + col.Name.L + "` is null"
}
}
sql := fmt.Sprintf("select 1 from `%s`.`%s` where %s limit 1;", schema.L, table.L, colsStr)
rows, _, err := ctx.(sqlexec.RestrictedSQLExecutor).ExecRestrictedSQL(sql)
if err != nil {
return errors.Trace(err)
}
rowCount := len(rows)
if rowCount != 0 {
if isDataTruncated {
return ErrWarnDataTruncated.GenWithStackByArgs(newCol.L, rowCount)
}
return errInvalidUseOfNull
}
return nil
}
func updateColumnDefaultValue(t *meta.Meta, job *model.Job, newCol *model.ColumnInfo, oldColName *model.CIStr) (ver int64, _ error) {
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, job.SchemaID)
if err != nil {
return ver, errors.Trace(err)
}
oldCol := model.FindColumnInfo(tblInfo.Columns, oldColName.L)
if oldCol == nil || oldCol.State != model.StatePublic {
job.State = model.JobStateCancelled
return ver, infoschema.ErrColumnNotExists.GenWithStackByArgs(newCol.Name, tblInfo.Name)
}
// The newCol's offset may be the value of the old schema version, so we can't use newCol directly.
oldCol.DefaultValue = newCol.DefaultValue
oldCol.DefaultValueBit = newCol.DefaultValueBit
oldCol.Flag = newCol.Flag
ver, err = updateVersionAndTableInfo(t, job, tblInfo, true)
if err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
job.FinishTableJob(model.JobStateDone, model.StatePublic, ver, tblInfo)
return ver, nil
}
func isColumnWithIndex(colName string, indices []*model.IndexInfo) bool {
for _, indexInfo := range indices {
for _, col := range indexInfo.Columns {
if col.Name.L == colName {
return true
}
}
}
return false
}
func isColumnCanDropWithIndex(colName string, indices []*model.IndexInfo) bool {
for _, indexInfo := range indices {
if indexInfo.Primary || len(indexInfo.Columns) > 1 {
for _, col := range indexInfo.Columns {
if col.Name.L == colName {
return false
}
}
}
}
return true
}
func listIndicesWithColumn(colName string, indices []*model.IndexInfo) []*model.IndexInfo {
ret := make([]*model.IndexInfo, 0)
for _, indexInfo := range indices {
if len(indexInfo.Columns) == 1 && colName == indexInfo.Columns[0].Name.L {
ret = append(ret, indexInfo)
}
}
return ret
}
func getColumnForeignKeyInfo(colName string, fkInfos []*model.FKInfo) *model.FKInfo {
for _, fkInfo := range fkInfos {
for _, col := range fkInfo.Cols {
if col.L == colName {
return fkInfo
}
}
}
return nil
}
func allocateColumnID(tblInfo *model.TableInfo) int64 {
tblInfo.MaxColumnID++
return tblInfo.MaxColumnID
}
func checkAddColumnTooManyColumns(colNum int) error {
if uint32(colNum) > atomic.LoadUint32(&TableColumnCountLimit) {
return errTooManyFields
}
return nil
}
// rollbackModifyColumnJob rollbacks the job when an error occurs.
func rollbackModifyColumnJob(t *meta.Meta, tblInfo *model.TableInfo, job *model.Job, oldCol *model.ColumnInfo, modifyColumnTp byte) (ver int64, _ error) {
var err error
if modifyColumnTp == mysql.TypeNull {
// field NotNullFlag flag reset.
tblInfo.Columns[oldCol.Offset].Flag = oldCol.Flag &^ mysql.NotNullFlag
// field PreventNullInsertFlag flag reset.
tblInfo.Columns[oldCol.Offset].Flag = oldCol.Flag &^ mysql.PreventNullInsertFlag
ver, err = updateVersionAndTableInfo(t, job, tblInfo, true)
if err != nil {
return ver, errors.Trace(err)
}
}
job.FinishTableJob(model.JobStateRollbackDone, model.StateNone, ver, tblInfo)
// For those column-type-change type which doesn't need reorg data, we should also mock the job args for delete range.
job.Args = []interface{}{[]int64{}, []int64{}}
return ver, nil
}
// modifyColsFromNull2NotNull modifies the type definitions of 'null' to 'not null'.
// Introduce the `mysql.PreventNullInsertFlag` flag to prevent users from inserting or updating null values.
func modifyColsFromNull2NotNull(w *worker, dbInfo *model.DBInfo, tblInfo *model.TableInfo, cols []*model.ColumnInfo,
newColName model.CIStr, isModifiedType bool) error {
// Get sessionctx from context resource pool.
var ctx sessionctx.Context
ctx, err := w.sessPool.get()
if err != nil {
return errors.Trace(err)
}
defer w.sessPool.put(ctx)
skipCheck := false
failpoint.Inject("skipMockContextDoExec", func(val failpoint.Value) {
if val.(bool) {
skipCheck = true
}
})
if !skipCheck {
// If there is a null value inserted, it cannot be modified and needs to be rollback.
err = checkForNullValue(ctx, isModifiedType, dbInfo.Name, tblInfo.Name, newColName, cols...)
if err != nil {
return errors.Trace(err)
}
}
// Prevent this field from inserting null values.
for _, col := range cols {
col.Flag |= mysql.PreventNullInsertFlag
}
return nil
}
func generateOriginDefaultValue(col *model.ColumnInfo) (interface{}, error) {
var err error
odValue := col.GetDefaultValue()
if odValue == nil && mysql.HasNotNullFlag(col.Flag) {
zeroVal := table.GetZeroValue(col)
odValue, err = zeroVal.ToString()
if err != nil {
return nil, errors.Trace(err)
}
}
if odValue == strings.ToUpper(ast.CurrentTimestamp) {
if col.Tp == mysql.TypeTimestamp {
odValue = time.Now().UTC().Format(types.TimeFormat)
} else if col.Tp == mysql.TypeDatetime {
odValue = time.Now().Format(types.TimeFormat)
}
}
return odValue, nil
}
func findColumnInIndexCols(c string, cols []*model.IndexColumn) *model.IndexColumn {
for _, c1 := range cols {
if c == c1.Name.L {
return c1
}
}
return nil
}
func getColumnInfoByName(tbInfo *model.TableInfo, column string) *model.ColumnInfo {
for _, colInfo := range tbInfo.Cols() {
if colInfo.Name.L == column {
return colInfo
}
}
return nil
}
// isVirtualGeneratedColumn checks the column if it is virtual.
func isVirtualGeneratedColumn(col *model.ColumnInfo) bool {
if col.IsGenerated() && !col.GeneratedStored {
return true
}
return false
}
func indexInfoContains(idxID int64, idxInfos []*model.IndexInfo) bool {
for _, idxInfo := range idxInfos {
if idxID == idxInfo.ID {
return true
}
}
return false
}
func indexInfosToIDList(idxInfos []*model.IndexInfo) []int64 {
ids := make([]int64, 0, len(idxInfos))
for _, idxInfo := range idxInfos {
ids = append(ids, idxInfo.ID)
}
return ids
}
func genChangingColumnUniqueName(tblInfo *model.TableInfo, oldCol *model.ColumnInfo) string {
suffix := 0
newColumnNamePrefix := fmt.Sprintf("%s%s", changingColumnPrefix, oldCol.Name.O)
newColumnLowerName := fmt.Sprintf("%s_%d", strings.ToLower(newColumnNamePrefix), suffix)
// Check whether the new column name is used.
columnNameMap := make(map[string]bool, len(tblInfo.Columns))
for _, col := range tblInfo.Columns {
columnNameMap[col.Name.L] = true
}
for columnNameMap[newColumnLowerName] {
suffix++
newColumnLowerName = fmt.Sprintf("%s_%d", strings.ToLower(newColumnNamePrefix), suffix)
}
return fmt.Sprintf("%s_%d", newColumnNamePrefix, suffix)
}
func genChangingIndexUniqueName(tblInfo *model.TableInfo, idxInfo *model.IndexInfo) string {
suffix := 0
newIndexNamePrefix := fmt.Sprintf("%s%s", changingIndexPrefix, idxInfo.Name.O)
newIndexLowerName := fmt.Sprintf("%s_%d", strings.ToLower(newIndexNamePrefix), suffix)
// Check whether the new index name is used.
indexNameMap := make(map[string]bool, len(tblInfo.Indices))
for _, idx := range tblInfo.Indices {
indexNameMap[idx.Name.L] = true
}
for indexNameMap[newIndexLowerName] {
suffix++
newIndexLowerName = fmt.Sprintf("%s_%d", strings.ToLower(newIndexNamePrefix), suffix)
}
return fmt.Sprintf("%s_%d", newIndexNamePrefix, suffix)
}
func getChangingColumnOriginName(changingCol *model.IndexColumn) string {
colName := strings.ToLower(strings.TrimPrefix(changingCol.Name.O, changingColumnPrefix))
// Since the unique colName may contain the suffix number (columnName_num), better trim the suffix.
var pos int
if pos = strings.LastIndex(colName, "_"); pos == -1 {
return colName
}
return colName[:pos]
}
func getChangingIndexOriginName(changingIdx *model.IndexInfo) string {
idxName := strings.TrimPrefix(changingIdx.Name.O, changingIndexPrefix)
// Since the unique idxName may contain the suffix number (indexName_num), better trim the suffix.
var pos int
if pos = strings.LastIndex(idxName, "_"); pos == -1 {
return idxName
}
return idxName[:pos]
}