ddl: Corrected index management during REORGANIZE PARTITION #56786
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -2185,9 +2185,7 @@ func (w *worker) rollbackLikeDropPartition(jobCtx *jobContext, job *model.Job) ( | |
|
|
||
| var dropIndices []*model.IndexInfo | ||
| for _, indexInfo := range tblInfo.Indices { | ||
| if indexInfo.State == model.StateWriteOnly { | ||
| dropIndices = append(dropIndices, indexInfo) | ||
| } | ||
| } | ||
|
|
@@ -3043,9 +3041,6 @@ func (w *worker) onExchangeTablePartition(jobCtx *jobContext, job *model.Job) (v | |
| } | ||
|
|
||
| func getNewGlobal(partInfo *model.PartitionInfo, idx *model.IndexInfo) bool { | ||
| for _, newIdx := range partInfo.DDLUpdateIndexes { | ||
| if strings.EqualFold(idx.Name.L, newIdx.IndexName) { | ||
| return newIdx.Global | ||
|
|
@@ -3151,6 +3146,9 @@ func getReorgPartitionInfo(t *meta.Mutator, job *model.Job, args *model.TablePar | |
| // | ||
| // Everything now looks as it should, no memory of old partitions/indexes, | ||
| // and no more double writing, since the previous state is only reading the new partitions/indexes. | ||
| // | ||
| // Note: Special handling is also required in tables.newPartitionedTable(), | ||
| // to get per partition indexes in the right state. | ||
| func (w *worker) onReorganizePartition(jobCtx *jobContext, job *model.Job) (ver int64, _ error) { | ||
| args, err := model.GetTablePartitionArgs(job) | ||
| if err != nil { | ||
|
|
@@ -3262,39 +3260,33 @@ func (w *worker) onReorganizePartition(jobCtx *jobContext, job *model.Job) (ver | |
| if err != nil { | ||
| return ver, errors.Trace(err) | ||
| } | ||
| // Currently only support Explicit Global indexes. | ||
| if !inAllPartitionColumns && !newGlobal { | ||
| job.State = model.JobStateCancelled | ||
| return ver, dbterror.ErrGlobalIndexNotExplicitlySet.GenWithStackByArgs(index.Name.O) | ||
| } | ||
| if !index.Global && !newGlobal { | ||
| // still local index, no need to duplicate index. | ||
| continue | ||
| } | ||
| if tblInfo.Partition.DDLChangedIndex == nil { | ||
| tblInfo.Partition.DDLChangedIndex = make(map[int64]bool) | ||
| } | ||
| // Duplicate the unique indexes with new index ids. | ||
| // If previously was Global or will be Global: | ||
| // it must be recreated with new index ID | ||
| // TODO: Could we allow that session in StateWriteReorganization, when StateDeleteReorganization | ||
| // has started, may not find changes through the global index that sessions in StateDeleteReorganization made? | ||
| // If so, then we could avoid copying the full Global Index if it has not changed from LOCAL! | ||
| // It might be possible to use the new, not yet public partitions to access those rows?! | ||
| // Just that it would not work with explicit partition select SELECT FROM t PARTITION (p,...) | ||
| newIndex := index.Clone() | ||
| newIndex.State = model.StateDeleteOnly | ||
| newIndex.ID = AllocateIndexID(tblInfo) | ||
| tblInfo.Partition.DDLChangedIndex[index.ID] = false | ||
| tblInfo.Partition.DDLChangedIndex[newIndex.ID] = true | ||
| newIndex.Global = newGlobal | ||
| tblInfo.Indices = append(tblInfo.Indices, newIndex) | ||
| } | ||
| failpoint.Inject("reorgPartCancel1", func(val failpoint.Value) { | ||
| if val.(bool) { | ||
|
|
@@ -3487,26 +3479,18 @@ func (w *worker) onReorganizePartition(jobCtx *jobContext, job *model.Job) (ver | |
| if !index.Unique { | ||
Defined2014 marked this conversation as resolved.
Show resolved
Hide resolved
|
||
| continue | ||
| } | ||
| isNew, ok := tblInfo.Partition.DDLChangedIndex[index.ID] | ||
| if !ok { | ||
| continue | ||
| } | ||
| if isNew { | ||
| // Newly created index, replacing old unique/global index | ||
| index.State = model.StatePublic | ||
| continue | ||
| } | ||
| // Old index, should not be visible any longer, | ||
| // but needs to be kept up-to-date in case rollback happens. | ||
| index.State = model.StateWriteOnly | ||
| } | ||
| firstPartIdx, lastPartIdx, idMap, err2 := getReplacedPartitionIDs(partNames, tblInfo.Partition) | ||
| if err2 != nil { | ||
|
|
@@ -3563,14 +3547,18 @@ func (w *worker) onReorganizePartition(jobCtx *jobContext, job *model.Job) (ver | |
|
|
||
| var dropIndices []*model.IndexInfo | ||
| for _, indexInfo := range tblInfo.Indices { | ||
| if indexInfo.Unique && indexInfo.State == model.StateWriteOnly { | ||
| // Drop the old unique (possible global) index, see onDropIndex | ||
| indexInfo.State = model.StateNone | ||
| DropIndexColumnFlag(tblInfo, indexInfo) | ||
| RemoveDependentHiddenColumns(tblInfo, indexInfo) | ||
| dropIndices = append(dropIndices, indexInfo) | ||
| } | ||
| } | ||
| // TODO: verify that the indexes are dropped, | ||
| // and that StateDeleteOnly+StateDeleteReorganization is not needed. | ||
|
There was a problem hiding this comment. Tests shows that at least the old partitions can be written (double write in case we need to rollback), so we need to add a new state to make sure all clients only use (read and write) the new partitions and new indexes, this should not be fixed in this PR, but a new PR for fixing #56819. |
||
| // local indexes is not an issue, since they will be gone with the dropped | ||
| // partitions, but replaced global indexes should be checked! | ||
| for _, indexInfo := range dropIndices { | ||
| removeIndexInfo(tblInfo, indexInfo) | ||
| } | ||
|
|
@@ -3632,6 +3620,9 @@ func (w *worker) onReorganizePartition(jobCtx *jobContext, job *model.Job) (ver | |
| failpoint.Return(ver, errors.New("Injected error by reorgPartFail5")) | ||
| } | ||
| }) | ||
| failpoint.Inject("updateVersionAndTableInfoErrInStateDeleteReorganization", func() { | ||
| failpoint.Return(ver, errors.New("Injected error in StateDeleteReorganization")) | ||
| }) | ||
| args.OldPhysicalTblIDs = physicalTableIDs | ||
| args.NewPartitionIDs = newIDs | ||
| ver, err = updateVersionAndTableInfo(jobCtx, job, tblInfo, true) | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why only drop the write-only indexes? What about other states? For example, what will happen when we rollback from the "none -> delete-only" step?
There may be
delete-onlyindexes before converting to rolling back job:There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There should only be either
StatePublicorStateWriteOnlyso only write-only indexes that should be removed. Also in the follow ups #57114 and #56974 (based on this PR) it will block the rollback here and here for the last state. One can see this PR a part 1 of 3, where the optimistic path works, then the failure/rollback/cleanup is fixed in part 2 and 3.