indexMerge with limit may cause tidb server OOM

[wjhuang2016]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

(pprof) list NewHandleHeap
Total: 19.90GB
ROUTINE ======================== github.com/pingcap/tidb/pkg/executor.(*indexMergeProcessWorker).NewHandleHeap in /Users/bba/pc/tidb/pkg/executor/index_merge_reader.go
   19.33GB    19.33GB (flat, cum) 97.13% of Total
         .          .   1001:func (w *indexMergeProcessWorker) NewHandleHeap(taskMap map[int][]*indexMergeTableTask, memTracker *memory.Tracker) *handleHeap {
         .          .   1002:	compareFuncs := make([]chunk.CompareFunc, 0, len(w.indexMerge.byItems))
         .          .   1003:	for _, item := range w.indexMerge.byItems {
         .          .   1004:		keyType := item.Expr.GetType()
         .          .   1005:		compareFuncs = append(compareFuncs, chunk.GetCompareFunc(keyType))
         .          .   1006:	}
         .          .   1007:
         .          .   1008:	requiredCnt := uint64(0)
         .          .   1009:	if w.indexMerge.pushedLimit != nil {
         .          .   1010:		requiredCnt = max(requiredCnt, w.indexMerge.pushedLimit.Count+w.indexMerge.pushedLimit.Offset)
         .          .   1011:	}
         .          .   1012:	return &handleHeap{
         .          .   1013:		requiredCnt: requiredCnt,
         .          .   1014:		tracker:     memTracker,
         .          .   1015:		taskMap:     taskMap,
   19.33GB    19.33GB   1016:		idx:         make([]rowIdx, 0, requiredCnt),
         .          .   1017:		compareFunc: compareFuncs,
         .          .   1018:		byItems:     w.indexMerge.byItems,
         .          .   1019:	}
         .          .   1020:}
         .          .   1021:

2. What did you expect to see? (Required)

No OOM

3. What did you see instead (Required)

19.33GB 19.33GB 1016: idx: make([]rowIdx, 0, requiredCnt),

4. What is your TiDB version? (Required)

1fc92b3

[Defined2014]

This only happens if the limit number is very large. It would also OOM if we had that many rows. So the problem only occurs when limit number is very large and the number of rows are much less. Change its serverity to moderate rather than major.