Find this when implementing
HBASE-21717. In some UT we put several good requests and bad requests together, and expect only the bad ones to fail. This usually depends on the grouping at rs side, if we group the good one and the bad one together as a batch, it will fail them all. So usually in test we will insert an increment or append in the middle to break them into two groups when executing at RS side.
So if we do not maintain the order, at the rs side, the increment or append may comes first or last, then the good ones and bad ones will be grouped and cause all of them to fail.