The tablet size determines when to split the tablet. By default, the tablet size is 4GB, so when the tablet is larger than 4GB, it’s split in into smaller tablets.
One feature of MapR-DB is to aggressively perform splits in the beginning. If the table is created without any splits (that is, only one tablet), it is split into 4 tablets as the table grows beyond 256MB (which has some sample of row keys). This behavior avoids hot spotting to some extent by getting 4 tablets quickly. Subsequently, the tablets split at 4GB.
A small nit.
It depends on what you are doing and your use case.
If your row key is always increasing, you will always be adding to the right side. (Going left to right.)
So when you split, roughly 1/2 of the data goes to tablet A, the other half goes to tablet B. Since the table has a clustered index on its key, no new rows will be added to tablet A. Instead they will be added to tablet B until it splits and again, you get 1/2 of a tablet filled while you keep adding to the other.
Note: This may or may not cause hot spotting. It depends on the velocity in which data is being added to the table and the volume of data. Also there are some use cases where you do not want to avoid hot spotting...
Retrieving data ...