Description
Quoting from the scan token design doc:
Eventually, the scan token API should allow applications to further split scan tokens so that inter-tablet parallelism can be acheived. Splitting tokens may be achieved by assigning the child tokens non-overlapping sections of the primary key range. Even without the token splitting feature built in to the API, applications can simulate the effect by building multiple sets of scan tokens using non-overlapping sets of primary key bounds. However, it is likely that in the future Kudu will be able to choose a more optimal primary key split point than the application, perhaps through an internal tablet statistics API. Additionally, having the API built in to the Kudu client further decreases the effort required to write high performance integrations for Kudu.
Attachments
Issue Links
- duplicates
-
KUDU-2437 Split a tablet into primary key ranges by size
- Resolved