Details
-
Improvement
-
Resolution: Fixed
-
Major
-
2.0-beta
-
Security Level: Public
-
None
Description
We'd like 2 optimizations to make reads faster and more efficient:
1. Stored the update seq id of every update inside the item rowId. This should be a very simple task.
2. Batch reads together. As background fetches get backed-up, there is an opportunity to batch the reads from a single vbucket into one bulk read request. This will require some restructuring of the background fetch code to batch them together, ensure that reads are scheduled fairly (perhaps try to batch by vbucket in a loop) so we don't make read requests wait a very long time.
This task has a dependency on MB-5163, which are the couchstore modifications necessary to make this happen.