Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
Description
Presently, the BatchReader uses a thread-pool to make synchronous calls to tablet servers to fetch data. That limits the number of tablet servers that can be doing queries in parallel. On very large clusters, clients may not initiate queries to all of the servers until some queries finish. It is not practical to create thousands of threads just to initiate queries.
Alternatively, asynchronous calls can efficiently initiate queries an wait for responses, without using threads just to wait on responses.