As per my understanding, clients do not have a server-preference for sending read queries and do either a round-robin or a random client selection.
If so, then each server acts as a coordinator node (one that scatter-gathers).
And if any one node is slow, then the queries on that slow node have to be slow (About 1/N queries where N is the total number of data nodes in the cluster).
Is there any way to prevent such a thing?
For example, Cassandra has speculative retries to overcome the same.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.