-
Notifications
You must be signed in to change notification settings - Fork 134
Open
Labels
performanceImproves performance of existing featuresImproves performance of existing features
Milestone
Description
Currently, if the node which handled the control connection goes down, the driver will walk over the list of remaining nodes, in random order, sequentially trying to re-restablish the control connection until it manages to do it. However, this the random order does not consider DCs at all. While the latency of requests on control connection isn't too important, it would be nice if the schema/topology information didn't have to involve cross-DC traffic.
While we don't have a concept of a local DC in the Session (yet?), we can approximate it by assuming that the contact nodes are from the local DC. After establishing the first control connection, we should always prefer nodes from the DC of the first node.
Metadata
Metadata
Assignees
Labels
performanceImproves performance of existing featuresImproves performance of existing features