Details
-
Improvement
-
Status: Resolved
-
Low
-
Resolution: Fixed
-
None
-
None
Description
deliverAllHints seems to be a "just in case we miss the onAlive signal to send hints over," but the constant scanning of the hint rows it generates is extremely painful.
I'd like to do 3 things here:
1. test that, if node X and Y are in the cluster, and Y goes down, then X goes down, then Y comes back up, then X comes back up [with enough time for each change that FD recognizes it], that X gets deliverHintsToEndpoint called for Y. I think it will, but we should test to be sure.
2. remove HHOM.deliverAllHints
3. expose HHOM.deliverHintsToEndpoint to jmx so it can be manually invoked, Just In Case