The coordinator will refuse to send writes/hints to a node if it has a large backlog of hints (128 * #cores) already and the destination replica is one of the nodes with hints destined to it. It will still send writes to any "healthy" node (a node with no outstanding hints).
The idea is to not further overload already overloaded nodes. If you see OverloadedExceptions you'll have to repair after all nodes become stable. See StorageProxy.java#L1327 <https://github.com/apache/cassandra/blob/6bcc60ae4c440be1ba9f6c2b83f697ac2f2b3f3e/src/java/org/apache/cassandra/service/StorageProxy.java#L1327> called from StorageProxy.java::sendToHintedEndpoints() <https://github.com/apache/cassandra/blob/6bcc60ae4c440be1ba9f6c2b83f697ac2f2b3f3e/src/java/org/apache/cassandra/service/StorageProxy.java#L1248> On 13 July 2018 at 05:38, Karthick V <karthick...@gmail.com> wrote: > Refs : https://docs.datastax.com/en/cassandra/3.0/cassandra/operations/ > opsRepairNodesHintedHandoff.html > > On Thu, Jul 12, 2018 at 7:46 PM Karthick V <karthick...@gmail.com> wrote: > >> Hi everyone, >> >> If several nodes experience brief outages simultaneously, substantial >>> memory pressure can build up on the coordinator.* The coordinator >>> tracks how many hints it is currently writing, and if the number increases >>> too much, the coordinator refuses writes and throws the * >>> *OverloadedException exception.* >> >> >> In the above statement, it is been said that after some extent(of >> hints) the* coordinator *will refuse to writes. can someone explain the >> depth of this limitations and its dependency if any (like disk size or any)? >> >> Regards >> Karthick V >> >> >>
