On Wed, Apr 27, 2011 at 5:23 PM, Terje Marthinussen
<tmarthinus...@gmail.com> wrote:
> I have two issues here.
> - The massive amount of mutation caused by the hints playback
I'm not sure how one node playing back hints could cause this. The
intent of the code in HintedHandoffManager is to send a single
mutation, wait for it to be completed, then proceed to the next. This
is deliberately not parallel to avoid overwhelming the recipient:
rm.add(cf);
IWriteResponseHandler responseHandler =
WriteResponseHandler.create(endpoint);
MessagingService.instance().sendRR(rm, endpoint, responseHandler);
try
{
responseHandler.get();
}
catch (TimeoutException e)
{
return false;
}
> - The long periods where there are no increase in completed mutations.
When all your memtables and flush queues are full, mutations have
stop. Otherwise it will OOM trying to hold an increasing number of
full memtables.
--
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of DataStax, the source for professional Cassandra support
http://www.datastax.com
