5000 rows in a mutli get is way, way, way (did I say way ? ) to many. Whenever you get a TimedOutException check the tp stats on the nodes, you will normally see a high pending count. Every row get get turns into an message in a TP. So if you ask for 5k rows you flood the TP with 5k messages which will often result in the node(s) been temporarily overloaded.
More is not always more. I would guess 100 as a starting point, i would be doubtful that you would see much benefit beyond 1000. pycassa defaults to 1024 https://github.com/pycassa/pycassa/blob/master/pycassa/columnfamily.py#L63 Cheers ----------------- Aaron Morton Freelance Cassandra Developer @aaronmorton http://www.thelastpickle.com On 6/10/2011, at 9:14 AM, Yuhan Zhang wrote: > Hi all, > > I have been experiencing the unavailableException and TimedOutException on a > 3-node cassandra cluster > during a multiGetSliceQuery with 1000 columns. Since there are many keys > involved in the query, I divided > them into groups of 5000 rows and process each group individually in a for > loop. but seems like it is not helping.. > Once the TimedOutException appears, further requests to cassandra will cause > UnavailableException. > However, the servers can recover after a while without intervention. > > Which settings should I pay attention to in order to fix the problem? This > problem becomes very frequent recently. > > > Thank you. > > Yuhan > > The exception looks like: > > 1/10/05 13:05:31 ERROR connection.HConnectionManager: Could not fullfill > request on this host > CassandraClient<ec2-75-101-238-70.compute-1.amazonaws.com:9160-33> > 11/10/05 13:05:31 ERROR connection.HConnectionManager: Exception: > me.prettyprint.hector.api.exceptions.HTimedOutException: TimedOutException() > at > me.prettyprint.cassandra.service.ExceptionsTranslatorImpl.translate(ExceptionsTranslatorImpl.java:32) > at > me.prettyprint.cassandra.service.KeyspaceServiceImpl$3.execute(KeyspaceServiceImpl.java:161) > at > me.prettyprint.cassandra.service.KeyspaceServiceImpl$3.execute(KeyspaceServiceImpl.java:143) > at > me.prettyprint.cassandra.service.Operation.executeAndSetResult(Operation.java:101) > at > me.prettyprint.cassandra.connection.HConnectionManager.operateWithFailover(HConnectionManager.java:155) > ... > Caused by: TimedOutException() > at > org.apache.cassandra.thrift.Cassandra$get_range_slices_result.read(Cassandra.java:12104) > at > org.apache.cassandra.thrift.Cassandra$Client.recv_get_range_slices(Cassandra.java:732) > > > 11/10/05 20:06:05 ERROR connection.HConnectionManager: Could not fullfill > request on this host > CassandraClient<ec2-184-73-116-237.compute-1.amazonaws.com:9160-15> > 11/10/05 20:06:05 ERROR connection.HConnectionManager: Exception: > me.prettyprint.hector.api.exceptions.HUnavailableException: > UnavailableException() > at > me.prettyprint.cassandra.service.ExceptionsTranslatorImpl.translate(ExceptionsTranslatorImpl.java:50) > at > me.prettyprint.cassandra.service.KeyspaceServiceImpl$10.execute(KeyspaceServiceImpl.java:397) > at > me.prettyprint.cassandra.service.KeyspaceServiceImpl$10.execute(KeyspaceServiceImpl.java:383) > at > me.prettyprint.cassandra.service.Operation.executeAndSetResult(Operation.java:101) > > Caused by: UnavailableException() > at > org.apache.cassandra.thrift.Cassandra$multiget_slice_result.read(Cassandra.java:9620) > at > org.apache.cassandra.thrift.Cassandra$Client.recv_multiget_slice(Cassandra.java:636) > at > org.apache.cassandra.thrift.Cassandra$Client.multiget_slice(Cassandra.java:608) > at > me.prettyprint.cassandra.service.KeyspaceServiceImpl$10.execute(KeyspaceServiceImpl.java:388) > ... 35 more >
