Thanks. You resolved my Problem. My error : I didn't see transport.open, open a new socket for each call. I did think it's reuse the same one.
2010/5/20 Jonathan Ellis <[email protected]> > "disseminating load info" is not related to your problem. > > certainly you should be using connection pooling rather than opening a > ton of sockets. > > You didn't say what CL you are using, but you should not use CL.ZERO > for load tests. > > On Thu, May 20, 2010 at 11:14 AM, xavier manach <[email protected]> wrote: > > Hi. > > > > I am neebie in Cassandra. I study and compare performance of databases > for > > choose my future architecture. > > > > I try to load a lot of datas in cassandra. > > I use python with protocol thrift (very simple, whithout threading) > > I do sequentials requests : client.batch_mutate and client.get_slice. > > (around 1M mutate y 0.3M get mixed for 10 columns family) > > I didn't wait between each request. > > I use only one node (I have the same probleme with 2 nodes) > > > > It's start fine, but in the middle of the load > > during an transport.open() > > I have this exception : > > TTransportException.NOT_OPEN error > > with the message : could not connect to <IP>:<PORT> > > > > The server didn't respond for seconds ( between 1 and 30) > > After, it's work again... > > > > I turn the log to DEBUG, > > When the server did'nt respond, I have this message : > > Disseminating load info... > > Whats mean this ? > > > > I had forget something ? What is wrong ? > > How avoid it ? > > > > > > Thanks for your help. > > > > > > > > -- > Jonathan Ellis > Project Chair, Apache Cassandra > co-founder of Riptano, the source for professional Cassandra support > http://riptano.com >
