It turned out to be due to a rolling restart that was being done (however
the management node had already been restarted, it was only data nodes that
were being done). As soon as the rolling restart completed, the timings
dropped to around a second.
On Wed, Dec 5, 2012 at 8:27 AM, Craig L Russell <craig.russell@stripped>wrote:
> Hi Graeme,
> On Dec 3, 2012, at 11:46 AM, Graeme Wallace wrote:
> On my cluster its currently taking 20-25 seconds to get a SessionFactory
>> instance back from ClusterJHelper.**getSessionFactory(props)
>> Is there anything I can do to make this faster ?
> Do you get the same response from e.g. starting a mysqld server? It's
> basically the same code.
> There's nothing special about the way clusterj gets a session factory
> (ndb_cluster_connection) but it does require contacting the management node
> and setting up tcp connections from the client node to all data nodes. If
> the management node is slow, overloaded, on a slow network, etc. I might
> expect the initial connection to be slow.
>> Graeme Wallace
>> O: 972 588 1414
>> M: 214 681 9018
> Craig L Russell
> Architect, Oracle
> 408 276-5638 mailto:Craig.Russell@oracle.**com <Craig.Russell@stripped>
> P.S. A good JDO? O, Gasp!
O: 972 588 1414
M: 214 681 9018