Guillaume, Ken,
I filed the issue under https://forge.continuent.org/jira/browse/SEQUOIA-905
One more time, you should consider adding another controller to your
configuration to avoid the introduction of a new SPOF.
Some people just use Sequoia as a load balancer and don't necessary need
multiple instances if their application is not clustered.
Emmanuel : the problem seems to be that
AbstractRequest.setLazyTransactionStart is only called in the
distributed case (in DistributedRequestManager.lazyTransactionStart).
It prevents the begin from being logged in the recovery log because in
RAIDb1.execWriteRequest (which is the method used in Ken's case),
logLazyTransactionBegin is only executed if
request.isLazyTransactionStart() returns true.
I fixed the RequestManager to properly set the value. Thanks for
tracking this down.
Shouldn't we simply force the users to use a <Distribution> element?
Perhaps it's already the case in Sequoia 3.0 beta? I have really a lot
of weird behaviours when I'm not in distributed mode and it seems that
there is a lot of bugs specific to this code path.
I contemplated for some time the option of enforcing the Distribution
element but there are a couple of scenarios that can make sense with a
single controller. This is mostly load balancing use cases and we don't
want to lose that flexibility. However, we need to improve the testing
of these configurations.
Let me know if the fix I commited in 3.0 works for you and I'll backport
it to 2.10.
Thanks again for the feedback,
Emmanuel
--
Emmanuel Cecchet
Chief Scientific Officer, Continuent
Blog: http://emanux.blogspot.com/
Open source: http://www.continuent.org
Corporate: http://www.continuent.com
Skype: emmanuel_cecchet
Cell: +33 687 342 685
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
