[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [xmlblaster] two xmlblaster servers with one floating IP address?
Dominique Petitpierre wrote:
is there a way to have redundancy at the level of the xmlblaster
servers that is transparent for the clients (i.e. the clients do not
need to be programmed specially)?
More specifically, I would like to use two server machines with a
floating service address managed by heartbeat
(http://www.linux-ha.org/heartbeat/ ). A client would see only one
xmlblaster server; the currently serving server would communicate with
the standby server in order to maintain exactly the same state on both
machines (i.e. a message is considered written to or read from a queue
only when both servers have completed the necessary operations). If
the serving server fails, then heartbeat initiates a failover, the
floating address is taken over by the standby server which becomes the
serving server and the client continues with that server
(not noticing anything if no operation was currently going on, or
possibly restarting the current failed operation).
When the failed server is available again, it catches up on
the states of the queues and becomes a usable standby server.
Sorry, mirroring is not yet implemented.
We almost started on it last winter but due to
lack of resources we had to defer it.
Our requirements where exactly as yours (your compendium
is a nice overview).
Our current cluster approach is a good base to extend it to
support such a mirroring, but coding testsuite, specifying details
coding the implementation and make a documentation takes its time ...
I'd like to use this in a PtX framework where multiple
clients read the same message sent by one publisher.
Load balancing is not a concern (yet :-)).
I read the cluster requirement page on
but could not determine if that approach was possible. In particular
the note mentionned under "Multiple masters for same domain"
| If multiple nodes acquire the master mode for a certain domain there
| are two approaches.
| Messages which are published are sent to all masters.
it seems strange that the client should handle many message
submissions (with all the synchronisation problems that it entails in
case of failure). I'd prefer the clients to think there is
only one server with which it has to sometimes restart
an operation that did not complete successfully.
Yes, what you say is ment there, the server to which the client
talks is responsible to forward it to other masters.
For the time being you can extend the cluster support
in xmlBlaster yourself or hire somebody (e.g. the xmlBlaster maintainers)
to do it for you.
As far as i know no xmlBlaster contributer is currently working on mirroring.
Thanks in advances for sharing you experience and knowledge!
Mr. Dominique Petitpierre Email: User at Domain
Division Informatique User=Dominique.Petitpierre
University of Geneva Domain=adm.unige.ch
(Switzerland) WWW : http://www.unige.ch/dinf/