[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [xmlblaster] RE: [xmlblaster] Mirrored Masters?

Craig McIlwee schrieb:
Thanks Marcel.  What we are looking for is nowhere near as elegant of a solution as you've outlined below.  Quoting the same page as David, we are most interested in

"Messages which are published are sent to all masters."

Because we really just want a second master node that stays in sync with the primary master.   In our scenario we don't need any type of smart failover or load balancing, we want to make sure that 2 nodes are receiving and persisting the same information.  In normal usage, the primary node will always receive messages.  In a catastrophic situation where the primary node fails (hardware error, etc) we need to be able to switch to the secondary node that should be in sync with the primary, and manual reconfiguration is acceptable.

So from your mail below, I take it that persisting at 2 nodes is not implemented yet?
No this is not implemented.
If not, it sounds like you recommend mirroring at the database level?
Yes. And I would recommend to go from the beginning the Linux HA (bonding,DRDB,Heartbeat etc, HP-ServiceGuard ...) way,
as all the failover problems will arise over time anyhow, so I suggest doing it right from beginning.


-----Original Message-----
From: owner-xmlblaster at server.xmlBlaster.org [mailto:owner-xmlblaster at server.xmlBlaster.org] On Behalf Of Marcel Ruff
Sent: Tuesday, December 08, 2009 4:41 AM
To: xmlblaster at server.xmlBlaster.org
Subject: Re: [xmlblaster] Mirrored Masters?

Hi David,

multi master is not implemented.

The I_LoadBalancer#getClusterNode() returns one node (NodeMasterInfo) which
is used to forward messages to the next node, but only to one.

If you need mirroring you could, depending if you need "failover" or "load balancing",
consider following:

1) Change the xmlBlaster cluster code to handle multiple masters
You need to decide in this case if the multiple masters are informed
in a transaction (sync) or not (async). Clients need to be coded
to choose a master depending on current load or failure.
-> failover and load balancing

2) You could create an extended I_Queue and I_Map plugin which
duplicates the data to two DBs
-> failover

3) You could do low-level DB mirroring (DRDB, Linux HA et al)
xmlBlaster gets on startup all information to re-establish its state.
In this case the mirrored xmlBlaster is started on failure of the first,
clients automatically find the mirrored as the IP is change (standard HA behaviour)
-> failover

4) Master/Slave operation (exists already, one master with many salves)
-> load balancing

5) Combination of 3) and 4) should work out of the box
-> failover and load balancing

It depends on what you want to achieve,

best regards

David R Robison schrieb:
In the discussion on clustering in the reference manual it talks about mirrored masters:

    /An xmlBlaster cluster allows to have more than one master server
    for a specific message domain. The master nodes are //mirrored
    instances for those messages. Published messages reach all master
    nodes. Subscribed messages are retrieved using a load balance

However, the discussion says that it has not been implemented. Has any work been done on this? Any suggestions on how to achieve this?
Thanks, David

-- Marcel Ruff http://www.xmlBlaster.org http://gpsvision.biz