Failover and updateClusterClients

[Mark White <ma...@somsip.com>]

I'm a newbie at this, so please excuse any limitations in my understanding.

We're looking at a load-balanced network of brokers with failover, using a
fixed list of URIs with 4 servers (A - D). We'll be using a common config
across brokers (except broker name). A JWS client will be used.

Client:
failover:(nio://A:61616,nio://B:61616,nio://C:61616,nio://D:61616)

Brokers:
<networkConnector name="example"
uri="static://(nio://A:61616,nio://B:61616,nio://C:61616,nio://D:61616)"/>
...
<transportConnector name="openwire"
uri="failover:(nio://A:61616,nio://B:61616,nio://C:61616,nio://D:61616)"
updateClusterClients="true" />

Questions:

1) Does it matter if the broker's own IP is included in networkConnector?

2) If broker A goes down, and a client attempts to connect to broker A, is
it 'timeout' that affects how quickly the client will failover to a
different broker?

3) The broker cluster is aware that A has dropped. How is
updateClusterClients represented at the client? Would it be similar to the
client connecting using a random broker from
"failover:(nio://B:61616,nio://C:61616,nio://D:61616)"?

4) If broker A is restored, will the client be advised of this on the next
successful communication with any broker? If not, when or how?

Thanks for any clarification anyone can give.


Mark...
-- 
View this message in context: http://old.nabble.com/Failover-and-updateClusterClients-tp28967819p28967819.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.

Re: Failover and updateClusterClients

[Mark White <ma...@somsip.com>]

> but on the client side - use tcp:// instead of nio - you will get better
performance.

Thanks for the advice on this, and the rest of your answer. Much
appreciated.


Mark...
-- 
View this message in context: http://old.nabble.com/Failover-and-updateClusterClients-tp28967819p28979962.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.

Re: Failover and updateClusterClients

[Mark White <ma...@somsip.com>]

rajdavies wrote:
> 
> This is a new feature in 5.4 - so we are all newbies to it ;) - answers
> inlined:
> 

Ah. I just noticed that 5.4 is not yet in stable branch. I didn't realise
these features were *that* new :)
-- 
View this message in context: http://old.nabble.com/Failover-and-updateClusterClients-tp28967819p28988990.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.

Re: Failover and updateClusterClients

[Rob Davies <ra...@gmail.com>]

This is a new feature in 5.4 - so we are all newbies to it ;) - answers inlined:
On 23 Jun 2010, at 06:02, Mark White wrote:

> 
> I'm a newbie at this, so please excuse any limitations in my understanding.
> 
> We're looking at a load-balanced network of brokers with failover, using a
> fixed list of URIs with 4 servers (A - D). We'll be using a common config
> across brokers (except broker name). A JWS client will be used.
> 
> Client:
> failover:(nio://A:61616,nio://B:61616,nio://C:61616,nio://D:61616)
> 
> Brokers:
> <networkConnector name="example"
> uri="static://(nio://A:61616,nio://B:61616,nio://C:61616,nio://D:61616)"/>
> ...
> <transportConnector name="openwire"
> uri="failover:(nio://A:61616,nio://B:61616,nio://C:61616,nio://D:61616)"
> updateClusterClients="true" />
> 
> Questions:
> 
> 1) Does it matter if the broker's own IP is included in networkConnector?
No - you will see a warning - but the broker won't connect to itself
> 
> 2) If broker A goes down, and a client attempts to connect to broker A, is
> it 'timeout' that affects how quickly the client will failover to a
> different broker?
Depends on the OS system - the underlying tcp protocol will usually throw an exception - first before a timeout in this case
> 
> 3) The broker cluster is aware that A has dropped. How is
> updateClusterClients represented at the client? Would it be similar to the
> client connecting using a random broker from
> "failover:(nio://B:61616,nio://C:61616,nio://D:61616)"?
You don't need all the brokers (though it doesn't matter if you do) - the client will be updated automatically - but
on the client side - use tcp:// instead of nio - you will get better performance.
> 
> 4) If broker A is restored, will the client be advised of this on the next
> successful communication with any broker? If not, when or how?
Yes the client will be informed about it
> 
> Thanks for any clarification anyone can give.
> 
> 
> Mark...
> -- 
> View this message in context: http://old.nabble.com/Failover-and-updateClusterClients-tp28967819p28967819.html
> Sent from the ActiveMQ - User mailing list archive at Nabble.com.
> 

cheers,

Rob

Rob Davies
follow me: http://twitter.com/rajdavies
I work here: http://fusesource.com
My Blog: http://rajdavies.blogspot.com/
I'm writing this: http://www.manning.com/snyder/