activemq-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Charles Anthony" <charles.anth...@hpdsoftware.com>
Subject RE: Failover Advice
Date Tue, 03 Oct 2006 14:54:33 GMT
Thanks.

One broker in the failover sounds promising; so if we kill the broker,
the clients & server will repeatedly try to connect to each broker in
the list - and if there's only one, they'll just "wait" until it
reappears ?

Perhaps a silly question, but do they ever give up ? Or is this where
some groovy back-off algorithm kicks in ?

> -----Original Message-----
> From: James Strachan [mailto:james.strachan@gmail.com] 
> Sent: 03 October 2006 15:50
> To: activemq-users@geronimo.apache.org
> Subject: Re: Failover Advice
> 
> :)
> 
> So yes the URL you use at the end sounds line to me. Run 2 
> brokers if you like, otherwise just use failver: on the one 
> broker (so that bad/hung TCP connections get recreated).
> 
> On 10/3/06, Charles Anthony <charles.anthony@hpdsoftware.com> wrote:
> > I'm looking for some advice, please, on possible failover 
> approaches.
> >
> > We have a single standalone ActiveMQ Broker, and both application 
> > clients and an application servers connective to the broker via tcp.
> > The messages are non-persistent, and we have persitence 
> turned off on 
> > the ActiveMQ Broker.
> >
> > The application server actually resides on the same server as the 
> > ActiveMQ server; the clients are coming in over a VPN connection.
> >
> > We have two internet lines coming in - and sometimes we 
> need to swap 
> > lines, due to routing problems. Our IT chaps tinker away with some 
> > VPN/router settings, and hey presto, the TCP traffic is 
> rerouted over 
> > the other line.
> >
> > Sometimes, though not always, it seems this causes the TCP 
> Connections 
> > to ActiveMQ to hang (sometimes they go to TCP_WAIT states).
> >
> > Now, obviously I'd rather the connections didn't go belly up - but 
> > anecodotally (i.e. through observation, not through proof) they do 
> > occasionally do this.
> >
> > How can I set up a fail over scenario here ? Our broker, 
> and all the 
> > clietns and servers are configured using URLs; the Broker currently 
> > listens on
> >
> > tcp://localhost:61616
> >
> > The application server connections to the same url
> > (tcp://localhost:61616)
> > The clients connect using a dns-resolved name,
> > (tcp://examplemachine.host.com:61616)
> >
> > I'm actually not bothered if existing messages get lost - 
> what I don't 
> > want to do is restart my application server, because we have some 
> > lengthy initialisation logic which can take up to half an hour.
> >
> > Would this work (yes, I know I should just try it - but SHOULD this
> > work) :
> >
> > configure two separate ActiveMQ Brokers on the same machine
> >
> > Broker 1 tcp://localhost:61616
> > Broker 2 tcp://localhost:61700
> >
> > Configure the clients/application server to use a  failover URL :
> > 
> failover://(tcp://examplemachine.host.com:61616,tcp://examplemachine.h
> > os
> > t.com:61616)?randomize=false
> >
> > Would that make any sense ?
> >
> > We're using a snapshopt of ActiveMQ 4.0 from May.
> >
> > Excuse the hurried email that probably doesn't make a great deal of 
> > sense; disgruntled clients are applying pressure...
> >
> > Cheers,
> >
> > Charles.
> >
> 
> 
> -- 
> 
> James
> -------
> http://radio.weblogs.com/0112098/
> 

Mime
View raw message