You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@qpid.apache.org by Marnie McCormack <ma...@googlemail.com> on 2006/10/06 09:47:57 UTC
Clustering - Current Setup/Questions
All,
In response to a user query, I've been looking (a little) at the existing
clustering code and wanted to check my assumptions about how a user would go
about setting up/starting a cluster.
Be good if people who've run a cluster or are more familiar with the code
could confirm please ?
- to start a broker which will participate in a cluster use the class
org.apache.qpid.server.cluster.Main and the -j option which should specify
host:port
- then start x other brokers to join the cluster
- the ClusteredProtocolHandler will then intercept the events usually
handled (in a non-clustered broker) by the ProtocolHandler and manages them
according to the state of the cluster member
I have a couple of quick questions:
- is there any way to configure a cluster using the broker config. I had
initially imagined that you'd be able to do that but couldn't spot a route
in that way ?
- how do you go about running up more than one cluster on the same host ? (I
had guessed that you'd specify an identifier for the cluster but don't think
that's correct)
- is there anyway to default failover behaviour in a cluster such that
clients don't have to specify the string of hosts for failover at connection
time, rather the cluster leader can manage failover from brokers that it
knows about ?
Does anyone have test scripts for starting up a cluster ? Couldn't see any
in svn.
Thanks & Regards,
Marnie
Re: Clustering - Current Setup/Questions
Posted by John O'Hara <jo...@gmail.com>.
If we can get 200,000 messages per second out of one broker, clustering for
HA on persistent queues and seamless failover to a shadow broker are more
important than clustering for performance (which is easier, and possibly a
distraction from the important stuff).
My 2c
John
On 06/10/06, Gordon Sim <gs...@redhat.com> wrote:
>
> Marnie McCormack wrote:
> > - to start a broker which will participate in a cluster use the class
> > org.apache.qpid.server.cluster.Main and the -j option which should
> specify
> > host:port
> > - then start x other brokers to join the cluster
>
> You start the first broker without the join option, then subsequent
> members are started with the join option giving the address of an
> existing cluster member (e.g. the first one started).
>
> > - is there any way to configure a cluster using the broker config. I had
> > initially imagined that you'd be able to do that but couldn't spot a
> route
> > in that way ?
>
> You can use a config file for each broker, but not a global file to
> startup a set of clustered brokers. The join address can't as yet be
> specified in the config either I don't think.
>
> > - how do you go about running up more than one cluster on the same host
> > ? (I
> > had guessed that you'd specify an identifier for the cluster but don't
> > think
> > that's correct)
>
> When you start a cluster broker it either forms a new cluster or joins
> the cluster referred to be the argument to the join option.
>
> > - is there anyway to default failover behaviour in a cluster such that
> > clients don't have to specify the string of hosts for failover at
> > connection
> > time, rather the cluster leader can manage failover from brokers that it
> > knows about ?
>
> Unfortunately there is not at present.
>
>
Re: Clustering - Current Setup/Questions
Posted by Gordon Sim <gs...@redhat.com>.
Marnie McCormack wrote:
> - to start a broker which will participate in a cluster use the class
> org.apache.qpid.server.cluster.Main and the -j option which should specify
> host:port
> - then start x other brokers to join the cluster
You start the first broker without the join option, then subsequent
members are started with the join option giving the address of an
existing cluster member (e.g. the first one started).
> - is there any way to configure a cluster using the broker config. I had
> initially imagined that you'd be able to do that but couldn't spot a route
> in that way ?
You can use a config file for each broker, but not a global file to
startup a set of clustered brokers. The join address can't as yet be
specified in the config either I don't think.
> - how do you go about running up more than one cluster on the same host
> ? (I
> had guessed that you'd specify an identifier for the cluster but don't
> think
> that's correct)
When you start a cluster broker it either forms a new cluster or joins
the cluster referred to be the argument to the join option.
> - is there anyway to default failover behaviour in a cluster such that
> clients don't have to specify the string of hosts for failover at
> connection
> time, rather the cluster leader can manage failover from brokers that it
> knows about ?
Unfortunately there is not at present.