openstack team mailing list archive

[Everett Toews <everett.toews@xxxxxxxxx>]

Yep, if you were just running two nova-api processes, you'd have two entry
points. Hence the need for something like HAProxy (or Pound) to act as your
single EC2 entry point but passing requests to your nova-api processes.
Everything being stateless you wouldn't get into a state where the DB &
messages were incorrect.

We haven't looked at replicating rabbitmq yet. If you come up with a
solution, please share.

Everett

On Thu, Aug 18, 2011 at 3:21 PM, Joshua Harlow <harlowja@xxxxxxxxxxxxx>wrote:

>  The way suggested here would setup them as different ec2 entry points
> though right? (the different port part).
>
> If you have one nova-api on one machine Y with X port and then another
> nova-api on another machine Z with X port would that work.
>
> It would seem possible, just not sure if the code would handle that
> correctly (is everything really stateless...)
>
> That way if Y machine dies, action will still occur on machine Z (just
> whatever Y may be doing is dead – which may result in DB state & messages
> being incorrect?).
>
> It seems like rabbitmq has active/passive failure, has anyone looked into
> that?
>
>
> On 8/18/11 12:27 PM, "Everett Toews" <everett.toews@xxxxxxxxx> wrote:
>
> Yes and maybe (we're doing it).
>
> The individual nova-api processes need to run on their own port though.
>
> nova-api --flagfile=/etc/nova/nova.conf
> --logfile=/var/log/nova/nova-api-<port1>.log --ec2_listen_port=<port1>
> nova-api --flagfile=/etc/nova/nova.conf
> --logfile=/var/log/nova/nova-api-<port2>.log --ec2_listen_port=<port2>
>
> you can fire up multiple nova-api processes with something like monit. Then
> you can load balance between them with something like haproxy.
>
> Everett
>
> On Thu, Aug 18, 2011 at 11:43 AM, Joshua Harlow <harlowja@xxxxxxxxxxxxx>
> wrote:
>
> Is it possible to have multiple nova-api’s, nova-networks running in the
> same instance (connected to the same db/rabbitmq)?
>
> Say that you want to have fault tolerance, so you would have multiple
> instances of these, so that if one failed the whole iaas system would not.
>
> Is that something typically done?
>
> _______________________________________________
> Mailing list: https://launchpad.net/~openstack
> Post to     : openstack@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~openstack
> More help   : https://help.launchpad.net/ListHelp
>
>
>
>