cluster master ip

[Serge van Ginderachter]

Hi,

Quick question. 

So far I set up a 1 node cluster. I noticed the cluster master ip isn't assigned to any nic/bridge. I assumed the cluster software would manage that.

- Is that a config problem, or 

- is there a particular reason for this in a 1 node setup?

Serge

[Guido Trotter]

On Thu, Aug 12, 2010 at 1:20 PM, Serge van Ginderachter
<s...@vanginderachter.be> wrote:

Hi,

Could you post the output of gnt-cluster info, and in particular check
what the master-netdev is?
Also check what the ip is, confirm that's not pingable, and try
restarting the ganeti daemons, and check if there are errors in the
daemon logs.

Thanks,

Guido

[Serge van Ginderachter]

On 12 August 2010 21:49, Guido Trotter <ultr...@gmail.com> wrote:

> So far I set up a 1 node cluster. I noticed the cluster master ip isn't
> assigned to any nic/bridge. I assumed the cluster software would manage
> that.

Could you post the output of gnt-cluster info,

http://paste.pocoo.org/show/249675/

 

and in particular check
what the master-netdev is?

br0, first bridge I set up, facing 'public' 

 

Also check what the ip is, confirm that's not pingable,

It's a dedicated IP, not allocated elswhere.

 

and try
restarting the ganeti daemons, and check if there are errors in the
daemon logs.

OK, restarting the daemons created br0:0, which pretty much answers my question.

On a side note, I'm aloso trying to understand the to what extent the networking parameters are important to the ganeti backend and how those interact.

- Would it be possible to assign the cluster ip from a range on the secondary network? My guess is that it wouldn't be 'the secondary network, but the network on br1, which now is the secondary network and would become master-netdev?

- Likewise, is it necessary for the node ip's (which whome the nodes are registered) to live on a public facing network? Would it make sense to put the master network, on a private segment? Of course, each node would need to retain a 'public' ip for ssh access.

Thanks for your support,

Serge

[Guido Trotter]

On Fri, Aug 13, 2010 at 7:47 AM, Serge van Ginderachter
<s...@vanginderachter.be> wrote:
> OK, restarting the daemons created br0:0, which pretty much answers my
> question.
>

Ah, perfect. Then the only remaining problem is why it didn't create
it the first time around...

> On a side note, I'm aloso trying to understand the to what extent the
> networking parameters are important to the ganeti backend and how those
> interact.
> - Would it be possible to assign the cluster ip from a range on the
> secondary network? My guess is that it wouldn't be 'the secondary network,
> but the network on br1, which now is the secondary network and would become
> master-netdev?

The secondary network doesn't need to be on bridges. That said, I'm
not completely sure if it's possible to assign the cluster master ip
from there as there might be parts of the code that check for its
existence by binding to the primary ip. Needs to be checked (don't
remember right now).

> - Likewise, is it necessary for the node ip's (which whome the nodes are
> registered) to live on a public facing network? Would it make sense to put
> the master network, on a private segment? Of course, each node would need to
> retain a 'public' ip for ssh access.

You can do that, no problem. As long as the nodes can talk to each
other, all is fine.

Thanks,

Guido

[Serge van Ginderachter]

On 16 August 2010 15:59, Guido Trotter <ultr...@gmail.com> wrote:

> On a side note, I'm aloso trying to understand the to what extent the
> networking parameters are important to the ganeti backend and how those
> interact.
> - Would it be possible to assign the cluster ip from a range on the
> secondary network? My guess is that it wouldn't be 'the secondary network,
> but the network on br1, which now is the secondary network and would become
> master-netdev?

The secondary network doesn't need to be on bridges. That said, I'm
not completely sure if it's possible to assign the cluster master ip
from there as there might be parts of the code that check for its
existence by binding to the primary ip. Needs to be checked (don't
remember right now).

> - Likewise, is it necessary for the node ip's (which whome the nodes are
> registered) to live on a public facing network? Would it make sense to put
> the master network, on a private segment? Of course, each node would need to
> retain a 'public' ip for ssh access.

You can do that, no problem. As long as the nodes can talk to each
other, all is fine.

I just noticed that communication for relocating a disk, or modifying a disk from plain to drbd, passes through the primary netdev.

In my test setup, primary netdev is the "public facing" network, running on 100Mbit network, whilst I have a dedicated secondary network for drbd.

Is this by design that this kind of trafiic doesn't get to use the secondary network? At least in my case this seems counter-productive?

For now, it would seem to me I'd better use the backend network (dedicated 1GB link) as primary network for everything, and configure other interfaces afterwards. Are there any recomandations of best practices you guys follow on this?

thanks,

Serge

[Guido Trotter]

On Thu, Aug 19, 2010 at 5:19 PM, Serge van Ginderachter
<s...@vanginderachter.be> wrote:

> I just noticed that communication for relocating a disk, or modifying a disk
> from plain to drbd, passes through the primary netdev.
> In my test setup, primary netdev is the "public facing" network, running on
> 100Mbit network, whilst I have a dedicated secondary network for drbd.
> Is this by design that this kind of trafiic doesn't get to use the secondary
> network? At least in my case this seems counter-productive?
> For now, it would seem to me I'd better use the backend network (dedicated
> 1GB link) as primary network for everything, and configure other interfaces
> afterwards. Are there any recomandations of best practices you guys follow
> on this?
>

What happens right now is that all communication passes through the
primary network, apart from actual DRBD traffic.
In your case perhaps you want to use your secondary ips as primary
ips, not have a replication network at all, and connect your instances
to a bridge connected to the "public facing" network, so they don't
share a network with what the cluster does at all.

Thanks,

Guido

> thanks,
>
> Serge
>
>

--
Guido Trotter
Google - Corporate Computing Services SRE

Google Ireland Ltd. : Registered in Ireland with company number 368047.
Gordon House, Barrow Street, Dublin 4, Ireland.