Setting up networking on a compute node

asked 2011-06-30 01:55:47 -0600

cubranic gravatar image

If install only nova-compute on a compute node of a multi-node cluster, is it supposed to set up the networking (bridge, routes, iptables) by itself like what happened on the controller node (which runs all five nova services: network, api, compute, objectstore, and scheduler)?

I use FlatDHCPManager, and on the controller just using "flat_interface" flag was sufficient to add a bridge to that NIC and set up the routing tables with no input from me. But on the compute node, none of this happens after I set up nova.conf and restarted the compute service. I can run commands like euca-describe-instances on the node, and it even shows up in "nova-manage service list" run on the controller. But no Nova-related networking was added: there is no br100 bridge, iptables are empty, and there is no route to the subnet used by guest instances.

edit retag flag offensive close merge delete

5 answers

Sort by ยป oldest newest most voted
0

answered 2011-07-21 17:48:29 -0600

cubranic gravatar image

Thanks Vish Ishaya, that solved my question.

edit flag offensive delete link more
0

answered 2011-06-30 15:14:30 -0600

vishvananda gravatar image

The bridge should be created automatically when you launch an instance, but you cannot route into the instances from the compute host.

Vish

On Jun 29, 2011, at 6:55 PM, Davor Cubranic wrote:

New question #163228 on OpenStack Compute (nova): https://answers.launchpad.net/nova/+q...

If install only nova-compute on a compute node of a multi-node cluster, is it supposed to set up the networking (bridge, routes, iptables) by itself like what happened on the controller node (which runs all five nova services: network, api, compute, objectstore, and scheduler)?

I use FlatDHCPManager, and on the controller just using "flat_interface" flag was sufficient to add a bridge to that NIC and set up the routing tables with no input from me. But on the compute node, none of this happens after I set up nova.conf and restarted the compute service. I can run commands like euca-describe-instances on the node, and it even shows up in "nova-manage service list" run on the controller. But no Nova-related networking was added: there is no br100 bridge, iptables are empty, and there is no route to the subnet used by guest instances.


You received this question notification because you are a member of Nova Core, which is an answer contact for OpenStack Compute (nova).

edit flag offensive delete link more
0

answered 2011-07-11 22:44:59 -0600

cubranic gravatar image

Launching an instance on the compute node runs it instead on the controller. No bridge is created on the compute node.

edit flag offensive delete link more
0

answered 2011-07-12 00:14:00 -0600

cubranic gravatar image

I'm not sure how the instances are scheduled, but after starting up a few of them, all of which ran on the controller, one finally did start up on the compute node. I'm using the default scheduler. Does it pick a node based on some kind of load balancing?

Also, once the compute node started an instance, it set up br100 there, and I can ping and ssh into the instance running on the compute node from the controller. But compute nodes can't ping/ssh any instances, including ones they host. Is this by design? (no public ips are assigned, it's all private ips)

edit flag offensive delete link more
0

answered 2011-07-12 11:57:09 -0600

vishvananda gravatar image

The default scheduler is chance scheduler which randomly picks a node. The simple scheduler will pick the least loaded host.

--scheduler_driver=nova.scheduler.simple.SimpleScheduler

compute hosts by default don't get an ip on the vm network or a route into the vms.

Vish

On Jul 11, 2011, at 5:16 PM, Davor Cubranic wrote:

Question #163228 on OpenStack Compute (nova) changed: https://answers.launchpad.net/nova/+q...

Davor Cubranic posted a new comment: I'm not sure how the instances are scheduled, but after starting up a few of them, all of which ran on the controller, one finally did start up on the compute node. I'm using the default scheduler. Does it pick a node based on some kind of load balancing?

Also, once the compute node started an instance, it set up br100 there, and I can ping and ssh into the instance running on the compute node from the controller. But compute nodes can't ping/ssh any instances, including ones they host. Is this by design? (no public ips are assigned, it's all private ips)


You received this question notification because you are a member of Nova Core, which is an answer contact for OpenStack Compute (nova).

edit flag offensive delete link more

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower

Stats

Asked: 2011-06-30 01:55:47 -0600

Seen: 85 times

Last updated: Jul 21 '11