Ask Your Question
0

Neutron high availability

asked 2015-03-01 03:53:53 -0500

fierabras gravatar image

Hi all.

I'm trying to build my Ubuntu Openstack Juno Cloud using HA. I'm going in trouble with Neutron, because I'm not able to implement it following the HA guide provided by Openstack. My conclusions are:

Openvswitch(ML2) may work (simultaneously) in all network nodes. Somewhere I've read that I may boot as many dhcp servers as network nodes, but really I don't know how to mplement it because I imagine they don't have to share ip ranges, etc ... For L3, the routers are going to be distributed across the network nodes, but no word about this distribution policy.

Is there any doc explaining this? The Openstack HA Guide is not enough.

Thanks.

edit retag flag offensive close merge delete

2 answers

Sort by ยป oldest newest most voted
0

answered 2015-03-01 06:37:37 -0500

SamYaple gravatar image

For DHCP that is simple. You can run as many DHCP agents as you would like. Since they are all backed by the database they all contain the same leases. This just means that they will all recieve and respond to the DHCPDISCOVER. The actual offer will occur between the DHCP agent that responded quickest.

For L3 you can have HA in an active/passive capacity fairly easily. There are some builtin Neutron helpers for this, but they are only half implemented. I would recommend the AT&T script (neutron-ha-tool.py) to help with that.

In Juno you have two new options for Openvswitch, L3HA + DVR.

L3HA uses VRRP/keepalived to move a vip between L3 router namespaces to allow high availability.

DVR attempts to use some trickery to allow each compute to be its own exit point to the internet. Effectively removing any networking bottleneck that could exist. It... does not work all that well right now. I would not recommend using it in production unless you have thouroughly tested it.

edit flag offensive delete link more

Comments

Thanks for your answer. I am really interested in active/passive option based on pacemaker for those plugins that cannot be executed at the same time in both nodes. I've red that you can do something using the same hostname for all the nodes. Thanks.

fierabras gravatar imagefierabras ( 2015-03-01 12:04:23 -0500 )edit

Yea if it is truly active/passive then you can have the same hostname. But you will have some down period. That AT&T script is the cleanest, quickest, most reliable way I have seen of doing it. But the active/passive can certainly be workable!

SamYaple gravatar imageSamYaple ( 2015-03-01 21:19:23 -0500 )edit
0

answered 2015-08-03 13:25:16 -0500

Gurri Bindra gravatar image

I have actualy used OVIRT (or XEN/VMWARE if you will) to create an HA environment and then have the Neutron VM sit in that environment. The HA is then taken care of by that layer. Works flawlessly unless a particular VM hangs (happened just one) In the production environment, a cron job from a remote server tests connectivity to the Neutron server on a specific application and just reboots the VM if need be. No problems so far !

edit flag offensive delete link more

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower

Stats

Asked: 2015-03-01 03:53:53 -0500

Seen: 462 times

Last updated: Mar 01 '15