Multi host DHCP networking and local DNS resolving

[Édouard Thuleau]

I have a multi host environment with VLAN networking running and all works fine except DNS resolving between instances on different compute nodes.

For example assume that there are 3 instances running (call them x, y and z) and 2 of them run on compute node A (x,y) and the third instance (z) runs on node B. The VMs x and y can access each other via their assigned DNS names (respectively x.novalocal and y.novalocal) but the VMs can not resolve the DNS for the third one (z.novalocal). And vice versa, the VM y cannot resolve VMs x and y.

Problem here is that each compute node has his own dnsmasq and mange its own local DNS entries but other compute nodes does not know how to access this information.

The workaround is to setup some external DNS if I want resolution between hosts.

[Édouard Thuleau (ethuleau)]

We can set up dnsmasq to use a configured hosts file which contains all DNS entries for the managed network.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/16153

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/16153
Committed: http://github.com/openstack/nova/commit/bdf53c30394f6fc6558d3be52db7166f3796b399
Submitter: Jenkins
Branch: master

commit bdf53c30394f6fc6558d3be52db7166f3796b399
Author: Édouard Thuleau <email address hidden>
Date: Wed Nov 14 18:59:01 2012 +0100

    Multi host DHCP networking and local DNS resolving

    Configure all dnsmasq instances to use the hosts file instead of the
    default local file '/etc/host' when the network is configured in
    'multi_host' mode.

    This hosts file contains all DNS entries of the network and it is
    replicated on all host where a 'nova-network' daemon is instantiated.
    There is a hosts file for each networks.

    It needs to add a new network API to update the host file on all
    network hosts.

    When a network host is called to (de)allocate a VM, it (de)allocate the
    instance fixed IP adress(es) on the concerned network host and when it's
    done, it calls all networks to update their DNS db entries if it's
    needed through a fanout cast.

    DocImpact: new config options
    Fixes LP bug #1078808

    Change-Id: I5c13bd895ebd31f276eb14e996025ddcfb67c3d9

[Stephan Fabel (sfabel)]

What's the possibility of this being fixed in Folsom?

[Jian Wen (wenjianhn)]

Remove the folsom-backport-potential tag

As Vish said in https://review.openstack.org/#/c/16153/
  "This is a feature so it won't be backported."

New feature is completely forbidden for stable branch.
https://wiki.openstack.org/wiki/StableBranch

[Stephan Fabel (sfabel)]

Not wanting to be pushy, but how exactly is a working DNS a 'new feature'?

[Jian Wen (wenjianhn)]

I think logically it's a bug, but it is fixed by adding new feature include
 a new network API and a new config option.

[Mark McLoughlin (markmc)]

I've targeted it to the folsom series so we can discuss here whether it's suitable for the branch. If not, we can just close this task as WontFix

[Dave Walker (davewalker)]

This bug certainly looks interesting as, instances on separate compute nodes can't resolve each others hostnames using multihost.. which would seem to be a pretty big issue for deployments. With the default config to be opt-in, it should mean no impact/config-changes for users that do not wish to use this behaviour?

Vishy, do you still feel this is inappropriate for folsom?

Thanks