[infra] NetworkManager on infra Fedora 29 and CentOS nodes
Hello, Just a heads-up; with Fedora 29 the legacy networking setup was moved into a separate, not-installed-by-default network-scripts package. This has prompted us to finally move to managing interfaces on our Fedora and CentOS CI hosts with NetworkManager (see [1]) Support for this is enabled with features added in glean 1.13.0 and diskimage-builder 1.19.0. The newly created Fedora 29 nodes [2] will have it enabled, and [3] will switch CentOS nodes shortly. This is tested by our nodepool jobs which build images, upload them into devstack and boot them, and then check the networking [4]. I don't really expect any problems, but be aware NetworkManager packages will appear on the CentOS 7 and Fedora base images with these changes. Thanks -i [1] https://bugzilla.redhat.com/show_bug.cgi?id=1643763#c2 [2] https://review.openstack.org/618672 [3] https://review.openstack.org/619960 [4] https://review.openstack.org/618671
On Tue, Jan 8, 2019 at 11:15 PM Ian Wienand <iwienand@redhat.com> wrote:
Hello,
Just a heads-up; with Fedora 29 the legacy networking setup was moved into a separate, not-installed-by-default network-scripts package. This has prompted us to finally move to managing interfaces on our Fedora and CentOS CI hosts with NetworkManager (see [1])
Support for this is enabled with features added in glean 1.13.0 and diskimage-builder 1.19.0.
The newly created Fedora 29 nodes [2] will have it enabled, and [3] will switch CentOS nodes shortly. This is tested by our nodepool jobs which build images, upload them into devstack and boot them, and then check the networking [4].
Don't suppose we could try this with tripleo jobs prior to cutting them all over could we? We don't use NetworkManager and infact os-net-config doesn't currently support NetworkManager. I don't think it'll cause problems, but I'd like to have some test prior to cutting them all over. Thanks, -Alex
I don't really expect any problems, but be aware NetworkManager packages will appear on the CentOS 7 and Fedora base images with these changes.
Thanks
-i
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1643763#c2 [2] https://review.openstack.org/618672 [3] https://review.openstack.org/619960 [4] https://review.openstack.org/618671
On Wed, Jan 09, 2019 at 09:02:57AM -0700, Alex Schultz wrote:
Don't suppose we could try this with tripleo jobs prior to cutting them all over could we? We don't use NetworkManager and infact os-net-config doesn't currently support NetworkManager. I don't think it'll cause problems, but I'd like to have some test prior to cutting them all over.
It is possible to stage this in by creating a new NetworkManager enabled node-type. I've proposed that in [1] but it's only useful if you want to then follow-up with setting up testing jobs to use the new node-type. We can then revert and apply the change to regular nodes. By just switching directly in [2], we can quite quickly revert if there should be an issue. We can immediately delete the new image, revert the config change and then worry about fixing it. Staging it is the conservative approach and more work all round but obviously safer; hoping for the best with the escape hatch is probably my preferred option given the low risk. I've WIP'd both reviews so just let us know in there your thoughts. Thanks, -i [1] https://review.openstack.org/629680 [2] https://review.openstack.org/619960
On Wed, Jan 9, 2019 at 4:26 PM Ian Wienand <iwienand@redhat.com> wrote:
On Wed, Jan 09, 2019 at 09:02:57AM -0700, Alex Schultz wrote:
Don't suppose we could try this with tripleo jobs prior to cutting them all over could we? We don't use NetworkManager and infact os-net-config doesn't currently support NetworkManager. I don't think it'll cause problems, but I'd like to have some test prior to cutting them all over.
It is possible to stage this in by creating a new NetworkManager enabled node-type. I've proposed that in [1] but it's only useful if you want to then follow-up with setting up testing jobs to use the new node-type. We can then revert and apply the change to regular nodes.
By just switching directly in [2], we can quite quickly revert if there should be an issue. We can immediately delete the new image, revert the config change and then worry about fixing it.
Staging it is the conservative approach and more work all round but obviously safer; hoping for the best with the escape hatch is probably my preferred option given the low risk. I've WIP'd both reviews so just let us know in there your thoughts.
For us to test I think we just need https://review.openstack.org/#/c/629685/ once the node pool change goes in. Then the jobs on that change will be the NetworkManager version. I would really prefer testing this way than possibly having to revert after breaking a bunch of in flight patches. I'll defer to others if they think it's OK to just land it and revert as needed. Thanks, -Alex
Thanks,
-i
[1] https://review.openstack.org/629680 [2] https://review.openstack.org/619960
On Wed, Jan 9, 2019 at 4:34 PM Alex Schultz <aschultz@redhat.com> wrote:
On Wed, Jan 9, 2019 at 4:26 PM Ian Wienand <iwienand@redhat.com> wrote:
On Wed, Jan 09, 2019 at 09:02:57AM -0700, Alex Schultz wrote:
Don't suppose we could try this with tripleo jobs prior to cutting them all over could we? We don't use NetworkManager and infact os-net-config doesn't currently support NetworkManager. I don't think it'll cause problems, but I'd like to have some test prior to cutting them all over.
It is possible to stage this in by creating a new NetworkManager enabled node-type. I've proposed that in [1] but it's only useful if you want to then follow-up with setting up testing jobs to use the new node-type. We can then revert and apply the change to regular nodes.
By just switching directly in [2], we can quite quickly revert if there should be an issue. We can immediately delete the new image, revert the config change and then worry about fixing it.
Staging it is the conservative approach and more work all round but obviously safer; hoping for the best with the escape hatch is probably my preferred option given the low risk. I've WIP'd both reviews so just let us know in there your thoughts.
For us to test I think we just need https://review.openstack.org/#/c/629685/ once the node pool change goes in. Then the jobs on that change will be the NetworkManager version. I would really prefer testing this way than possibly having to revert after breaking a bunch of in flight patches. I'll defer to others if they think it's OK to just land it and revert as needed.
Looks like it should be ok based on the test results from https://review.openstack.org/#/c/629685/. Thanks, -Alex
Thanks, -Alex
Thanks,
-i
[1] https://review.openstack.org/629680 [2] https://review.openstack.org/619960
On Mon, Jan 14, 2019 at 07:51:07AM -0700, Alex Schultz wrote:
Looks like it should be ok based on the test results from https://review.openstack.org/#/c/629685/.
Thanks; just to confirm this is now the default so CentOS and Fedora images now have NetworkManager in the base image. You can report any problems here or in #openstack-infra. Cheers, -i
participants (2)
-
Alex Schultz
-
Ian Wienand