Wesley Hayutin whayutin at redhat.com
Tue Nov 12 17:13:39 UTC 2019


Thanks for all those Red Hatters who attended the OpenStack Summit and PTG
in Shanghai!

A special thanks to those who presented their topics and discussed work
items with the folks in attendance.  As the current PTL for TripleO I will
do my best here to summarize those conversations and items others should be
made aware of.

Over the course of Thursday and Friday roughly 7-10 folks discussed the
identified topics [1], at the following times with my raw notes attached
[2].  My apologies if I did not accurately represent your topic here,
please feel free to correct me.

Thursday, Giulio Fidente: Edge ( SDS storage )

Guilio walked us through some background work with regards to support for
storage in remote sites / edge deployments.  Working through support for
Cinder was straight forward enough with no real collaboration required.
Support for ceph copy on write for nova guests was also added with the
glance image added to remote sites.  Where Guilio needed input was with
regards to having change the ctrl plane config for glance for each remote
site [3]. This ctrl plane update would force operators to put the cloud in
maintenance mode for a stack update.  It was determined this could not be
avoided at this time. It was noted that the TripleO simplification project
and rework puppet-apply, please help us achieve that by reviewing the
following two topics [4][5]. Thanks Giulio!

Thursday, Guilio Fidente: Virtual IP / Storage

Guilio walked us through some challenges with hosting a shared file system
on remote/edge sites using manilla.  The idea was to use Ganesha
translation with CephFS. The proposal was that Ganesha and pacemaker would
be managed in the ctrl plane but there was a question with regards to the
virtual ip on edge sites.  This was an interesting conversation that ended
up with a suggestion from Kevin Carter to use a host-only local route on
the edge to properly route the ip. This seemed to everyone to be a very
clever solution to the problem :)  Thanks Guilio, Thanks Kevin!

Thursday, Martin Schuppert: Nova CellV2 multicell

Martin walked the group through the current status and plans for the
multicell implementation.

Background: Nova multicells are used to help scale a cloud and partition it
in such a way to get the messaging queue closer to the compute cell,
essentially rabbit, galera, collector, vnc proxy and a number of compute
nodes.  This architecture is already in use but with only one default cell,
pike was the switch to cellv2.

The work started in Stein and continued through train using a similar
approach as DCN.  Some of the specs are that there is one cell per stack
that is initially created from an export of the central stack, more ansible
is place for the deployment as well.  Two different architectures were
noted, all cells in one heat stack [6], and one that splits the cell
controllers and computes into different heat stacks w/ multiple stacks on
the edge sites [7].   The development work for updates is complete and
upgrades is still a WIP.

Plans for the future included integrating TLS everywhere and enabling
storage in the cell ( cinder, ceph, glance).  Tony Breeds pointed out this
architecture should just work in multiarch but would like the teams help in
designing / advice while creating a test environment.

Please review the following patches [13]

Thanks Martin!!

We tried to get more folks to switch to their topics to Thursday but were
not able to.  On to Friday.

Friday, Edge ( DCN ) roadmap: David Paterson

This conversation was informally walked through Thursday mainly with Arkady
and Guilio and was followed up on Friday with a joint edge session
regarding booting edge nodes. Several questions were raised on Thursday
regarding the networking and connectivity for edge sites as it relates to
provisioning.  Validations were discussed as a way to address the minimum
requirements for booting edge nodes. David did not end up presenting here,
but was available at the joint session. See the “edge booting” section
later in the document for details.

Friday, Backup and Restore: Carlos Camacho

The project started in Newton. Initially the backup consisted of a database
dump and files being backed up for a defined set of use cases.  In the
field it was discovered that customers had many different kinds of
deployments and the feature did not work well for all the intended use
cases.   An improved plan included to move to full disk image backups
utilizing REAR [8]. Carlos also noted that customers are now trying to use
( or misuse ) this feature to perform baremetal to virt migrations.   One
of the current issues with the current solution include that it’s not clear
how services behave after backup and restore.. E.G. OSD mons. Wes Hayutin
noted that we have an opportunity to test the full image backup and restore
solution by moving to a more image based internal CI system currently being
designed by Jesse Pretorious and others.

Thanks Carlos!!

Friday, Failure Domains: Kevin Carter

Unfortunately Kevin was in high demand across PTG events and was unable to
present this topic.  This should be discussed in a mid-cycle ( virtual or
in person ) and written up as a blueprint.   Essentially Kevin is proposing
in large deployments to allow some number or percentage of nodes to fail a
deployment while not failing the entire deployment.  If a few non-critical
nodes fail a large scale deployment TripleO should be better able to handle
that, report back and move on. It was pointed out to me there is a related
customer bug <https://bugzilla.redhat.com/show_bug.cgi?id=1721089> as well.

Thanks Kevin!!

Friday, Cross project:  Edge Booting: Julia Kreger

You can find notes on this session here [9].  I will only summarize the
questions proposed on earlier edge ( DCN ) topic.  With regards to when
does TripleO need to support redfish there was no immediate or extremely
urgent requests ( please correct me if I do not have the correct
information there).  Redfish IMHO did seem to be a nice improvement when
compared to IPMI.

This was my first introduction to Redfish, and I of course curious what
steps we had to take in order to CI it.  Luckily after doc diving I found
several helpful links that include steps with setting Redfish up with our
own OVB tooling ( hooray \0/ ).  Links can be found here [10], and it seems
like others have done some hard work to make that possible so thank you!!

Thank you Julia!!

Friday, Further TripleO Ansible Integration: Wes Hayutin

The idea here would be allow the TripleO project to govern how TripleO is
deployed with ansible as an operator.  The TripleO project would ship
ansible modules and roles that directly import python-tripleoclient to
support ansible to cli parity [12].  Using or modeling a new repo, perhaps
called tripleo-operator-ansible would be used to host these modules and
roles and include the same requirements and features of tripleo-ansible’s
linting, molecule tests, and auto documentation.   This could tie in well
with an initiative from Dan Macpherson to ship Ansible playbooks as part of
our OSP documentation. Julia Kreger noted that we should not ignore the
Ansible OpenStackSDK for part of the deployment process which is a very
valid point.  Most everyone at the PTG agreed this was a good direction
moving forward and would help consolidate the public and internal tooling
around TripleO’s CLI in ansible.

Thanks Dan, Julia!!

Friday, TLS-E standalone gate: Ade Lee

Ade Lee walked us through a proposal to help test and CI TLS upstream which
has been very difficult to date ( I can personally vouch for this ).  Using
a two node setup upstream with one node as the IPA server and the other a
TripleO standalone deployment. The keystone team is setting the right
example for other projects and teams that are finding it difficult to keep
outside patches from breaking their code, and that is to find a way to get
something voting and gating upstream even if it’s not installed and
deployed in the same exact ways customers may use it.

Please help by reviewing the keystone / security teams patches here [14]

Thanks Ade!!

Friday, Octavia tempest plugin support in TripleO-CI: Chandan Kumar

Chandan was off fighting battles with the infra team and other projects.
Here are some of his notes:

Have a rdo third party standalone job with full octavia tempest triggered
against octavia patches from stein onwards. FS062

Look into multinode job for queens releases as a third party from queens
and rocky. FS038

Have a support of octavia tempest plugin in os_tempest.

We certainly should have a conversation offline regarding these topics.
I’ll note the TripleO-CI community meeting or the #tripleo meeting on
Tuesdays are a good way to continue collaborating here.

Thanks Chandan!!

Friday, Replace clouds.yaml with an openrc ansible module: Chandan Kumar

Open Question:  is this module [15] from the openstack-ansible project
something we can reuse in TripleO via tripleo-ansible?

Friday, Zuul jobs and ansible roles to handle rpm packaging: Chandan Kumar

The background and context can be found here

https://pagure.io/zuul-distro-jobs - collection of ansible roles to deal
with rpm packaging

-> creating rpms from different projects depends-on in commit message


Move it to zuul-jobs

Making the rpm packaging more generic for centos/fedora/rhel and move it to
zuul jobs

* Move mock and rpmbuild related roles to zuul jobs repo

* Adding a mention of third party zuul jobs to main zuul jobs doc

* Build-set_registry setup http server and start the job

* Details are here

Thanks Chandan!! This is indeed a very interesting and powerful proposal.
We should definitely continue this conversation with the broader community.

Did you make it all the way down here?  Well done! I should add an easter
egg :)


[1] https://etherpad.openstack.org/p/tripleo-ussuri-topics

[2] https://etherpad.openstack.org/p/tripleo-ptg-ussuri


[4 ]





[8] https://access.redhat.com/solutions/2115051

[9] https://etherpad.openstack.org/p/PVG-ECG-PTG

[10] https://github.com/openstack/sushy




[12] https://hackmd.io/caRlGha7SueZxDRcyq9eGA?both



[15] https://opendev.org/openstack/openstack-ansible-openstack_openrc

[16] https://etherpad.openstack.org/p/PVG-keystone-forum-policy

[17 https://datko.pl/zuul.pdf


[19] https://github.com/openstack/openstack-virtual-baremetal

Thanks all!!

Wes Hayutin

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20191112/da52952f/attachment-0001.html>

More information about the openstack-discuss mailing list