Consider 3.13 kernel for master node

Bug #1322641 reported by Matthew Mosesohn
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Fuel for OpenStack
5.1.x
Won't Fix
High
Matthew Mosesohn

Bug Description

This should be self-evident that new kernel runs docker and LXC commands much better.

2.6.32 kernel:
Loading docker image astute.tar...
Loading docker image busybox.tar...
Loading docker image cobbler.tar...
Loading docker image mcollective.tar...
Loading docker image nailgun.tar...
Loading docker image nginx.tar...
Loading docker image ostf.tar...
Loading docker image postgres.tar...
Loading docker image rabbitmq.tar...
Loading docker image rsync.tar...
Loading docker image rsyslog.tar...

real 3m0.653s
user 0m1.154s
sys 0m9.887s

3..10 kernel:
# time for image in *tar ; do echo "Loading docker image ${image}..."; docker load -i "$image"; done
Loading docker image astute.tar...
Loading docker image busybox.tar...
Loading docker image cobbler.tar...
Loading docker image mcollective.tar...
Loading docker image nailgun.tar...
Loading docker image nginx.tar...
Loading docker image ostf.tar...
Loading docker image postgres.tar...
Loading docker image rabbitmq.tar...
Loading docker image rsync.tar...
Loading docker image rsyslog.tar...

real 1m27.304s
user 0m0.670s
sys 0m13.674s

50% performance boost is worth it to get master node provisioned more quickly.

We should see what real blockers there are here. We could add an install flag for legacy kernel and document it for those rare use cases.

Changed in fuel:
importance: Undecided → High
status: New → Triaged
Changed in fuel:
assignee: Fuel Library Team (fuel-library) → Matthew Mosesohn (raytrac3r)
Revision history for this message
OpenStack Infra (hudson-openstack) wrote : Fix proposed to fuel-main (master)

Fix proposed to branch: master
Review: https://review.openstack.org/100501

Changed in fuel:
status: Triaged → In Progress
Revision history for this message
Roman Alekseenkov (ralekseenkov) wrote : Re: Consider 3.10 kernel for master node

This change is too big for 5.0.1.

Upgrading to a new kernel one week before the release is (a) way too risky, (b) out of scope for a point/maintenance release, and (c) has a chance of delaying the release which we don't want to happen.

Moreover, migration to a new kernel needs to happen consistently across all components:
1) master node
2) bootstrap
3) centos (target nodes)
4) ubuntu (slave nodes)

And a new kernel should be 3.11, not 3.10 (it has been confirmed that 3.10 still has issues, while 3.12, 3.13 and 3.14 are not yet stable). 3.11 is a known good version according to multiple sources - community, HP, Alex Shaposhnikov, etc.

Since this is a performance improvement and not a blocker, my suggestion is to:
1) Document as a known performance issue for 5.0.1. Close as "won't fix" for 5.0.1
2) Address it in 5.1 in a consistent way (as I mentioned above)

tags: added: release-notes
summary: - Consider 3.10 kernel for master node
+ Consider 3.11 kernel for master node
Revision history for this message
Matthew Mosesohn (raytrac3r) wrote : Re: Consider 3.11 kernel for master node

Roman, it's only 2 days from 5.1 Feature Freeze and nobody has even considered implementing 3.11. We should decide today to move forward with 3.10 for CentOS on Master node (packages should now be ready) or to abandon these plans and wait for 5.2 or 6.0 for 3.11. I don't think it's nearly possible for 5.1 now. We're giving up a quick win for performance gains in deploying Fuel Master in order to reach something we're not yet ready for.

Mike Scherbakov (mihgen)
Changed in fuel:
status: In Progress → Won't Fix
no longer affects: fuel
Revision history for this message
Dmitry Borodaenko (angdraug) wrote :

Should we rename this to consider 3.13 now?

Revision history for this message
Aleksander Mogylchenko (amogylchenko) wrote :

3.13 kernel is being tested at the moment.

summary: - Consider 3.11 kernel for master node
+ Consider 3.13 kernel for master node
no longer affects: fuel/5.0.x
Revision history for this message
Mike Scherbakov (mihgen) wrote :

It is too late for 6.0 release. We are about to call HCF, while this change is considered to be a huge change for master node - so it had to be done before Feature Freeze, a month ago.
If we change the kernel now, I'm pretty sure we will end up with Fuel upgrade issues, and there is a high risk for other defects to occur.

no longer affects: fuel
Revision history for this message
Aleksander Mogylchenko (amogylchenko) wrote :

This will be postponed till we have proper separation between master and slave repos.

no longer affects: fuel/6.0.x
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.