[library] Deployment of new controllers causes services downtime

Bug #1370067 reported by Artem Panchenko
10
This bug affects 2 people
Affects Status Importance Assigned to Milestone
Fuel for OpenStack
Invalid
Medium
Fuel QA Team

Bug Description

Fuel version info: http://paste.openstack.org/show/112149/

Steps to reproduce:

1. Create new cluster (Ubuntu, HA, NovaFlatDHCP, Ceph for images, Sahara, Ceilometer)
2. Add 3 controller+mongo, 4 compute and 4 cinder+ceph nodes (I tested on hardware)
3. Deploy changes. Run long-time (at least concurrency * 100 requests) Rally benchmark test on deployed cloud. I used concurrency = 80, but it could be smaller or greater depending on hardware specifications (cluster should be able to handle all incoming requests without overload/errors).
4. During performance testing add 2 new controllers to the environment and deploy changes

Expected result:

- deployment is successful, Rally test ends with 100% success (all requests to nova for create/delete instance were handled)

Actual result:

- deployment is successful, but Rally got a lot of 504/502 errors from HAProxy (Nova-api) and test was aborted:

http://paste.openstack.org/show/112154/

As you can see during few minutes nova-api was unreachable and most of requests from Rally failed. Here you can find the part of syslog on primary controller (node-18):

http://paste.openstack.org/show/112167/

And the part of HAProxy stats:

http://paste.openstack.org/show/112166/

Link to the diagnostic snapshot: https://yadi.sk/d/8yQXZY7EbWrmQ

summary: - [library] Deploying of new controllers cause services downtime
+ [library] Deployment of new controllers causes services downtime
Changed in fuel:
importance: Undecided → Medium
description: updated
Revision history for this message
Irina Povolotskaya (ipovolotskaya) wrote :

Should this be included into Release notes?
Is there any workaround?

Changed in fuel:
status: New → Confirmed
Revision history for this message
Dmitry Borodaenko (angdraug) wrote :

This is a Medium priority bug so it doesn't qualify for release notes. The workaround is to re-run the failed API calls after controller deployment is finished.

tags: added: release-notes
Changed in fuel:
milestone: 6.0 → next
Dmitry Pyzhov (dpyzhov)
Changed in fuel:
milestone: next → 7.0
Revision history for this message
Vladimir Kuklin (vkuklin) wrote :

This bug is one year old - there has been a lot of fixes since then. Marking as incomplete and asking qa guys to recheck this behaviour.

Changed in fuel:
assignee: Fuel Library Team (fuel-library) → Fuel QA Team (fuel-qa)
status: Confirmed → Incomplete
Revision history for this message
Timur Nurlygayanov (tnurlygayanov) wrote :

This bug was in incomplete state more than 1 month, status changed to Invalid.
Please, change the status to actual value if this bug is actual for now and add some comments.

Changed in fuel:
status: Incomplete → Invalid
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.