yahoo-eng-team team mailing list archive

["Markus Zoeller \(markus_z\)" <mzoeller@xxxxxxxxxx>]

Sean Dague's quote from the abandoned review [1]:

> Booting an instance is expensive, and triggers io load, which may
> increase failure rates of these booting. Doing in parallel is not
> really guarunteed to be faster than serial, and in some situations
> will actually be slower.
> 
> So this isn't a simple bug fix. It's something which we really should
> have a spec for. There definitely has to be a max number which 
> is < 200. It should also have some real world boot data to see how
> this plays out in real world situations.

I'm closing it as "Opinion/Wishlist". If you decide to work on this
consider using a blueprint [2] (with a spec [3]). I'll recommend to 
read [4] if not yet done. 

If there are any questions left, feel free to contact me (markus_z)
in the IRC channel #openstack-nova

References:
[1] https://review.openstack.org/#/c/223572/
[2] https://blueprints.launchpad.net/nova/
[3] https://github.com/openstack/nova-specs
[4] https://wiki.openstack.org/wiki/Blueprints

** Changed in: nova
       Status: In Progress => Opinion

** Changed in: nova
   Importance: Undecided => Wishlist

** Changed in: nova
     Assignee: Rui Chen (kiwik-chenrui) => (unassigned)

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/1496334

Title:
  Nova-compute launch slowly because lots of instances is init one by
  one

Status in OpenStack Compute (nova):
  Opinion

Bug description:
  1. code base

  $ git log -1
  commit b492942744e09276e3ba4dcf0196143c521a1662
  Merge: 920abc9 9706454
  Author: Jenkins <jenkins@xxxxxxxxxxxxxxxxxxxx>
  Date:   Thu Sep 3 00:05:04 2015 +0000

      Merge "Fix bodies on consolidate-console-api"

  2. Reproduce steps:

  The issue happen on VMware driver, think about the following case:
  * 200 active instances run in one nova-compute host that map to one vCenter Cluster, batch delete all instances, all of them are in "deleting" task_state.
  * nova-compute process stop and restart when all instances are in "deleting" task_state.
  * nova-compute start to init 200 deleting instances one by one. The workflow of VMware driver is power-off instance, then wait task finish, then delete the instance.
  * After all the deleting instances are handled, nova-compute is set to "up" state, continue to work.

  step 3 will spend lots of time on serial init_instance. In my
  performance test environment, the nova-compute spend about 15 minutes
  to finish init_instance.

  In other drivers, like: libvirt, nova-compute manage less instances
  than VMware driver (maybe less than 50 instances), so these drivers
  have less chance to face the issue.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1496334/+subscriptions