yahoo-eng-team team mailing list archive

[Sean Dague <sean@xxxxxxxxx>]

This isn't really a bug. If you have a reproduce with an expected
behavior and actually behavior maybe we can turn it into a bug.

** Changed in: nova
       Status: Triaged => Opinion

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

** Tags added: scheduler

-- 
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/1161664

Title:
  Rescheduling can DDOS itself

Status in OpenStack Compute (Nova):
  Opinion

Bug description:
  Due to the way nova currently handles rescheduling there is a
  tendency, when a large number of compute nodes need to reschedule (for
  whatever error) that they then swamp the message queue (and nova
  scheduler) with rescheduling messages. This can cascade to where
  further rescheduling messages will occur (and repeat...) or until the
  MQ piles up and/or the scheduler falls over.

  Even with a reschedule 'count' under situations when rescheduling is
  happening on mass the rescheduling itself can cause more problems for
  your system than it helps solve (aka, just leave the scheduled
  instance in error state). Likely the way to do this in a more
  centralized manner (aka with a orchestration unit that can do this
  rescheduling on behalf of the request) can help rate limit itself and
  its requests to the scheduler for new locations to schedule to. Having
  each compute node perform this same operation means rate limiting is
  not possible (and allows for your own system to DDOS itself).

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1161664/+subscriptions