yahoo-eng-team team mailing list archive

[Sean Dague <sean@xxxxxxxxx>]

So this is an interesting possible feature, but that means it should
come in via a spec, not really via launchpad.

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

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

-- 
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/1373478

Title:
  filter scheduler makes invalid assumption of monotonicity

Status in OpenStack Compute (Nova):
  Opinion

Bug description:
  The current filter scheduler handles the scheduling of a homogenous
  batch of N instances with a loop that assumes that a host ruled out in
  one iteration can not be desirable in a later iteration --- but that
  is a false assumption.

  Consider the case of a filter whose purpose is to achieve balance
  across some sort of areas.  These might be AZs, host aggregates,
  racks, whatever.  Consider a request to schedule 4 identical
  instances; suppose that there are two hosts, one in each of two
  different areas, initially hosting nothing.  For the first iteration,
  both hosts pass this filter.  One gets picked, call it host A.  On the
  second iteration, only the other host (call it B) passes the filter.
  So the second instance goes on B.  On the third iteration, both hosts
  would pass the filter but the filter is only asked about host B.  So
  the third instance goes on B.  On the fourth iteration, host B is
  unacceptable but that is the only host about which the filter is
  asked.  So the scheduling fails with a complaint about no acceptable
  host found.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1373478/+subscriptions