Comment 9 for bug 928910

Reviewed: https://review.openstack.org/4061
Committed: http://github.com/openstack/nova/commit/1c8ad4553b4b8d404f941c5297e3f6e42c9f7e6a
Submitter: Jenkins
Branch: master

commit 1c8ad4553b4b8d404f941c5297e3f6e42c9f7e6a
Author: Jay Pipes <email address hidden>
Date: Sun Feb 12 13:34:14 2012 -0500

    Completes fix for LP #928910 - libvirt performance

    This patch adds the remainder of the recommended fixes
    from the original bug report:

    * Modifies methods in the compute manager that relied on
      the DB power state to be in sync with the virt driver to
      instead just query the power state of the instance from the
      virt driver. This enables us to set the periodic tick to 10
      for the problematic compute.manager.Manager._sync_power_states()
      method.
    * Modifies the _sync_power_states method in the following ways:
     ** Replace the call to driver.list_instances_detail() to a new,
        driver-overrideable get_num_instances() call
     ** For each instance known by the database, call driver.get_info()
        separately inside the loop instead of calling the expensive
        list_instances_detail() method that can take a very long time
        to complete on hosts with lots of instances
     ** Call greenthread.sleep(0) before each call to update the
        database power state, enabling other periodic tasks to do work

    Once again, I left an inefficient default implementation of the
    new driver.get_num_instances() method in the base driver class. I
    need help from folks who understand the Xen/VMWare drivers to do
    an override for get_num_instances() in those drivers that calls
    the underlying XenAPI or VMWare API.

    Change-Id: I88002689cdda32124423da320f8c542e286be51b