Comment 10 for bug 1367826

Reviewed: https://review.openstack.org/121422
Committed: https://git.openstack.org/cgit/openstack/swift/commit/?id=20e9ad538b6ba90674a75310d9cc184162ba9398
Submitter: Jenkins
Branch: master

commit 20e9ad538b6ba90674a75310d9cc184162ba9398
Author: Christian Schwede <email address hidden>
Date: Sun Sep 14 20:48:32 2014 +0000

    Limit partition movement when adding a new tier

    When adding a new tier (region, zone, node, device) to an existing,
    already balanced ring all existing partitions in the existing tiers of
    the same level are gathered for reassigning, even when there is
    not enough space in the new tier. This will create a lot of unnecessary
    replication traffic in the backend network.

    For example, when only one region exists in the ring and a new region is
    added, all existing parts are selected to reassign, even when the new
    region has a total weight of 0. Same for zones, nodes and devices.

    This patch limits the number of partitions that are choosen to reassign
    by checking for devices on other tiers that are asking for more
    partitions.

    Failed devices are not considered when applying the limit.

    Co-Authored By: Florent Flament <email address hidden>
    Change-Id: I6178452e47492da4677a8ffe4fb24917b5968cd9
    Closes-Bug: 1367826