Comment 10 for bug 1786703

Reviewed: https://review.openstack.org/592654
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=2d7ed309ec4e656ce9d6f21f03ea158278f2526d
Submitter: Zuul
Branch: master

commit 2d7ed309ec4e656ce9d6f21f03ea158278f2526d
Author: Jay Pipes <email address hidden>
Date: Thu Aug 16 14:56:47 2018 -0400

    placement: use single-shot INSERT/DELETE agg

    When replacing a provider's set of aggregate associations, we were
    issuing a call to:

     DELETE resource_provider_aggregates WHERE resource_provider_id = $rp

    and then a single call to:

     INSERT INTO resource_provider_aggregates
     SELECT $rp, aggs.id
     FROM provider_aggregates AS aggs
     WHERE aggs.uuid IN ($agg_uuids)

    This patch changes the _set_aggregates() function in a few ways.
    First, we grab the aggregate's internal ID value when creating new
    aggregate records (or grabbing a provider's existing aggregate
    associations). This eliminates the need for any join to
    provider_aggregates in an INSERT/DELETE statement.

    Second, instead of a multi-row INSERT .. SELECT statement, we do
    single-shot INSERT ... VALUES statements, one for each added aggregate.

    Third, we no longer DELETE all aggregate associations for the provider
    in question. Instead, we issue single-shot DELETE statements for only
    the aggregates that are being disassociated.

    Finally, I've added a number of log debug statements so that we can have
    a little more information if this particular patch does not fix the
    deadlock issue described in the associated bug.

    Change-Id: I87e765305017eae1424005f7d6f419f42a2f8370
    Closes-bug: #1786703