auto-select rpc version info logging is too chatty

[Matt Riedemann]

This shows up 392 this in the n-api log run here:

"Automatically selected compute RPC version 4.5 from minimum service version 2"

http://logs.openstack.org/92/253192/6/check/gate-grenade-dsvm/76ad561/logs/new/screen-n-api.txt.gz#_2015-12-09_16_50_42_905

About 1.3 million times in gate runs in 24 hours:

http://logstash.openstack.org/#dashboard/file/logstash.json?query=message:%5C%22Automatically%20selected%20compute%20RPC%20version%5C%22%20AND%20message:%5C%22RPC%20version%5C%22

We should be smarter about logging that, or it could indicate there is something wrong here in how often we're looking that up.

[Matt Riedemann (mriedem)]

https://review.openstack.org/#/c/255404/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/255404
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=242ee4b5e774ad12a001e5a1482fc4c2e592615f
Submitter: Jenkins
Branch: master

commit 242ee4b5e774ad12a001e5a1482fc4c2e592615f
Author: Dan Smith <email address hidden>
Date: Wed Dec 9 10:01:05 2015 -0800

    Cache the automatic version pin to avoid repeated lookups

    The compute rpcapi module was looking up the desired automatic
    version pin each time it was initialized. Turns out, we create and
    destroy this object too often, which means we do a lot of those
    lookups unnecessarily. That's a larger problem that will be harder
    to solve, but we can be smarter about the DB impact of that.

    This adds a cache of the last version we discovered and avoids doing
    the lookup when possible. It makes the reset() method invalidate the
    cache prior to reinitializing the object so that we will do the lookup
    on SIGHUP as desired.

    Closes-Bug: #1524444
    Change-Id: Ia1c58efa7105d32973f6921c2144b17c30b764c2

[Thierry Carrez (ttx)]

This issue was fixed in the openstack/nova 13.0.0.0b2 development milestone.