Redis driver fails to claim messages (for a subset of requests) under high concurrency

[Kurt Griffiths]

With 1000 message producers + 1000 consumer workers trying to claim simultaneously across two redis instances in my bare metal test rig, some claims start to fail because they reach the maximum retry duration. Adding a backoff sleep ala the Mongo driver helps a lot (I did a quick hack to try it), but I'd like to explore using Lua as well to fix this.

[Flavio Percoco (flaper87)]

Lets make sure we check the redis version on startup as we check mongodb's version. We should prevent the service from starting if the minimum required version is not met

[Kurt Griffiths (kgriffs)]

Good idea, makes sense.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/123230

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/123230
Committed: https://git.openstack.org/cgit/openstack/zaqar/commit/?id=72598b23891ce4c9e9c5583e8177d99be0f56eb4
Submitter: Jenkins
Branch: master

commit 72598b23891ce4c9e9c5583e8177d99be0f56eb4
Author: kgriffs <email address hidden>
Date: Mon Sep 22 14:27:38 2014 -0500

    Move Redis driver's claim transaction to Lua

    This avoids having to set a watch and retry if someone else beats us to
    the punch; the retry is less efficient than doing it with Lua, and would
    require adding a backoff sleep to avoid lots of failed requests under
    high concurrency.

    This patch also adds a Redis server version check in the driver since
    Lua requires redis-server >= 2.6

    Change-Id: I0c34e267a6358965a54179a4b2fd03ac15b194e9
    Closes-Bug: #1372510