Cinder client use has no timeout

[Brian Elliott]

Nova-api calls cinder in a synchronous fashion on volume-list and snapshot-list calls. There is a baked in param defaulting to 3 retries, but no ability to control the timeout on each attempt. In order to achieve more predictable nova-api response times, a timeout should be configurable.

[Matt Riedemann (mriedem)]

https://review.openstack.org/#/c/99211/

[Matt Riedemann (mriedem)]

If cinder api is underperforming maybe you need more osapi_volume workers, see this patch:

https://review.openstack.org/#/c/98492/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/99211
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=618ae7a4392dce4359c47c7e4b60bebc1c8d853b
Submitter: Jenkins
Branch: master

commit 618ae7a4392dce4359c47c7e4b60bebc1c8d853b
Author: Brian Elliott <email address hidden>
Date: Tue Jun 10 16:23:27 2014 -0500

    Add configurable HTTP timeout to cinder API calls

    Some Nova API calls require synchronous use of cinder. (e.g. volume
    list) If the call to cinder lands on a slow performing or otherwise bad
    cinder API node, the user of Nova API will see poor performance.

    This patch adds the ability to control HTTP timeouts when calling the
    cinder API. By configuring a timeout, nova will be able to more quickly
    retry its cinder request and hopefully land on a better performing
    cinder API node. This will help make the API response times more
    predictable.

    For backward compatibility, the default timeout value is None.

    Closes-Bug: 1329797
    DocImpact

    Change-Id: I5cccced524d9a9d69ed41f755781d0b7ae3dc970