OS API issues when calling network API

[Chris Behrens]

OS API uses instance['info_cache']['network_info'] to use for server details, however it'll fall back to doing network API queries if network_info is None or []. This means that immediate GETs after POST /server will call to the network API.. When using melange and melange is down, CONNREFUSED errors are returned via the API.

We should trust the cache.. Perhaps we should implement some self healing.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/4269

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/4269
Committed: http://github.com/openstack/nova/commit/08fa534a0d28fa1be48aef927584161becb936c7
Submitter: Jenkins
Branch: master

commit 08fa534a0d28fa1be48aef927584161becb936c7
Author: Chris Behrens <email address hidden>
Date: Fri Feb 17 00:42:10 2012 +0000

    Remove network_api fallback for info_cache from APIs

    Fixes bug 932395

    OS API and EC2 would query the network API when instance['info_cache']
    had no network info.. or network info was an empty list. The ideal was
    to fall back to querying the network IP in case the cache was busted.

    However, reality says this case is extremely common as it is the case
    when instances are freshly built and haven't had network info assigned
    yet. The calls to network API are expensive and goes against the whole
    idea of this "cache".

    So, this patch removes the fallback to querying the network API. In its
    place, it adds a periodic task to the compute manager to periodically
    sync the cache entry.

    Since I had to fix a number of tests, I consolidated some things there
    as well with regards to instance stubbing.

    Change-Id: I493f811bcba4e99ac6a5756bcab473557d5c0104