Potential codec errors during rescheduling operations

[Joe Cropper]

In the compute manager's 'retry' logic (i.e., handling RescheduledException) in _build_and_run_instance, it has a catch-all case for all exceptions and sets reason=str(e). This works well in most cases, but for cases in which lower level code may generate locale-specific messages, it's possible to see the "UnicodeEncodeError: 'ascii' codec can't encode character..." error, which ultimately masks the message in the compute logs, etc.

It also means the instance won't be rescheduled as it fails with an error similar to...

WARNING nova.compute.manager [req-7fe662b1-c947-4303-b43f-114fbdafe875 None] [instance: 333474eb-dd07-4e05-b920-787da0fd5b32] Unexpected build failure, not rescheduling build.
(i.e., as a result of the UnicodeEncodeError previously mentioned)

The simple solution is to use six.text_type(e) instead of str(e).

[Davanum Srinivas (DIMS) (dims-v)]

Looks like the review is this one - https://review.openstack.org/#/c/115846/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/115846
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=9f1f9a8b6fadbba2eb6733e0f13b727f14a52a3f
Submitter: Jenkins
Branch: master

commit 9f1f9a8b6fadbba2eb6733e0f13b727f14a52a3f
Author: Joe Cropper <email address hidden>
Date: Wed Aug 20 21:18:17 2014 -0500

    Handle non-ascii characters in spawn exception msg

    If a compute driver's spawn function returns an exception whose text
    has any non-ascii characters (e.g., for globalization), the compute
    manager's "reschedule logic" doesn't handle it properly as it runs
    the text through the str() function, which causes UnicodeEncodeError
    exceptions.

    This simply sends the exception text through six.text_type so as to
    no longer blow up and proceed with rescheduling as intended.

    Closes-Bug: #1359427

    Change-Id: I3fef783feffbe99c6971543c881e3fce2b60693b