exception tracebacks are lost when using raise

[Vish Ishaya]

Since switching the db code to use an eventlet library, we can no longer re-raise exceptions if there is any db access in the exception handler. There are a number of places in the code where this is done.

we need to change instances of

except <SomeException>:
  # code that uses db or rpc
  raise

to:

except <SomeException>:
  with excutils.save_and_reraise_exception():
    # code that uses db or rpc

Also adding a logging.exception('useful message') is probably a good idea in most cases if the original traceback is important.

[Brian Lamar (blamar)]

Arg, this is unfortunate. This whole eventlet + SQLAlchemy thing has been a pain...but I'm not sure how the rest of the world deals with it. Maybe I'll make a summit brainstorming session over this because it's been the source of a number of bugs.

[Russell Bryant (russellb)]

Added oslo as this affects impl_zmq

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/23565

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/23566

[Johannes Erdfelt (johannes.erdfelt)]

This is actually a bug in eventlet. Here's the issue (and bug fix) for it:

https://bitbucket.org/eventlet/eventlet/issue/129/exception-info-clearing-is-problematic-and

Unfortunately, it hasn't been touched in months :(

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/23565
Committed: http://github.com/openstack/oslo-incubator/commit/f32af7b42e7801338583f67e794792f22df70a5d
Submitter: Jenkins
Branch: master

commit f32af7b42e7801338583f67e794792f22df70a5d
Author: Russell Bryant <email address hidden>
Date: Tue Mar 5 10:50:15 2013 -0500

    Preserve exceptions in impl_zmq.

    There is complication with re-raising exceptions and our usage of
    eventlet. If the code in the exception handler accesses the db or rpc
    in the exception handler, it will no longer be able to re-raise the
    exception. Using excutils.save_and_reraise_exception() works aorund
    this issue.

    The most common error is calling LOG.error() or LOG.exception(), as it
    is possible for these to go access rpc. There is an option to turn on
    notifications for these errors.

    Fix bug 845866.

    Change-Id: Ic914bba4703200ed114a42e3a55402883b430407

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/23566
Committed: http://github.com/openstack/nova/commit/2f983793574de07269c4a79852d75a9059465ee0
Submitter: Jenkins
Branch: master

commit 2f983793574de07269c4a79852d75a9059465ee0
Author: Russell Bryant <email address hidden>
Date: Tue Mar 5 10:55:18 2013 -0500

    Fix issues with re-raising exceptions.

    There is complication with re-raising exceptions and our usage of
    eventlet. If the code in the exception handler accesses the db or rpc
    in the exception handler, it will no longer be able to re-raise the
    exception. Using excutils.save_and_reraise_exception() works aorund
    this issue.

    The most common error is calling LOG.error() or LOG.exception(), as it
    is possible for these to go access rpc. There is an option to turn on
    notifications for these errors.

    Fix bug 845866.

    Change-Id: Icfca2af63805711229249aa7abe60a938edd1b35