[stable][glance] backport exception request for external HTTP API response code change

Dan Smith dms at danplanet.com
Thu May 19 22:05:00 UTC 2022

> We recently noticed that an Image API call implemented in Yoga, which
> returns no content, is returning a 200 instead of a 202 (as the design
> [0] had specified).  We would like to correct this in master (Zed
> development) and in the stable/yoga branch.  Here are our reasons why
> we think this is a legitimate exception to the
> no-http-api-changes-backport policy:
> (1) Even ignoring the lack of a response body, a 200 (OK) is
> misleading because operation initiated by the PUT /v2/cache/{image_id}
> request may not succeed.  A 202 (Accepted) signals more clearly to the
> operator that some followup is necessary to be sure the image has been
> cached.

...may not succeed, will be processed in the background, and may never
actually happen. In other words, the textbook definition of 202 :)

> (2) This is intended as an admin-only API call; the default policy is
> admin-only.  So it has not been exposed to end-users.
> (3) The change [1] was first released in Yoga and there has not been
> much time for admins to begin consuming this feature.  Further, there
> has not yet been a release from stable/yoga after glance 24.0.0.
> For these reasons, we request that the glance team be allowed to make
> this change.

Yep, my feeling is that we caught this very early and that it's a very
limited-scope operation which means the number of people potentially
impacted is very small. It's unlikely many people are yet running Yoga
in production, and of those, only some admins and scripts could be
affected, and only those that were super excited to use this brand new
rather minor feature right away. It's an API for a function that was
already being handled by a different tool, so people would have had to
*convert* their existing stuff to this already, which is even less
likely than some new exciting functionality that didn't exist before.

I think the calculus comes down to:

If we do backport it, a small number of super bleeding-edge admins *may*
have already jumped on writing against this API and may notice. The vast
majority of people that end up deploying Yoga will never experience

If we don't backport it, we'll ensure that many more people will be
affected. People _will_ eventually deploy Yoga, and then they _will_
deploy Zed. Those people _will_ experience a change, and I think it's
clear that this class of people is larger than the one described above.

It sucks, but I think it's less pain overall to backport ASAP, reno the
mea culpa clearly, and limit the number of affected people to the much
smaller number.


More information about the openstack-discuss mailing list