On 3/9/2016 8:54 AM, Amrith Kumar wrote:
Matt,
As I understand it, you have 4 changes up for review.
Change [1] seeks to revert the change [6] to python-troveclient on
master and reinstate the slave_of parameter.
Change [2] is doing for stable/liberty, the same things that were
done for master in [9] on master, and [7] on Kilo earlier.
Change [3] is looking to blacklist python-troveclient 2.1.0 on
stable/liberty.
Change [4] is looking to bump the minimum python-troveclient version
on master to 2.1.0.
Right, and there was actually another for that beat me to this:
https://review.openstack.org/#/c/290115/
Here are three questions I have:
(1) Wouldn't backward compatibility also require that you revert the
other change that removed slave_of on the server [5] ? After all, just
like a python client that calls create() with slave_of, a REST client
could call the endpoint with slave_of. What is the policy for REST API
backward compatibility?
I guess it depends on the Trove team. In Nova, backward compatibility in
the API is serious business and that's why we have left all sorts of
warts in the v2.0 API and couldn't remove them. But with the v2.1 API
and microversions, we're able to move the API forward (Ironic also has
microversions, and I think cinder/neutron/keystone are working on adding
that support).
So if maintaining backward compatibility in the trove API is important
to the trove team and it's users, then yes the API change should
probably be reverted. For anyone doing CD with Trove, they are already
broken, but people upgrading from liberty to mitaka could save
themselves the break.
(2) Wouldn't [4] just block 2.1.0 in global-requirements on master
(mitaka)?
I'm not sure I understand, [4] here bumps the minimum required version
of python-troveclient to be 2.1.0, not block it. As noted in the review,
I don't know that it's really necessary to bump to 2.1.0 iff we land and
release [1].
(3) Something, potentially your patch set [2], should also fix the test
that is invoking create() with slave_of, no?
[2] is stable/liberty which still has slave_of in the create API, so I
don't think that needs to be fixed in stable/liberty.
-amrith
[1] https://review.openstack.org/290048/
[2] https://review.openstack.org/290023/
[3] https://review.openstack.org/290021/
[4] https://review.openstack.org/290168/
[5] https://review.openstack.org/#/c/245845/
[6] https://review.openstack.org/#/c/245738/
[7] https://review.openstack.org/#/c/246735/
On 03/08/2016 08:24 PM, Matt Riedemann wrote:
On 3/8/2016 4:44 PM, Matt Riedemann wrote:
On 3/8/2016 3:17 PM, Matt Riedemann wrote:
On 3/8/2016 1:18 PM, Amrith Kumar wrote:
Matt,
See inline below.
-amrith
-----Original Message-----
From: Matt Riedemann [mailto:mrie...@linux.vnet.ibm.com]
Sent: Tuesday, March 08, 2016 2:00 PM
To: openstack-dev@lists.openstack.org
Subject: Re: [openstack-dev] [trove][stable] proboscis tests randomly
failing in stable branches; bug 1538506
On 3/8/2016 12:52 PM, Matt Riedemann wrote:
On 3/8/2016 12:35 PM, Amrith Kumar wrote:
Matt,
The correct solution for liberty is that we should fix the tests.
Here's why I believe that this is the case.
In pertinent part, the backtrace from your bug includes:
2016-01-27 07:02:06.777 | File
"/home/jenkins/workspace/periodic-trove-python27-liberty/trove/tests/
api/replication.py",
line 83, in create_slave
2016-01-27 07:02:06.777 | slave_of=instance_info.id)
This is in the tests, not the client.
The test is now generating a parameter that the client no longer
understands.
For a user, here are the various situations that could occur.
1. User running python-troveclient from the latest 2.1.0 and a
server
from liberty. All is good.
2. User running an old python-troveclient and a server from
liberty.
All is good.
Will this work with python-troveclient 1.2.0 which is the minimum
required troveclient for stable/liberty?
https://github.com/openstack/requirements/blob/stable/liberty/global-r
equirements.txt#L174
3. User running an old python-troveclient and a new server from
mitaka, this is not supported.
How is this not supported? If it's not supported, the minimum
required version of troveclient in master g-r is wrong, since it
hasn't changed since liberty:
https://github.com/openstack/requirements/blob/master/global-requireme
nts.txt#L202
I've confirmed that running master (mitaka) unit tests for trove
against
python-troveclient 1.2.0 don't work:
http://paste.openstack.org/show/489719/
[amrith] Just like the bug you listed, this appears to be a case of a
broken test. Would you please LP it.
4. User running a current python-troveclient from the latest 2.1.0
and a server from mitaka, All is good.
What you have here is a case where a test is generating an argument
(slave_of) that the client (master, 2.1.0) no longer understands.
There's nothing amiss there, except that the test needs to be
fixed.
When you get back, let's continue the conversation either in
email or
IRC #openstack-dev. Looking forward to hearing your feedback on
this.
Thanks,
-amrith
-----Original Message-----
From: Matt Riedemann [mailto:mrie...@linux.vnet.ibm.com]
Sent: Tuesday, March 08, 2016 12:11 PM
To: openstack-dev@lists.openstack.org
Subject: Re: [openstack-dev] [trove][stable] proboscis tests
randomly failing in stable branches; bug 1538506
On 3/8/2016 10:17 AM, Matt Riedemann wrote:
This is a call for help on resolving bug 1538506 [1] where the
proboscis tests randomly fail on the stable branches with
something
like:
TypeError: create() got an unexpected keyword argument 'slave_of'
Craig Vyvial has a proposed stable/kilo change [2] but it has
some
issues, at least from me as a reviewer of that change.
The tests are failing the periodic-stable job daily [3].
Can we get someone to help out with this? Otherwise I'm
inclined to
say the tests should be disabled on the stable branches, but
that's
pretty drastic.
Note that this is the kind of thing that breaks stable branch
policy, which is going to be part of a review when applying
for the
stable:follows-policy tag. [4] And the stable:follows-policy tag
may be used to determine when a stable branch for a project goes
end of life if it's not being maintained.
[1] https://bugs.launchpad.net/trove/+bug/1538506
[2] https://review.openstack.org/#/c/276934/
[3] http://goo.gl/fqf11U
[4]
http://governance.openstack.org/reference/tags/stable_follows-polic
y.h
tml
This is my solution for liberty, cap python-troveclient<2.1.0 in
global- requirements on stable/liberty [1].
Then there is a change to trove on stable/liberty to use the g-r
version range for python-troveclient for unit tests [2].
Alternatively, we could avoid the cap in stable/liberty by:
1. Reverting https://review.openstack.org/#/c/245738/
2. Blacklisting python-troveclient 2.1.0 in g-r 3. Release python-
troveclient 2.2.0.
It's getting late in the mitaka release to be messing with this
though since we're already past client freeze.
[1] https://review.openstack.org/#/c/290021/
[2] https://review.openstack.org/#/c/290023/
--
Thanks,
Matt Riedemann
____________________________________________________________________
______
OpenStack Development Mailing List (not for usage questions)
Unsubscribe:
openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
_____________________________________________________________________
_____
OpenStack Development Mailing List (not for usage questions)
Unsubscribe:
openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
--
Thanks,
Matt Riedemann
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe:
openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe:
openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Well, the point I'm trying to get to is, I don't think troveclient <
2.1.0 will work with the trove-api in mitaka.
For example, in my change to revert the slave_of removal in the client:
https://review.openstack.org/#/c/290048/
That fails in the API:
http://logs.openstack.org/48/290048/2/check/gate-trove-functional-dsvm-mysql/ddd3089/logs/screen-tr-api.txt.gz#_2016-03-08_19_30_32_790
"Validation error: instance Additional properties are not allowed
(u'slave_of' was unexpected)"
So if my client application that I wrote in liberty was using slave_of,
it's happily working until the cloud that my application is running on
upgrades to mitaka, then things break.
Part of the issue here is the backward incompatible change in the trove
API. The other part is, the trove API in mitaka requires
troveclient>=2.1.0 because that's what removed slave_of.
Right?
I'm trying to sort out what a valid troveclient is for master,
because I
don't really trust what's in global-requirements since trove hasn't
been
testing with those versions, at least not at the lower bound of 1.2.0.
Here is a change that bumps the minimum required version of
python-troveclient for mitaka to 2.1.0:
https://review.openstack.org/#/c/290168/
Here is my preferred option now.
1. Revert the slave_of removal from python-troveclient and provide it
as a compat shim until liberty-eol:
https://review.openstack.org/#/c/290048/
If slave_of is used, a deprecation warning is emitted (which we should
have been doing when this was deprecated in the first place). It also
never sends slave_of to the API, it's just a proxy to replica_of. So
this should work with both the liberty and mitaka versions of the
trove API.
2. Block python-troveclient 2.1.0 from stable/liberty g-r:
https://review.openstack.org/#/c/290021/
That blocks the bad 2.1.0 version from being used in stable/liberty
which is needed for...
3. Using the g-r versions of python-troveclient when testing on
stable/liberty:
https://review.openstack.org/#/c/290023/
If trove is following global-requirements, it should be using the
versions of python-troveclient from global-requirements rather than
installing it from the latest tarball on trunk. master and stable/kilo
are already doing this.
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
--
Thanks,
Matt Riedemann
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
