[openstack-dev] [devstack][infra] pip vs psutil

Jens Harbott j.harbott at x-ion.de
Mon Apr 16 14:33:44 UTC 2018


2018-04-16 7:46 GMT+00:00 Ian Wienand <iwienand at redhat.com>:
> On 04/15/2018 09:32 PM, Gary Kotton wrote:
>>
>> The gate is currently broken with
>>  https://launchpad.net/bugs/1763966.
>> https://review.openstack.org/#/c/561427/
>>  Can unblock us in the short term. Any other ideas?
>
>
> I'm thinking this is probably along the lines of the best idea.  I
> left a fairly long comment on this in [1], but the root issue here is
> that if a system package is created using distutils (rather than
> setuptools) we end up with this problem with pip10.
>
> That means the problem occurs when we a) try to overwrite a system
> package and b) that package has been created using distutils.  This
> means it is a small(er) subset of packages that cause this problem.
> Ergo, our best option might be to see if we can avoid such packages on
> a one-by-one basis, like here.
>
> In some cases, we could just delete the .egg-info file, which is
> approximately what was happening before anyway.
>
> In this particular case, the psutils package is used by glance & the
> peakmem tracker.  Under USE_PYTHON3, devstack's pip_install_gr only
> installs the python3 library; however the peakmem tracker always uses
> python2 -- leaing to missing library the failures in [2].  I have two
> thoughts; either install for both python2 & 3 always [3] or make
> peakmem tracker obey USE_PYTHON3 [4].  We can discuss the approach in
> the reviews.
>
> The other option is to move everything to virtualenv's, so we never
> conflict with a system package, as suggested by clarkb [5] or
> pabelanger [6].  These are more invasive changes, but also arguably
> more correct.
>
> Note diskimage-builder, and hence our image generation for some
> platforms, is also broken.  Working on that in [7].

The cap in devstack has been merged in master and stable/queens, other
merges are being help up by unstable volume checks or so it seems.

There is also another issue caused by pip 10 treating some former
warning as error now. I've tried to list all "global" (Infra+QA)
related issues in [8], feel free to amend as needed.

[8] https://etherpad.openstack.org/p/pip10-mitigation



More information about the OpenStack-dev mailing list