[openstack-dev] [all] create periodic-ci-reports mailing-list
Matthew Treinish
mtreinish at kortar.org
Wed Apr 13 16:58:47 UTC 2016
On Wed, Apr 13, 2016 at 06:22:50PM +0200, Ihar Hrachyshka wrote:
> Matthew Treinish <mtreinish at kortar.org> wrote:
>
> > On Wed, Apr 13, 2016 at 10:59:10AM -0400, Emilien Macchi wrote:
> > > Hi,
> > >
> > > Current OpenStack Infra Periodic jobs do not send e-mails (only
> > > periodic-stable do), so I propose to create periodic-ci-reports
> > > mailing list [1] and to use it when our periodic jobs fail [2].
> > > If accepted, people who care about periodic jobs would like to
> > > subscribe to this new ML so they can read quick feedback from
> > > failures, thanks to e-mail filters.
> >
> > So a big motivation behind openstack-health was to make doing this not
> > necessary. In practice the ML posts never get any real attention from
> > people
> > and things just sit. [3] So, instead of trying to create another ML here
> > I think it would be better to figure out why openstack-health isn't working
> > for doing this and figure out how to improve it.
> >
> > -Matt Treinish
>
> What I like in ML is that you get notified, instead of going to a website
> each day to check job health, while it passes. In the end, you go there not
> daily, but weekly or monthly, and so the failure response is not immediate.
>
So, sure I understand the attraction of an active notification. But, for that to
work that assumes people actively looking at things and taking action as soon as
they get an email. That very rarely happens, even for people who look at the
failures regularly. Let's take a look at the stable maint list:
http://lists.openstack.org/pipermail/openstack-stable-maint/2016-February/thread.html
http://lists.openstack.org/pipermail/openstack-stable-maint/2016-March/thread.html
That's 2 months, where is the immediate response on those? Only a handful only
have a single isolated failure, and looking at them they are mostly transient
failure (like a down mirror, git failure, upstream release, etc) which was
likely blocking other development and was almost certainly fixed without looking
at the ML results. The majority of the failure just sat for a long time.
The issue with the ML is that it's not actually used as a notification but
instead people only look at the details periodically. The ML provides a really
bad interface for visualizing these results over time. That's why using
openstack-health is a better solution for doing this. Now, I don't pretend that
it's complete or perfect, it's still a relatively young project. But, the thing
is everyone can help us work on fixing any issues or gaps with it:
http://git.openstack.org/cgit/openstack/openstack-health/
Instead, of pretending the ML work for doing this I think it'll be better if we
concentrate on making the dashboard for this better.
-Matt Treinish
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20160413/0bb35fad/attachment.pgp>
More information about the OpenStack-dev
mailing list