openstack-qa-team team mailing list archive

[corvus@xxxxxxxxxxxx (James E. Blair)]

Dan Smith <danms@xxxxxxxxxx> writes:

> I just verified this by looking at the zuul queue link you provided. I
> see in the check queue things in the following order:
>
> 11411, 11500, 11823, 11754, 11440, 11351, 11353, 11482

Oh, I thought you were talking about gating; the check pipeline tests
patches independently, exactly as they are uploaded.  So if you upload
three dependent patches, it will run tests three times.  Since it's
completely parallel, efficiency isn't quite as much of a concern.  We
could apply the same optimization ("this patch won't land because the
one it depends on fails"), but it may be useful to do a full test/report
anyway on patchset upload because a different set of tests may fail for
that change -- you wouldn't want to have to get the one ahead of it
working just to find out what's wrong with a change.

> but according to git, those should be in this order:
>
> 11411, 11351, 11352, 11353, 11410, 11440, 11481, 11482
>
> Not a huge deal, of course, but getting that ordering right is key to
> succeeding in the goal of not running dependent patches if their
> ancestors fail.

So in short, the changes in the check pipeline are un-ordered, but the
ones in the gate pipeline are ordered.  I'm sorry the information
display there is confusing; it could use some improvement (it was a
logging routine that quickly got put into service as a status screen).

-Jim