Patch Pilot report

[Martin Pool]

2009/11/30 Andrew Bennetts <andrew.bennetts at canonical.com>:
> Robert Collins wrote:
>> On Mon, 2009-11-30 at 15:12 +1100, Martin Pool wrote:
>> >
>> > I think, unless and until Launchpad gets a concept of 'responsible
>> > person' for a review
>> > <https://bugs.edge.launchpad.net/launchpad-code/+bug/488540>, it
>> > doesn't make sense for pilots to retain ownership of a patch once
>> > their week ends.  So I'm going to look at all open reviews, and then
>> > from next Monday not do any piloting.  Obviously if people have a
>> > personal interest in a particular patch they can keep looking at it.
>>
>> I disagree with this quite strongly; the number of patches to pilot is
>> pretty small - within a week its very tolerable - and handoffs are where
>> you get friction such as differences of opinion causing unnecessary
>> rework.
>>
>> Its not hard for a pilot to have a tomboy note open with the urls of
>> things they haven't guided through to 'in' or 'you need to do more
>> work'.
>
> There are two separate goals being aimed for here, both important:
>
>  1) make sure that new patches coming in are helped along in a timely fashion
>  2) clearing out our backlog of patches, some of which have been lingering for
>    months
>
> I agree with Martin's concern that a pilot that tries to clear the backlog would
> be overly burdened if they remained responsible for that backlog after their
> week in the pilot's seat.  I also think Robert makes a good point that handoffs
> are costly and should be minimised.

Essentially the problem with having people retain pilot responsibility
for some patches is that it complicates the queue state in a way that
isn't modeled in the software.  The rule for the pp is no longer "try
to get the queue emptied."

> So perhaps the thing to do is treat these separately.  For 1), do as Robert
> says: the pilot on duty when the patch arrives will keep piloting that patch
> until it's finished (landed or rejected).  For 2), pilots should also try to
> reduce the backlog, but no ongoing commitment is required for these.

I think if someone wants to say "I'll keep helping you with this next
week" that's great, and obviously the pp shouldn't just ride roughshod
over what another reviewer has said.

Maybe I'm failing to understand what kind of churn Robert thinks will
be introduced.

> Thoughts?

I think we should try what seems to me to be the simpler algorithm,
and then if it turns out in practice to be inefficient, change it.

-- 
Martin <http://launchpad.net/~mbp/>