Re: Review request: XLogInsert

Robert Haas wrote:
> Not to distract from the issue at hand, but that goal doesn't seem
> quite aggressive enough, considering that this is the last week of the
> CommitFest, or near enough. Do you have a plan for wrapping this up?
To step back for a second, the fact that I have to create a plan shows a
failure in our getting this turned into a real process. That what I've
been trying to do here--step back each week and figure out what I should
have done, and try to make it more likely that will happen the next
time. We should have a clear plan charted that says what will happen at
each step here.

It looks to me like the end of the previous CF work finished via a huge
amount of "patch chasing" work from you and possibly other helpers (I
don't know, as I haven't gotten any such help myself the last few
weeks). That was fine since you were blazing a trail here, but that's
not a sustainable model. This whole thing needs clear written deadlines
and process if it's going to run more automatically in the future. CF
manager and helper labor isn't easy to find an unlimited amount of. I'd
like to turn this all into something more like a state machine whose
transitions are marked out on the calendar from day one.

At around three weeks, where we're at now, I think what should happen
next is:

1) All "waiting for author patches" turn into "returned with feedback"
as of some deadline. Since there wasn't one in advance, maybe I
announce one on the hackers list today?

2) Poll the reviewer of every patch that's had an updated version who
hasn't submitted a re-review asking whether they think that version is
"ready for comitter" now, if they have more feedback, or if they feel
it's just not ready yet and should be rejected. In any case but "ready
for committer", it goes into "returned with feedback" pile.

3) Any patches in this state that we haven't heard back from the
reviewer on within a couple of days get decided on ("ready" /
"returned") at the CommitFest manager's discretion. If anyone feels
wronged by that, they can always ask that a committer take a look
anyway. The CF manager won't always have as much information as we
expect the reviewers to, and can be presumed to have a thicker skin
about people getting mad at them for making a bad decision too.

I have a deliverable to ship today, once I'm done with that I'll start
rattling people more. Feedback about tweaking the above before I start
executing on it would be appreciated.

--
Greg Smith 2ndQuadrant Baltimore, MD
PostgreSQL Training, Services and Support
greg(at)2ndQuadrant(dot)com www.2ndQuadrant.com