On Tue, May 27, 2014 at 12:19 PM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
>> Fair enough, but I don't think that affects correctness either way (I
>> don't think that you meant to imply that this was a necessary
>> precaution that you'd taken - right?).
>
>
> Right.
So, the comments within _bt_search() suggest that the _bt_moveright()
call will perform a _bt_finish_split() call opportunistically iff it's
called from _bt_doinsert() (i.e. access == BT_WRITE). There is no
reason to not do so in all circumstances though, assuming that it's
desirable to do so as soon as possible (which I *don't* actually
assume). If I'm not mistaken, it's also true that it would be strictly
speaking correct to never do it there. Do you think it would be a fair
stress-test if I was to hack Postgres so that this call never happens
(within _bt_moveright())? I'd also have an incomplete page split occur
at random with a probability of, say, 1% per split. The mechanism
would be much the same as your original test-case for the split patch
- I'd throw an error at the wrong place, although only 1% of the time,
and over many hours.
The concern with the _bt_moveright() call of _bt_finish_split() is
that it might conceal a problem without reliably fixing it,
potentially making isolating that problem much harder.
--
Peter Geoghegan