Re: dsm_unpin_segment

On Mon, Aug 8, 2016 at 8:53 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Thomas Munro <thomas(dot)munro(at)enterprisedb(dot)com> writes:
>> Yeah, I was considering unbalanced pin/unpin requests to be a
>> programming error. To be more defensive about that, how about I add a
>> boolean 'pinned' to dsm_control_item, and elog(ERROR, ...) if it's not
>> in the expected state when you try to pin or unpin?
>
> Well, what you have there is a one-bit-wide pin request counter.
> I do not see why that's better than an actual counter, but if that's
> what you want to do, fine.
>
> The larger picture here is that Robert is exhibiting a touching but
> unfounded faith that extensions using this feature will contain zero bugs.

That's an overstatement of my position. I think it is quite likely
that extensions using this feature will have bugs, because essentially
all code has bugs, but whether they are likely have the specific bug
of unpinning a segment that is already unpinned is not quite so clear.
That's not to say I object to Thomas's v2 patch, which will catch that
mistake if it happens. Defensive programming never killed anybody, as
far as I know. However, I don't see the need for a full-blown request
counter here; we've had this API for several releases now and to my
knowledge nobody has complained about the fact that you aren't
supposed to call dsm_pin_segment() multiple times for the same
segment. Therefore, I think the evidence supports the contention that
it's not broken and doesn't need to be fixed. If we do decide it
needs to be fixed, I think that's material for a separate patch.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company