Re: [PATCH] Unremovable tuple monitoring

On Tue, Nov 15, 2011 at 10:29 AM, Alvaro Herrera
<alvherre(at)commandprompt(dot)com> wrote:
> Excerpts from Robert Haas's message of mar nov 15 12:16:54 -0300 2011:
>> On Tue, Nov 15, 2011 at 10:05 AM, Yeb Havinga <yebhavinga(at)gmail(dot)com> wrote:
>> > I reviewed your patch. I think it is in good shape, my two main remarks
>> > (name of n_unremovable_tup and a remark about documentation at the end of
>> > this review) are highly subjective and I wouldn't spend time on it unless
>> > other people have the same opinion.
>>
>> I share your opinion; it's not obvious to me what this means either.
>> I guess this is a dumb question, but why don't we remove all the dead
>> tuples?
>
> They were deleted but there are transactions with older snapshots.

Oh. I was thinking "dead" meant "no longer visible to anyone". But
it sounds what we call "unremovable" here is what we elsewhere call
"recently dead".

> I think vacuum uses the term "nondeletable" or "nonremovable".  Not sure
> which one is less bad.  Not being a native speaker, they all sound
> horrible to me.

"nondeletable" is surely terrible, since they may well have got into
this state by being deleted. "nonremovable" is better, but still not
great.

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