Re: ALTER TABLE lock strength reduction patch is unsafe

From: Robert Haas <robertmhaas(at)gmail(dot)com>
To: Alvaro Herrera <alvherre(at)2ndquadrant(dot)com>
Cc: Bruce Momjian <bruce(at)momjian(dot)us>, Simon Riggs <simon(at)2ndquadrant(dot)com>, Peter Geoghegan <pg(at)heroku(dot)com>, Noah Misch <noah(at)leadboat(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: ALTER TABLE lock strength reduction patch is unsafe
Date: 2014-01-29 05:04:51
Message-ID: CA+TgmoZjc=gcS_1uPrGU8h2TBj3qTKsJNmwz3pxbmSYLw779Yw@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Tue, Jan 28, 2014 at 12:36 PM, Alvaro Herrera
<alvherre(at)2ndquadrant(dot)com> wrote:
> Bruce Momjian escribió:
>> > I have no problem removing the parameter if required to. In that case,
>> > I would like to leave the parameter in until mid beta, to allow
>> > greater certainty.
>
> Uhm. If we remove a GUC during beta we don't need to force an initdb.
> At worst we will need to keep a no-op GUC variable that is removed in
> the next devel cycle. That said, if we're going to have a GUC that's
> going to disappear later, I think it's better to wait for 2 releases as
> proposed, not remove mid-beta.
>
>> > In any case, I would wish to retain as a minimum an extern bool
>> > variable allowing it to be turned off by C function if desired.
>
> I think this amounts to having an undocumented GUC that is hard to
> change. I prefer the GUC, myself.
>
>> Anything changed to postgresql.conf during beta is going to require an
>> initdb.
>> Also, lots of backward-compatibility infrastructure, as you are
>> suggesting above, add complexity to the system.
>>
>> I am basically against a GUC on this. We have far larger problems with
>> 9.3 than backward compatibility, and limited resources.
>
> If we have a clear plan on removing the parameter, it's easy enough to
> follow through. I don't think lack of resources is a good argument,
> because at that point there will be little to discuss and it's an easy
> change to make. And I think the plan is clear: if no bug is found the
> parameter is removed. If a bug is found, it is fixed and the parameter
> is removed anyway.
>
> Honestly, I would prefer that we push a patch that has been thoroughly
> reviewed and in which we have more confidence, so that we can push
> without a GUC. This means mark in CF as needs-review, then some other
> developer reviews it and marks it as ready-for-committer.

I also believe that would be the correct procedure. Personally, I
think it would be great if Noah can review this, as he's very good at
finding the kind of problems that are likely to crop up here, and has
examined previous versions. I also have some interest in looking at
it myself. But I doubt that either of us (or any other senior hacker)
can do that by Thursday. I think all such people are hip-deep in
other patches at the moment; I certainly am.

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

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Amit Kapila 2014-01-29 05:26:34 Re: Retain dynamic shared memory segments for postmaster lifetime
Previous Message Amit Kapila 2014-01-29 05:04:12 Re: [bug fix] pg_ctl always uses the same event source