Date: Thu, 20 Aug 2015 06:19:02 -0700
From: David Fetter <david@fetter.org>
To: Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>
Cc: Pg Hackers <pgsql-hackers@postgresql.org>
Subject: Re: Declarative partitioning
Message-ID: <20150820131902.GB8691@fetter.org>
References: <55D3093C.5010800@lab.ntt.co.jp>
 <20150818134323.GC18054@fetter.org>
 <55D4309F.8030601@lab.ntt.co.jp>
 <20150819125239.GE11258@fetter.org>
 <55D5A4C0.4040900@lab.ntt.co.jp>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <55D5A4C0.4040900@lab.ntt.co.jp>
User-Agent: Mutt/1.5.23 (2014-03-12)
Precedence: bulk
Sender: pgsql-hackers-owner@postgresql.org

On Thu, Aug 20, 2015 at 06:58:24PM +0900, Amit Langote wrote:
> On 2015-08-19 PM 09:52, David Fetter wrote:
> > On Wed, Aug 19, 2015 at 04:30:39PM +0900, Amit Langote wrote:
> > 
> > There are use cases where we need to warn people that their assertions
> > need to be true, and if those assertions are not true, this will
> > explode, leaving them to pick the resulting shrapnel out of their
> > faces.  There are other parts of the system where this is true, as
> > when people write UDFs in C.
> > 
> > As I understand it, NOT VALID means, "I assert that the tuples already
> > here fit the constraint.  Any changes will be checked against the
> > constraint."
> > 
> > I've seen cases where a gigantic amount of data is coming out of some
> > distributed system which holds the constraint as an invariant.  This
> > let a DBA decide to add a NOT VALID constraint in order not to take
> > the hit of a second full scan of the data, which might have made the
> > import, and possibly the entire project, untenable.
> 
> Ah, I understand the point of parameterization (TRUST). Seems like it
> would be good to have with appropriate documentation of the same. Perhaps,
> it might as well a parameter to the step 1 itself.

So TRUST would obviate step 2?  Better still!

> >>>> 5. Detach partition
> >>>>
> >>>> ALTER TABLE partitioned_table
> >>>> DETACH PARTITION partition_name [USING table_name]
> >>>>
> >>>> This removes partition_name as partition of partitioned_table.
> >>>> The table continues to exist with the same name or 'table_name',
> >>>> if specified.  pg_class.relispartition is set to false for the
> >>>> table, so it behaves like a normal table.
> >>>
> >>> Could this take anything short of an access exclusive lock on the
> >>> parent?
> >>
> >> Yes, both the step 1 of ATTACH command and DETACH command take
> >> access exclusive lock on the parent. They are rather quick metadata
> >> changes, so should not stall others significantly, I think.
> > 
> > So no.  Weakening required locks has been something of an ongoing
> > project, project-wide, and need not be part of the first cut of this
> > long-needed feature.
> > 
> 
> Do you mean ATTACH and DETACH, if they require access exclusive lock on
> the parent, should not be in the first cut? Or am I misreading?

Sorry I was unclear.

ATTACH and DETACH should be in the first cut even if they require an
access exclusive lock.

Cheers,
David.
-- 
David Fetter <david@fetter.org> http://fetter.org/
Phone: +1 415 235 3778  AIM: dfetter666  Yahoo!: dfetter
Skype: davidfetter      XMPP: david.fetter@gmail.com

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers