Date: Wed, 14 Oct 2020 22:08:40 -0300
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Andy Fan <zhihui.fan1213@gmail.com>
Cc: Pg Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: ALTER TABLE .. DETACH PARTITION CONCURRENTLY
Message-ID: <20201015010840.GA20821@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CAKU4AWrme6KPToAGJrzPFwaLu_aFqAoXbiSKLR8HTnDwB00z2A@mail.gmail.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On 2020-Oct-15, Andy Fan wrote:

> I think if it is possible to implement the detech with a NoWait option .
> 
> ALTER TABLE ... DETACH PARTITION ..  [NoWait].
> 
> if it can't get the lock, raise "Resource is Busy" immediately,
> without blocking others.  this should be a default behavior.   If
> people do want to keep trying, it can set a ddl_lock_timeout to
> 'some-interval',  in this case, it will still block others(so it can't
> be as good as what you are doing, but very simple),  however the user
> would know what would happen exactly and can coordinate with their
> application accordingly.   I'm sorry about this since it is a bit of
> off-topics or it has been discussed already.

Hi.  I don't think this has been discussed, but it doesn't really solve
the use case I want to -- in many cases where the tables are
continuously busy, this would lead to starvation.  I think the proposal
to make the algorithm work with reduced lock level is much more useful.

I think you can already do NOWAIT behavior, with LOCK TABLE .. NOWAIT
followed by DETACH PARTITION, perhaps with a nonzero statement timeout.