Re: LOCK DATABASE

On Wed, May 18, 2011 at 7:11 PM, Alvaro Herrera
<alvherre(at)commandprompt(dot)com> wrote:
>> 1.  I suggested that this looks a lot like the controls of pg_hba.conf
>>
>> When our DBAs are doing major management of replication, they are
>> known to reconfigure pg_hba.conf to lock out all users save for the
>> one used by Slony.
>
> Yeah, I mentioned this but I think it actually sucks.

Why? I don't really see why this sucks.

> Well, I don't intend to lock "everything except the present connection".
> The only thing this LOCK DATABASE does is prevent the establishment of
> new connections.  Existing connections can continue to exist and work.
> So you do the LOCK DATABASE, then boot whoever shouldn't be allowed
> (which is a separate step that needs to be taken), then do your deed.
> If you want the slon connections to persist, just don't terminate them.

This strikes me as a hack. First, it's completely inconsistent with
how we lock tables or rows. A lock means you are the only one
accessing an object, not just that new accesses are locked out.
Second, it relies on the fact that a new connection briefly grabs a
lock on the database that is then released. If we happened (for
whatever reason) to want to change that to a session lock, or get rid
of it entirely, then this would break.

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