Re: Should I implement DROP INDEX CONCURRENTLY?

On Wed, Feb 1, 2012 at 2:39 PM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:
> On Wed, Feb 1, 2012 at 7:31 PM, Peter Eisentraut <peter_e(at)gmx(dot)net> wrote:
>> On sön, 2012-01-29 at 22:01 +0000, Simon Riggs wrote:
>>> Patch now locks index in AccessExclusiveLock in final stage of drop.
>>
>> Doesn't that defeat the point of doing the CONCURRENTLY business in the
>> first place?
>
> That was my initial reaction.
>
> We lock the index in AccessExclusiveLock only once we are certain
> nobody else is looking at it any more.
>
> So its a Kansas City Shuffle, with safe locking in case of people
> doing strange low level things.

Yeah, I think this is much safer, and in this version that doesn't
seem to harm concurrency.

Given our previous experiences in this area, I wouldn't like to bet my
life savings on this having no remaining bugs - but if it does, I
can't find them.

I'll mark this "Ready for Committer".

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