Re: ALTER TABLE lock strength reduction patch is unsafe

On Mon, Jan 2, 2012 at 7:13 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Simon Riggs <simon@2ndQuadrant.com> writes:
>> On Mon, Jan 2, 2012 at 6:06 PM, Noah Misch <noah@leadboat.com> wrote:
>>> On Mon, Jan 02, 2012 at 05:09:16PM +0000, Simon Riggs wrote:
>>>> Attached patch makes SnapshotNow into an MVCC snapshot,
>
>>> That's a neat trick.  However, if you start a new SnapshotNow scan while one is
>>> ongoing, the primordial scan's snapshot will change mid-stream.
>
>> Do we ever do that?
>
> Almost certainly yes.  For example, a catcache load may invoke catcache
> or relcache reload operations on its way to opening the table or index
> needed to fetch the desired row.
>
> I think you can only safely do this if each caller has its own snapshot
> variable, a la SnapshotDirty, and that's going to be hugely more
> invasive.


I think I can avoid doing things that way and keep it non-invasive.

If we  #define  SnapshotNow   (GetSnapshotNow())

and make GetSnapshotNow() reuse the static SnapshotNowData if possible.
If not available we can allocate a new snapshot in TopTransactionContext.

We can free the snapshot at the end of scan.

That hides most of the complexity without causing any problems, ISTM.

Working on this now.

--
 Simon Riggs                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services