Re: Still more REINDEX fun

Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
> On Wed, Apr 20, 2011 at 7:41 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> It's still true though that you have to be REINDEXing system catalogs to
>> be at risk, else you shouldn't be seeing any IN_PROGRESS tuples.

> So the fix seems to be that we make REINDEX on a system catalog lock
> the whole catalog table.

It already does. The way to make system catalogs act more like user
tables for this purpose is at the other end: never release locks early
during DDL. And that patch would be pretty damn invasive too, not to
mention likely to introduce deadlock problems that don't exist today.
We've looked at that idea before and rejected it --- so I'm not inclined
to accept it now on no analysis, especially not in a fix that appears to
need to be backpatched for several releases.

regards, tom lane