Re: BUG #4208: Server crashes on insert into gist index

[Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>]

Ron Mackley <ronm(at)signalpatterns(dot)com> writes:
> Thanks for the quick response. No:

> sp_hub_production=# select distinct cube_dim(attribute_vector) from  
> user_attribute_vectors;
>   cube_dim
> ----------
>          5
> (1 row)

> In the past we had a problem where runt vectors found their way into  
> the table, but we deleted them and try to detect them on their way in.

Well, if there were such entries in the past then they could still exist
in the index, I believe.  One or two such keys lurking in dusty corners
of the index would fit with the observed fact that you don't see the
crash often.  As a workaround, it'd probably be worth your trouble to
REINDEX that index to get rid of any such entries.

In any case, the swap bug is definitely real; for instance in cube's
regression database try this:

contrib_regression=# select cube_inter(c,'(1,2,3,4,5,6)') from test_cube;
server closed the connection unexpectedly
        This probably means the server terminated abnormally
        before or while processing the request.

I will go fix that ...

			regards, tom lane