[Info-ingres] Interpreting Ingres 11 QEP

[Karl and Betty Schendel]

> On May 7, 2019, at 10:18 AM, Roy Hann <specially at processed.almost.meat> wrote:
> 
> [snip]
> 
> I am wondering if I really am seeing a change in the plan. Is this
> something to do with displaying a "reusable fragment"? 

Try it with "set noreuse"; I'd expect the reusable fragment to go away.

I'm pretty sure you are seeing a QEP change.  I'm not sure why it would be
happening, though.   The fragment is extracted in the rewriting phase, before
enumeration, and it looks like what was extracted was a restricted projection
of the customer table.   Why the subquery is sometimes planned with an
index fetch and sometimes with a customer scan, I've no idea;  the cost
numbers don't look anywhere close.  I assume that the effect is random?
I suspect a bug of some sort that is occasionally preventing the index
from being considered at all.

Karl