>
> I'm still confused how the optimizer in any scenario would choose an index
> with no relevant columns to complete a join.  Is there ever a scenario
> where such a choice makes sense?
>

I used to ask such questions until about 15 years ago. I never got an
entirely satisfactory answer.

To add to the mix, how table stats would change in such a way to lead the
> optimizer in such a direction, where for years it behaved reasonably.
>

Statistics that drive the optimised heuristics change over time. And every
time they get updated there is a very small but not 0 chance that the
random dive into the index tree will get sufficiently unlucky to get an
estimate that is far enough removed from reality to misinform the optimiser
into a suboptimal execution plan.
_______________________________________________
discuss mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to