Re: AW: AW: Call for alpha testing: planner statistics revi sion s

Zeugswetter Andreas SB <ZeugswetterA(at)wien(dot)spardat(dot)at> writes:
>> And on what are you going to base "sufficiently altered"?

> Probably current table size vs table size in statistics and maybe
> timestamp when statistics were last updated. Good would also be the
> active row count, but we don't have cheap access to the current value.

Once we get done with online VACUUM and internal free space re-use
(which is next on my to-do list), growth of the physical file will be
a poor guide to number of updated tuples, too. So the above proposal
reduces to "time since last update", for which we do not need any
backend support: people already run VACUUM ANALYZE from cron tasks.

> The point is, that if the combined effort of all "hackers" (with the
> help of some large scale users) cannot come to a more or less
> generally adequate answer, the field dba most certainly won't eighter.

True, but I regard your "if" as unproven. The reason for this call for
alpha testing is to find out whether we have a good enough solution or
not. I feel no compulsion to assume that it's not good enough on the
basis of no evidence.

regards, tom lane