"Zeugswetter Andreas SB SD" <ZeugswetterA@spardat.at> writes:
> The pg_stats values are only, um, statistical. If 99.9% of the table is
> the same value and the other four values appear only once or twice, it's
> certainly possible for ANALYZE's sample to include only the common value
> and miss the rare ones. AFAIK that will not break anything; if you have
> an example where the planner seems to be fooled because of this, let's
> see it.
> Hmm ? How about select * from xxx where profiles_faith = 7
> would estimate all rows, no ? Instead of 2.
Not in 7.2 ... nor in previous versions AFAIR.
> That is why I think a bin for "very uncommon" values could also be
> useful sometimes.
Perhaps you should experiment or read the code before opining...
regards, tom lane