"Sam Liddicott" <sam.liddicott@ananova.com> writes:
>> Indeed. What do 7.2 and 7.2.1 have in the pg_stats row for
>> broadcast.channelregionid? What is the real distribution of that
>> column?
> 7.2 says:
> [snip]
Well, that's annoying: the new ANALYZE code is in fact more accurate
than the old (it's estimating 429 distinct values, vs. formerly 4532,
when the true figure is 636) but it's nonetheless making a worse final
choice of plan.
If you turn off enable_seqscan, what EXPLAIN results do you get from
7.2.1? How about if you leave enable_seqscan on, but reduce
random_page_cost from the default 4.0 to perhaps 3.0?
regards, tom lane