gzh <gzhcoder(at)126(dot)com> writes:
> I've run analyze(not vacuum analyze), but it doesn't seem to work.
When you're asking for help, please don't give us vague statements
like "doesn't seem to work". Did the plan (including rowcount
estimates) change at all? To what? How far off is that rowcount
estimate, anyway --- that is, how many rows actually have cseid = 94?
If the estimate is far off, then increasing the table's statistics
target might help.
Another thing that would be worth checking is whether
"set enable_seqscan = off" prods it to choose the plan you want.
If not, then there's something else going on besides poor estimates.
regards, tom lane