Thanks for the insight, Tom. I had read that Postgres hash indexes didn't work too well, but thought that had been remedied in 8.2. Do you have any thoughts about btree versus GiST or GIN for text data with very few duplicate values?
"David Monarchi" <david.e.monarchi@gmail.com> writes: > I believe (but don't know) that a hash index would be better for this than a > btree.
A fairly reliable rule of thumb is that there isn't *any* situation where a Postgres hash index outperforms a btree.
Why this is so is not entirely clear, and various people keep poking at the code in hopes of making it better. Sooner or later we'll either succeed in getting hash indexes to be a win for specific use cases, or give up and drop them entirely.
But at present it is undeniable that Postgres hash indexes are not production quality. (Even if they had a performance win, their current lack of WAL backup makes them unfit for production use...)
