[benmmurphy]

I think that is a good question. It could be the queries were not very amenable to using indexes. In the article they also mentioned the query was taking 'minutes' which seems to indicate it is not an issue due to DB load or because the database is under resourced. I can understand that even if you have everything properly indexed if the DB isn't big enough or you are receiving too many queries then PG can still struggle with queries hitting 28 million rows. We have at least an order of magnitude more rows in one of our biggest tables and are doing paging queries against it and never receive timeouts. We use RDS which I guess is not super great for PG performance and are not even close to using the biggest instance.

[mewpmewp2]

If it's hitting an index properly it should perform binary search on it, and just be able to locate the row, correct?

They also say "We had a very large table (28 million) rows that we were getting timeouts querying on anything other than the primary key."

Which implies that indexing did work if that was primary key. How could other indexes fail? They should work with similar performance. Doesn't that imply they must've done something wrong or there was some sort of bug somewhere? Or were the column in some fashion I can't think of?