On Tue, Jul 20, 2021 at 1:26 PM Amit Kapila <amit.kapil...@gmail.com> wrote:
> One more thing we need to think about here is when to find the right > bucket page in the chain where we can insert the new tuple. Do we > first try to complete the uniqueness check (which needs to scan > through the entire bucket chain) and then again scan the bucket with > space to insert or do we want to do it along with uniqueness check > scan and remember it? The latter approach, but that is just a performance tweak for later. On a unique hash index, regular splitting means there are almost no bucket chains more than 2 long (bucket plus overflow), so it seems like mostly wasted effort at this point. -- Simon Riggs http://www.EnterpriseDB.com/