On Wed, Oct 8, 2025 at 1:37 PM Melanie Plageman <[email protected]> wrote: > > I have updated my patch to fix the mistakes above. I also noticed then > that I wasn't doubling space_allowed in the loop but instead setting > it to hash_table_bytes at the end. This doesn't produce a power of 2 > because we subtract skew_mcvs from the hash_table_bytes. So, we have > to keep using space_allowed if we want a power of 2 in the end. > > I've changed my patch to do this, but this made me wonder if we want > to be doing this or instead take hash_table_bytes at the end and round > it up to a power of 2 and set space_allowed to that. If the skew > hashtable is large, we may be allocating way more space_allowed than > we need for new hash_table_bytes + skew hashtable buckets.
Oh wait, that doesn't make sense because each batch could have a skew hashtable. - Melanie
