On Tue, Jun 18, 2024 at 2:20 PM Thomas Munro <thomas(dot)munro(at)gmail(dot)com> wrote:
> Or enable_parallel_hash=off.
And if that allows the query to run, and it still uses the same basic
query plan with a Hash Right Join there, I wonder if it also uses a
huge number of batches, and if it does but manages to succeed, I
wonder if that could be just due to a difference in size of the
per-batch objects? I vaguely recall that there was a proposal to cap
nbatch appropriately for that allocation, but the cap would be
different for serial and parallel cases due to the element sizes.