On Tue, Jul 02, 2019 at 01:26:26AM +1200, David Rowley wrote: > I've pushed the original patch plus a small change to only call > table_finish_bulk_insert() for the target of the copy when we're using > bulk inserts.
Yes, sorry for coming late at the party here. What I meant previously is that I did not find the version published at [1] to be natural with its structure to define an executor callback which then calls a callback for table AMs, still I get your point that it would be better to try to avoid unnecessary fsync calls on partitions where no tuples have been redirected with a COPY. The version 1 of the patch attached at [2] felt much more straight-forward and cleaner by keeping all the table AM callbacks within copy.c. This has been reverted as of f5db56f, still it seems to me that this was moving in the right direction. [1]: https://postgr.es/m/CAKJS1f95sB21LBF=1mcsev+xlta_jc3mtxx5kgduhdsogow...@mail.gmail.com [2]: https://postgr.es/m/cakjs1f_0t-k0_3xe+erxpq-jgaob6trzayercxf2rpgduvm...@mail.gmail.com -- Michael
signature.asc
Description: PGP signature
