Re: speedup COPY TO for partitioned table.

On Thu, Jun 26, 2025 at 9:43 AM torikoshia <torikoshia(at)oss(dot)nttdata(dot)com> wrote:
>
> After applying the patch, blank lines exist between these statements as
> below. Do we really need these blank lines?
>
> ```
> scan_rel = table_open(scan_oid,
> AccessShareLock);
>
> CopyThisRelTo(cstate, scan_rel, cstate->rel,
> &processed);
>
> table_close(scan_rel, AccessShareLock);
> ``
>
we can remove these empty new lines.
actually, I realized we don't need to use AccessShareLock here—we can use NoLock
instead, since BeginCopyTo has already acquired AccessShareLock via
find_all_inheritors.

> > +/*
> > + * rel: the relation from which the actual data will be copied.
> > + * root_rel: if not NULL, it indicates that we are copying partitioned
> > relation
> > + * data to the destination, and "rel" is the partition of "root_rel".
> > + * processed: number of tuples processed.
> > +*/
> > +static void
> > +CopyThisRelTo(CopyToState cstate, Relation rel, Relation root_rel,
>
> This comment only describes the parameters. Wouldn't it better to add a
> brief summary of what this function does overall?
>

what do you think the following

/*
* CopyThisRelTo:
* This will scanning a single table (which may be a partition) and exporting
* its rows to a COPY destination.
*
* rel: the relation from which the actual data will be copied.
* root_rel: if not NULL, it indicates that we are copying partitioned relation
* data to the destination, and "rel" is the partition of "root_rel".
* processed: number of tuples processed.
*/
static void
CopyThisRelTo(CopyToState cstate, Relation rel, Relation root_rel,
uint64 *processed)