Re: Parallel INSERT (INTO ... SELECT ...)

On Wed, Sep 23, 2020 at 2:21 PM Greg Nancarrow <gregn4422(at)gmail(dot)com> wrote:
>
> > - When INSERTs are made parallel, currently the reported row-count in
> > the "INSERT 0 <row-count>" status only reflects the rows that the
> > leader has processed (not the workers) - so it is obviously less than
> > the actual number of rows inserted.
>
> Attached an updated patch which fixes this issue (for parallel
> INSERTs, each worker's processed tuple count is communicated in shared
> memory back to the leader, where it is added to the global
> "es_processed" count).

I noticed that we are not having any check for skipping temporary
table insertion.

/* Check if the target relation has foreign keys; if so, avoid
* creating a parallel Insert plan (because inserting into
* such tables would result in creation of new CommandIds, and
* this isn't supported by parallel workers).
* Similarly, avoid creating a parallel Insert plan if ON
* CONFLICT ... DO UPDATE ... has been specified, because
* parallel UPDATE is not supported.
* However, do allow any underlying query to be run by parallel
* workers in these cases.
*/

You should also include temporary tables check here, as parallel
workers might not have access to temporary tables.

Regards,
Vignesh
EnterpriseDB: http://www.enterprisedb.com