Re: [HACKERS] reload-through-the-top-parent switch the partition table

On Thu, Aug 3, 2017 at 9:01 AM, Rushabh Lathia <rushabh.lathia@gmail.com> wrote:
> --unpartition-partitioned-table is very confusing.

+1.

> I liked the previous option which is --use-partitioned-table
> [partition-name, ...].
> The Only problem with --use-partitioned-table is, a user needs to specify
> the
> partition-name in the options list. Imagine if someone having 100's of
> partitions then specifying those name is pg_dump option is a pain.

Yeah, that won't work.

> Rather than that:
>
> --use-partitioned-table [partitioned_name, ...] # if
> names are omitted it defaults to all the partitioned tables.
>
> Here user need to specify the root relation name in the option - and any
> partition table have that as a ROOT, will load the data through
> top-parent-relation.

We could do that, but I'm not sure it's a good idea to use
getopt_long() with optional options.  Sometimes that creates confusion
-- is pg_dump --use-partitioned-table salad an attempt to dump the
salad database with the --use-partitioned-table option, or an attempt
to apply --use-partitioned-table only to partitions whose parent is
the salad table?  getopt_long() has an answer, but some people may
guess incorrectly about what it is.

I would be more inclined to make this a global option than something
that modifies the behavior for certain tables; the only per-table
flags we have right now are just to include/exclude individual tables.
You could make --inserts or --no-unlogged-table-data apply to some but
not all tables, but we didn't; why start here?

I don't like the specific name --use-partitioned-table much either.
Use it for what?

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company