Re: TRUNCATE on foreign table - Mailing list pgsql-hackers

2021年3月30日(火) 3:45 Fujii Masao <masao.fujii@oss.nttdata.com>:
>
> On 2021/03/28 2:37, Kazutaka Onishi wrote:
> > Fujii-san,
> >
> > Thank you for your review!
> > Now I prepare v5 patch and I'll answer to your each comment. please
> > check this again.
> > m(_ _)m
> >
> > 1. In postgres-fdw.sgml, "and truncatable" should be appended into the
> > above first description?
> > 2. truncate.sgml should be updated because, for example, it contains
> > the above descriptions.
> >
> > Yeah, you're right. I've fixed it.
> >
> >
> >
> > 3.  Don't we need to document the detail information about frels_extra?
> >
> > I've written about frels_extra into fdwhander.sgml.
> >
> >
> >
> > 4. postgres_fdw determines whether to specify ONLY or not by checking
> > whether the passed extra value is zero or not.
> >
> > Please refer this:
> > https://www.postgresql.org/message-id/CAOP8fzb-t3WVNLjGMC%2B4sV4AZa9S%3DMAQ7Q6pQoADMCf_1jp4ew%40mail.gmail.com
> >> Negative value means that foreign-tables are not specified in the TRUNCATE
> >> command, but truncated due to dependency (like partition's child leaf).
> >
> > I've added this information into fdwhandler.sgml.
>
> Even when a foreign table is specified explicitly in TRUNCATE command,
> its extra value can be negative if it's found as an inherited children firstly
> (i.e., in the case where the partitioned table having that foreign table as
> its partition is specified explicitly in TRUNCATE command).
> Isn't this a problem?
>
> Please imagine the following example;
>
> ----------------------------------
> create extension postgres_fdw;
> create server loopback foreign data wrapper postgres_fdw;
> create user mapping for public server loopback;
>
> create table t (i int, j int) partition by hash (j);
> create table t0 partition of t for values with (modulus 2, remainder 0);
> create table t1 partition of t for values with (modulus 2, remainder 1);
>
> create table test (i int, j int) partition by hash (i);
> create table test0 partition of test for values with (modulus 2, remainder 0);
> create foreign table ft partition of test for values with (modulus 2, remainder 1) server loopback options
(table_name't'); 
> ----------------------------------
>
> In this example, "truncate ft, test" works fine, but "truncate test, ft" causes
> an error though they should work in the same way basically.
>
(Although it was originally designed by me...)
If frels_extra would be a bit-masked value, we can avoid the problem.

Please assume the three labels below:
#define TRUNCATE_REL_CONTEXT__NORMAL         0x01
#define TRUNCATE_REL_CONTEXT__ONLY               0x02
#define TRUNCATE_REL_CONTEXT__CASCADED     0x04

Then, assign these labels on the extra flag according to the context where
the foreign-tables appeared in the truncate command.
Even if it is specified multiple times in the different context, FDW extension
can handle the best option according to the flags.

> In this example, "truncate ft, test" works fine, but "truncate test, ft" causes

In both cases, ExecForeignTruncate shall be invoked to "ft" with
(NORMAL | CASCADED),
thus, postgres_fdw can determine the remote truncate command shall be
executed without "ONLY" clause.

How about the idea?

Best regards,
--
HeteroDB, Inc / The PG-Strom Project
KaiGai Kohei <kaigai@heterodb.com>