Horton
Menu

Re: Perform streaming logical transactions by background workers and parallel apply

Dilip Kumar <dilipbalaut@gmail.com>

View thread
From: Dilip Kumar <dilipbalaut@gmail.com>
To: Masahiko Sawada <sawada.mshk@gmail.com>
Cc: "houzj.fnst@fujitsu.com" <houzj.fnst@fujitsu.com>, Amit Kapila <amit.kapila16@gmail.com>, "wangw.fnst@fujitsu.com" <wangw.fnst@fujitsu.com>, Peter Smith <smithpb2250@gmail.com>, "shiy.fnst@fujitsu.com" <shiy.fnst@fujitsu.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2022-10-18T11:52:39Z
Lists: pgsql-hackers 
On Thu, Oct 6, 2022 at 1:37 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>

> While looking at v35 patch, I realized that there are some cases where
> the logical replication gets stuck depending on partitioned table
> structure. For instance, there are following tables, publication, and
> subscription:
>
> * On publisher
> create table p (c int) partition by list (c);
> create table c1 partition of p for values in (1);
> create table c2 (c int);
> create publication test_pub for table p, c1, c2 with
> (publish_via_partition_root = 'true');
>
> * On subscriber
> create table p (c int) partition by list (c);
> create table c1 partition of p for values In (2);
> create table c2 partition of p for values In (1);
> create subscription test_sub connection 'port=5551 dbname=postgres'
> publication test_pub with (streaming = 'parallel', copy_data =
> 'false');
>
> Note that while both the publisher and the subscriber have the same
> name tables the partition structure is different and rows go to a
> different table on the subscriber (eg, row c=1 will go to c2 table on
> the subscriber). If two current transactions are executed as follows,
> the apply worker (ig, the leader apply worker) waits for a lock on c2
> held by its parallel apply worker:
>
> * TX-1
> BEGIN;
> INSERT INTO p SELECT 1 FROM generate_series(1, 10000); --- changes are streamed
>
>     * TX-2
>     BEGIN;
>     TRUNCATE c2; --- wait for a lock on c2
>
> * TX-1
> INSERT INTO p SELECT 1 FROM generate_series(1, 10000);
> COMMIT;
>
> This might not be a common case in practice but it could mean that
> there is a restriction on how partitioned tables should be structured
> on the publisher and the subscriber when using streaming = 'parallel'.
> When this happens, since the logical replication cannot move forward
> the users need to disable parallel-apply mode or increase
> logical_decoding_work_mem. We could describe this limitation in the
> doc but it would be hard for users to detect problematic table
> structure.

Interesting case.  So I think the root of the problem is the same as
what we have for a column is marked unique to the subscriber but not
to the publisher.  In short, two transactions which are independent of
each other on the publisher are dependent on each other on the
subscriber side because table definition is different on the
subscriber.  So can't we handle this case in the same way by marking
this table unsafe for parallel-apply?

-- 
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix invalid memory access during the shutdown of the parallel apply worker.

  2. Fix assertion failure in apply worker.

  3. Use elog to report unexpected action in handle_streamed_transaction().

  4. Use appropriate wait event when sending data in the apply worker.

  5. Allow the logical_replication_mode to be used on the subscriber.

  6. Rename GUC logical_decoding_mode to logical_replication_mode.

  7. Display the leader apply worker's PID for parallel apply workers.

  8. Improve the code to decide and process the apply action.

  9. Document the newly added wait events added by commit 216a784829.

  10. Perform apply of large transactions by parallel workers.

  11. Wake up a subscription's replication worker processes after DDL.

  12. Add copyright notices to meson files

  13. Better document logical replication parameters

  14. Add a common function to generate the origin name.

  15. Harmonize parameter names in storage and AM code.

  16. Avoid using list_length() to test for empty list.

  17. Improve two comments related to a boolean DefElem's value

  18. Fix partition table's REPLICA IDENTITY checking on the subscriber.

  19. Fix data inconsistency between publisher and subscriber.

  20. Fix cache look-up failures while applying changes in logical replication.