Starting logical replication at arbitrary point that's available in WAL

On Friday, December 27, 2024 7:39 PM Alexander Uvizhev <uvizhe@posteo.net> wrote:

Hi,

I'm doing a logical replication using streaming replication protocol and
I'm trying to start a stream from a certain arbitrary point that's
available in WAL. However, both CREATE_REPLICATION_SLOT and
pg_create_logical_replication_slot() create slot with nearly last LSN/XID.

Is it possible to create a replication slot with arbitrary LSN/XID values?

I think it's not supported to specify arbitrary LSN/XID values for a newly
created slot in core. However, if you have an existing slot with an older LSN/XID, you
can copy it using pg_copy_logical_replication_slot, and then advance the copied
slot to your desired position with pg_replication_slot_advance.

Also pg_create_logical_replication_slot() for some reason gives
different result than CREATE_REPLICATION_SLOT: new slot's `catalog_xmin`
is set to the smallest `catalog_xmin` among already existing slots.
Looks like a bug.

Could you provide a script to reproduce this issue ?
That would be helpful in diagnosing the reason.

Best Regards,
Hou zj

On 12/27/24 13:20, Zhijie Hou (Fujitsu) wrote:

On Friday, December 27, 2024 7:39 PM Alexander Uvizhev<uvizhe@posteo.net> wrote:

Hi,

I'm doing a logical replication using streaming replication protocol and
I'm trying to start a stream from a certain arbitrary point that's
available in WAL. However, both CREATE_REPLICATION_SLOT and
pg_create_logical_replication_slot() create slot with nearly last LSN/XID.

Is it possible to create a replication slot with arbitrary LSN/XID values?

I think it's not supported to specify arbitrary LSN/XID values for a newly
created slot in core. However, if you have an existing slot with an older LSN/XID, you
can copy it using pg_copy_logical_replication_slot, and then advance the copied
slot to your desired position with pg_replication_slot_advance.

Thanks for the idea, I've already discovered it and it requires me to
implement some script to advance that dedicated slot, which I was hoping
to avoid.

Also pg_create_logical_replication_slot() for some reason gives
different result than CREATE_REPLICATION_SLOT: new slot's `catalog_xmin`
is set to the smallest `catalog_xmin` among already existing slots.
Looks like a bug.

Could you provide a script to reproduce this issue ?
That would be helpful in diagnosing the reason.

Unfortunately, I have no such script. But it looks like this:

test=> select slot_name, catalog_xmin, restart_lsn, confirmed_flush_lsn
from pg_replication_slots;
  slot_name    | catalog_xmin | restart_lsn | confirmed_flush_lsn
---------------+--------------+-------------+---------------------
 inactive_slot |        10073 | 0/1101DB88  | 0/1101DB88
 active_slot |        42607 | 0/17A87410  | 0/17A87410
 physical      |              | 0/17A875A0  |

test=> select pg_create_logical_replication_slot('test_slot',
'decoderbufs');
  slot_name    | catalog_xmin | restart_lsn | confirmed_flush_lsn
---------------+--------------+-------------+---------------------
inactive_slot|        10073 | 0/1101DB88  | 0/1101DB88
active_slot |        42607 | 0/17A87410  | 0/17A87410
 physical      |              | 0/17A875D8  |
 test_slot |        10073 | 0/17A875A0  | 0/17A875D8

While using CREATE_REPLICATION_SLOT would give me the same slot with
`catalog_xmin` = 42607.

Hope this helps.

Best Regards,
Hou zj

--
AU