OK: so I think I've got some new insight by using test_decoding to
peek into the changes being replicate, here is what i think is
happening:
- there is a background job generating a Huge transaction (10K
changes) one after another that basically does DELETE on 1 table,
INSERT on 1 table and UPDAT
ccing the mailist again in case someone else would have a idea how to debug:
Here is what I see in the pg_locks on subscribe at all time:
As you can see, it mostly concern following 3 tables, accounts,
ledger, pending_ledger_fees, which I have tried analyze, vacuum them
etc, none of them helped m
Hi Justin, thanks for the response!
> REPLICA IDENTITY DEFAULT will only use primary keys, if the publisher
> includes those tables, the subscriber when replaying the WAL will stop
> throwing an error not knowing how to replay the UPDATE/DELETE.
But I don't see any errors being thrown out
On Tue, Feb 7, 2023 at 8:07 PM sunyuc...@gmail.com
wrote:
> Hi Justin:
>
> - i checked that I have 2 tables using replication identity FULL, but
> one table is empty and one table has only 1 row
> - 7 tables using index
> - overall I have ~100 tables in the publication: But I do find that
> some
Hi Justin:
- i checked that I have 2 tables using replication identity FULL, but
one table is empty and one table has only 1 row
- 7 tables using index
- overall I have ~100 tables in the publication: But I do find that
some tables doesn't have PK , but using replica identity default, I'm
abit co
On Tue, Feb 7, 2023 at 6:38 PM sunyuc...@gmail.com
wrote:
> Hi there,
>
> I am using PG 14.14 on both primary and secondary DB on AWS, setup
> using a logical replication, I'm having trouble with huge replication
> lag.
>
> My setup is as follows:
>
> P1 - physical - P1-R
> | (logical)
> P2 - p
