Re: Logical Replication speed-up initial data

[Nikhil Shetty <nikhil.dba04@xxxxxxxxx>]

Hi Stefano,

Thank you for the information.

Regards,

Nikhil

On Wed, Aug 4, 2021 at 9:25 PM Stefano Amoroso <stefano.amoroso@xxxxxxxxx> wrote:

Hello,

in my experience, to speed up the initial load, I had to drop UKs and FKs.

Unfortunately, the initial load doesn't work in parallel and, for each table, there is only one sync worker.

Regards

Stefano Amoroso

Il giorno mer 4 ago 2021 alle ore 17:24 Hüseyin Demir <demirhuseyinn.94@xxxxxxxxx> ha scritto:

Hello,

I also faced a similar issue. Try removing the indexes on the destination first if possible. After that, you can add the indexes.

Regards.

Nikhil Shetty <nikhil.dba04@xxxxxxxxx>, 4 Ağu 2021 Çar, 18:07 tarihinde şunu yazdı:

Hi Team,

We have a highly transactional system as the source of logical replication and the database size is 500GB+. We are replicating all tables from source using logical replication. 

For two tables the initial data load is very slow and it never completes even after 24hrs+ 

Table size is under 100GB and index size is around 400GB. 

How can we increase the speed of the initial data load without dropping the indexes on destination?

We increased max_sync_workers_per_subscription to 3 but it didn't help much for single tables

Thanks,

Nikhil

--

Hüseyin Demir

Senior Database Platform Engineer

Twitter:  https://twitter.com/d3rh5n

Linkedin: hseyindemir

Github: https://github.com/hseyindemir

Gitlab: https://gitlab.com/demirhuseyinn.94

Medium: https://demirhuseyinn-94.medium.com/