10-100 Times Higher Latency With Distributed Postgres

10-100 Times Higher Latency With Distributed Postgres | Scaling Postgres 298

Рет қаралды 722

Күн бұрын

To get the show notes as well as get notified of new episodes, visit:
www.scalingpostgres.com/episo...
In this episode of Scaling Postgres, we discuss how distributed Postgres solutions can result in 10-100 times higher latency, building incremental backup, DBMS of the Year and understanding explain.
Want to learn more about Postgres performance?
Join my FREE mini-course called the PostgreSQL Performance Starter Kit here: www.scalingpostgres.com/cours...
Timestamps:
00:00 - Intro
00:14 - An Overview of Distributed PostgreSQL Architectures
09:07 - Incremental Backup: What To Copy?
12:56 - Incremental Backups: Evergreen and Other Use Cases
13:56 - Waiting for PostgreSQL 17 - Add support for incremental backup.
14:17 - PostgreSQL is the DBMS of the Year 2023
14:43 - EXPLAIN
15:03 - Create an AI Expert With Open Source Tools and pgvector
15:51 - Best pg_dump compression settings for Postgres in 2024
17:24 - My PostgreSQL wishlist
18:39 - Remote Ruby - Unleashing the Power of Postgres with Andrew Atkinson
18:55 - The Rails Changelog - #014: PostgreSQL for Rails Developers with Andrew Atkinson
19:02 - Outro
#postgres #postgresq

Пікірлер: 8

@andatki 6 ай бұрын

Thanks for mentioning the book and the podcast episodes Creston! Appreciate it.

@ScalingPostgres 5 ай бұрын

You bet!

@professortrog7742 6 ай бұрын

If you are using external tooling for the zstd compression, the you can compare oranges to oranges by using pigz for multithreaded gzip compression.

@ScalingPostgres 6 ай бұрын

For the use case I discussed, we were using pigz and moved to zstd since it outperformed it by speed and size.

@professortrog7742 6 ай бұрын

@@ScalingPostgres nice to know, thanks 👍🏻

@smyrnian_ 6 ай бұрын

Re. distributed options, I am designing an eLearning system that is distributed to 3 regions (US/EU/AU) where courses are always designed in a single region so they are only distributed across regions for faster local reads, but writes to a course by learners can occur in any region but learners will always write to their own region so there shouldn't be any row-level conflicts across regions. I figure eventual consistency via logical replication from each region to the other two regions would be ok in this use case.Thoughts?

@ScalingPostgres 6 ай бұрын

This sounds like a successful use case for using an active-active solution. If constructed as you describe and with care, you should be able to avoid replication conflicts between regions.

@DylanYoung 5 ай бұрын

The issue when eventual consistency in that case is that you need to make sure that any reads following writes go to the region that takes the writes (that or you need to merge your local data with the not-yet-consistent region you're reading from).