PGConf.Russia 2021

Everyone knows very well what a failover cluster PostgreSQL is and how such a cluster protects against failures within the same data center. However, recently, more and more enterprises have placed increased demands on their services, these requirements include disaster tolerance. We call such clusters a GEO-Cluster (KUK). In the report, I will talk about the varieties, principles and approaches to building GEO-Clusters PostgreSQL based on the Corosync/Pacemaker cluster software.

The built-in partitioning mechanism in PostgreSQL has been developing for several years. However, there is no magic button that would transform an ordinary table into a partitioned one. I will explain how to implement partitioning in a production system, how to prepare your table and apps for partitioning, and what pitfalls can make a DBA's life harder. We will also take a deep dive into various techniques of transferring data between partitions, also considering their pros, cons and limitation.

TimescaleDB extension allows to turn good old Postgres into a real distributed cluster for storing time series data while maintaining the relational model, convenient SQL and a time-tested ecosystem. And additional features such as continuous materialized views and data compression allow to build truly powerful telematic hubs.

We maintain large PostgreSQL databases with the total data volume exceeding 180 TB. Every instance in the cluster gets a workload of 15 thousand requests per second. These circumstances certainly impose limitations on the classical approach to updating database structures. The cost of a DBA error gets higher, as even a small error can lead to a sleepless night) In my talk, I will cover our existing limitations for changes in the production environment, non-classical DBMS behavior under heavy workloads, and will discuss PostgreSQL in general.