PgConf.Russia 2020

Patroni and Stolon are two of the most popular and advanced solutions for orchestrating PostgreSQL and ensuring auto-failover for clusters with the Leader-Followers configuration. However, engineers who migrate from good old tools like Corosync&Pacemaker or built-in solutions of other DBMS often run into issues when installing these tools and figuring out the purpose of each component.
In this tutorial, we will go through a typical installation procedure of Patroni and Stolon clusters on virtual machines (not in containers), as well as review their behavior in case of various infrastructure failures. The whole process will be demonstrated on three vagrant virtual machines with pre-built images. You can follow along on your own system if you set up the required environment.

In the report, we want to talk about the experience of migrating one large system from Oracle to PostgreSQL. The system itself was built on the PHP + Oracle stack; its distinguishing feature was that all business logic was implemented in PL / SQL code. In a DBMS, there are more than 3000 packages with 4-10 functions (procedures) in each. In PHP - more than 10,000 forms with inserts of anonymous blocks used to receive data, process and save results in Oracle.

To solve this extremely voluminous work, we took ANTLR4 (a powerful parser generator for reading, processing, executing, or translating structured text), PL/SQL grammar, and created an automatic converter that allows you to convert all objects in the schema and our system from Oracle to working code for PostgreSQL.

This talk discusses the open source components we use at Adjust to manage a massive number (5+PB) of application log messages on PostgreSQL in a massively multi-parallel way. It provides both a use case for PostgreSQL in a big data (high volume/velocity/variety) environment, and can be used to show the power of PostgreSQL with JSONB, GIN, and more.

This talk covers the capabilities of the components in depth, sufficient to inspire similar solutions.

pg_repack is one of the most popular instruments for removing bloat of tables and indexes in Postgres. In most cases, it works perfectly. But if you use such a feature of Postgres as deferred constraints, using pg_repack becomes more difficult or even impossible. I will talk about how we encountered the problem and will describe some workarounds - from internal instruments of Postgres to a small patch for pg_repack.