Five shades of sharding

The importance of sharding is now colossal. The size of today's databases exceeds 100 terabytes, scaling up vertically, and adding replicas containing a complete physical copy of the database is becoming difficult, especially when computational resources are scarce. Database sharding is a way to scale horizontally by partitioning data between compute nodes that are independent of each other.

In the PostgreSQL world, there are well-known scaling tools - CitusDB, and Greenplum - as well as new generation solutions - Cockroach DB, Yugabyte DB, SPQR, and Shardman.

In this talk, we will discuss the difference between these implementations, the advantages, and disadvantages of these solutions, examine the current state of implementation of sharding in vanilla PostgreSQL, and also touch on another important topic - providing data integrity and consistency guarantees across a distributed cluster.

New UUID versions

Primary key generation is a known issue quite efficiently though not ideally solved by sequences. There are still problems with distributed generation, real uniqueness and predictability. Initially, UUID was not intended for use as a DBMS key, but its new versions (6, 7 and 8) make it quite convenient to use them as both synthetic and natural keys.

Using Debezium as a tool for delta migration

Delta migration case using Debezium is considered in details as well as the process of change data capture, sinking, applying configurations and load testing.

How we switched Oracle for PostgreSQL for a client, before it became mainstream

The history of import substitution did not start last year after well-known events. Its launch dates back to 2014. It was from this year that state and near-state companies began to think of switching to the so called "recommended software". One of these companies approached us back in 2020 with a project to move from Oracle to PostgreSQL. This project was designed to solve the accumulated architectural problems (imperfect storage system for telemetry data, the DBMS itself worked inside a virtual machine), and optimize the use of disk space (make space in the main storage, debug saving archived data, ensure correct backup). Since the customer's system should have worked uninterrupted 24/7, it was necessary to switch from one DBMS to another "seamlessly" without downtime, with simultaneous operation of both to ensure step-by-step translation of subsystems and the ability to control the correctness of the data. And, of course, the work had to be completed as quickly as possible.

In the report we will discuss how we managed to solve this case.