PgConf.Russia 2017

A short talk about collecting data and monitoring database workload in Avito. Exporting metrics from stored procedures to Graphite. Collecting and visualizing pg_stat* metrics in Grafana. Case studies.

VIDEO

Most of the main Skype services use a database platform based on PostgreSQL and other open-source technologies, such as Skytools, plProxy, pgBouncer, etc. This platform consists of several hundreds of servers with thousands of databases, which process hundreds of thousands of transactions per second. At the same time, the platform architecture allows its users (applications and their developers) to work with "logical" databases, without any worries about their real “physical” structure.

Our Skype Database Platform team is responsible for the database platform infrastructure. We develop automation systems for various processes that help us ensure service reliability and facilitate development, testing, and deployment of code. In this presentation, I will outline the database platform architecture, review its main components, and tell you about the methods we use in our every-day work to ensure high availability, scalability, replication, fault tolerance, and more.

Sets are apparently a useful data type for many kinds of applications. While PostgreSQL offers no built-in set data type, sets may be emulated to some degree with its built-in array and JSONB data types. Also, acceleration of respective containment (subset) queries is readily available as a built-in feature of the GIN index type.

Starting with the above, we will then explore the performance gains enabled by custom set data types, and especially by customisation code in C ("operator classes") for the GIN and GiST index types.