Fault-tolerant State Machine in Java and SQL that does not allow meetings in Yandex Telemost to split

Media servers sometimes go down. Autohealing tries to replace them, but users may end up on different media servers and the conferences split as a result.

How do you protect against this? With the help of a high-performance and mathematically correct finite automaton implementation in Java and SQL!

We will consider examples of using analytical tools that are necessary in a programmer's arsenal. As well as best practices that allowed our team to solve the problem with minimal load on the system.