Percona Live Online

In this talk, I will discuss tools for advanced testing of Distributed Databases. Kubernetes offers new interesting ways on how to emulate failures or even chain of catastrophic events like disk or network errors.
Using https://chaos-mesh.org/ Chaos Mesh I will emulate failures in Percona XtraDB Cluster Kubernetes Operator and will show how Operator recovers (or not) from them. This benefits not only database kernel developers but also is helpful to understand how your application behaves under failures.

For a considerable set of critical applications at Wise (former TransferWise), database high availability is a must to ensure that we don't let our customers down.

At Wise, we offload mostly of our relational databases operational toil into AWS RDS managed services. However, for some use-specific cases which need different availability requirements, we run some clusters on EC2.

In this presentation, I will show how we implemented high availability for our MariaDB clusters running on EC2 with an integration with Envoy and Orchestrator in order to decrease failovers and maintenances from few minutes to just a few seconds.

Widespread adoption of ProxySQL, the high performance, high availability, protocol-aware proxy for MySQL has lead to a plethora of different and highly innovative solutions for the inevitable scalability issues that MySQL DBAs run into with highly demanding workloads and the fast paced data growth of this modern era.

This talk aims to provide insights into real world MySQL scalability solutions implemented using ProxySQL by deep diving into the key areas to focus on when rolling out ProxySQL in production, examples on how to bulletproof the failover process and many other pertinent tuning recommendations.

- Where should I deploy ProxySQL?
- What hardware does ProxySQL need to run on?
- What does a typical production grade deployment look like?
- How should I design my ProxySQL query rules to ensure an efficient yet granularly controlled ruleset?
- What are the steps in planning a failover process?
- What should I do to achieve a transparent failover?
- What are the key variables I must tune on my production deployment?