Percona XtraDB Cluster: Failure Scenarios and their Recovery

Percona XtraDB Cluster (a.k.a PXC) is an open source, multi-master, high availability MySQL clustering solution. PXC works with your MySQL / Percona Server-created database. Given the multi-master aspect, there are multi-guards to protect a cluster from entering an inconsistent state. Most of these guards are configurable based on their user environment. However, if they are not configured properly they could cause the cluster to stall, fail or error-out. In this session, we’ll discuss failure scenarios, including a MySQL cluster entering a non-primary state due to network partitioning. We’ll also discuss a cluster stall due to flow control, data inconsistency causing the shutdown of a node and common problems during the initial catch up – a.k.a State Snapshot Transfer (SST). Other issues include delays in the purging of a transaction, a blocking DDL causing the entire cluster to stall and a misconfigured cluster. We will also go over how to solve some of these problems and how to safely recover from these failures.