Percona Live ONLINE

Kunlun Distributed Database Cluster is a distributed DBMS that aims to combine the best of both MySQL and PostgreSQL to meet the challenges of an ever-growing amount of data to manage and provide users with a highly performant, highly available, highly scalable and fault-tolerant, easy to use and manage database system that requires minimal human maintenance.

Kunlun Distributed Database Cluster enables users to define table sharding rules so that it automatically distributes tablets properly to available storage shards. We implemented the two-phase commit protocol to do distributed transaction commit; We use MySQL group replication for high availability in storage shards. We fixed a series of MySQL XA bugs to make Kunlun's distributed transaction processing highly reliable. We modified PostgreSQL to do far more than its FDW capabilities --- we implemented not only the global transaction coordinator but also concurrent asynchronous communication with MySQL storage shards and enhanced its query processing for best performance when accessing remote data in storage shards. Also, we implemented reliable DDL synchronization among multiple computing nodes of a cluster, the benefit is that all computing nodes have consistent metadata stored locally so that they can do query processing accurately with the best performance.

Come and see:

1. Kunlun design and architecture, key features, performance metrics, and how&why Kunlun can make the use of MySQL-8.0 group replication convenient and easy.
2. What issues exist with MySQL-8.0 XA transaction processing and how they are fixed in Kunlun Distributed Database Cluster, besides performance enhancements to MySQL-8.0.
3. How distributed query optimization in Kunlun makes queries performant and minimizes distribute query execution costs.
4. How dynamic table sharding distributes tablets properly as more storage shards are added to the cluster.
5. How global deadlocks come into being and how they are detected and resolved.
6. How DDL statements are synchronized among all computing nodes of a Kunlun cluster and how to make the synchronization fault-tolerant and reliable.