Yoshinori Matsunobu

In this session, the speaker will talk about how MyRocks's online backups work. Both physical (myrockshotbackup) and logical (mysqldump) backups will be covered, and physical backups will be more focused. MyRocks (and RocksDB)'s physical backups basically work by 1. invoking RocksDB checkpoint that creates hard links then 2. copying data files (SST), redo logs (WAL) and others. myrockshotbackup has additional features like "renewing checkpoint" that makes replication catchup shorter, streaming backups, and auto-detecting O_DIRECT. Logical backups in MyRocks/RocksDB is more efficient than InnoDB because RocksDB more efficiently keeps track of transaction histories. How physical/logical MyRocks/RocksDB backups work are not widely discussed in public, and the speaker will demystify in this session.

In this session, the speaker will talk about how MyRocks's online backups work. Both physical (myrockshotbackup) and logical (mysqldump) backups will be covered, and physical backups will be more focused. MyRocks (and RocksDB)'s physical backups basically work by 1. invoking RocksDB checkpoint that creates hard links then 2. copying data files (SST), redo logs (WAL) and others. myrockshotbackup has additional features like "renewing checkpoint" that makes replication catchup shorter, streaming backups, and auto-detecting O_DIRECT. Logical backups in MyRocks/RocksDB is more efficient than InnoDB because RocksDB more efficiently keeps track of transaction histories. How physical/logical MyRocks/RocksDB backups work are not widely discussed in public, and the speaker will demystify in this session.

We recently finished migrating from InnoDB to MyRocks in our user database (UDB) at Facebook. We have been running MyRocks in production for a while and we have learned several lessons. In this talk, I will share several interesting lessons learned from production deployment and operations, and will introduce future MyRocks development roadmaps.

We recently finished migrating from InnoDB to MyRocks in our user database (UDB) at Facebook. We have been running MyRocks in production for a while and we have learned several lessons. In this talk, I will share several interesting lessons learned from production deployment and operations, and will introduce future MyRocks development roadmaps.

In this session, I'll talk about how we migrated from InnoDB to MyRocks in production at Facebook, for both slave and master instances, without stopping services, without losing data, and without returning wrong results. We created many features for easier migration, such as fast data loading, online data correctness check tool. I will share with you about lessons learned from our initial migrations and impacts what we have seen.

At Facebook, we have been developing and using the RocksDB storage engine in a number of different services and applications.

The goal of this talk is to introduce some background on the MyRocks project and some of the features we've added to MySQL and RocksDB to support our production environment.

At Facebook, we have been developing and using the RocksDB storage engine in a number of different services and applications. The MySQL on RocksDB (MyRocks) project integrates this library as a plugin into MySQL, the database server which we use to store our social graph data. We have found using a write optimized log structured merge tree design, when compared to the B+ tree implementation of InnoDB, reduces the space utilization and write rates of our production servers by half.

The goal of this talk is to introduce some background on the MyRocks project and some of the features we've added to MySQL and RocksDB to support our production environment. We will also introduce some of the performance data gathered from our testing and discuss what the tradeoffs are for using RocksDB vs. InnoDB.