What is going on with RonDB development. Actually a lot, but most happens under the radar at the moment. So this blog will give any interested some idea about what is going on.
RonDB core development is further development of the fork of MySQL NDB Cluster. For the most part this development is focused on our production version RonDB 21.04 that is used at numerous companies in production. Development is very centered around supporting the Hopsworks platform. This means that we now have added 27 new features on top of MySQL NDB Cluster and 127 bug fixes. The latest feature is an improvement of the node recovery. This improvement can bring up to 4-8x shorter restart times. This was seen as an important improvement to ensure that Online Reconfiguration of RonDB in our cloud setting is speedy.
We now have 3 main versions of RonDB core. The RonDB 21.04 that we use in production. RonDB 22.10 that is prepared for use in production. It brings the possibility to store 10x more data in RonDB compared to RonDB 21.04 important for large customers and large applications. We have also started work on the next RonDB generation in RonDB 23.04 that is integrated with MySQL 8.0.33 already.
Managed RonDB has been delivered in two steps. The first integrated the possibility to start up, backup, stop and restore a RonDB database. The configuration is specified in numbers of replicas, number of MySQL Servers and type of VMs for the various node types. One can start the cluster either through a UI or through Terraform.
Now the second step is working as well, this step introduces Online RonDB Reconfiguration. One can change the number of replicas, change the VM types of the nodes and increase/decrease the number of MySQL Servers. This is currently an experimental feature available to our customers on request. The change is fully online and has been verified in internal Hackathons where our developers test various Hopsworks features while the RonDB cluster is reconfiguring.
We are now working on a third step that makes changes more efficient and uses the Kubernetes model with desired state. So the cloud specifies the new desired state and the agent software will ensure that the RonDB cluster moves to this new desired state. Anyone can run RonDB in Docker and try out those new changes on their own laptops.
Those steps are also available using Docker with the rondb-docker github tree. We use Docker as a development platform making it easy to test thousands of state transformations at various levels. Soon there will be videos and blogs describing how to use Docker to test RonDB Reconciliation that will be accessible from the github tree.
It doesn't stop there, a major focus is currently on developing the first version of the RonDB REST API server. This makes it easy to access RonDB using a REST service in parallel with the MySQL Server and more efficient NDB API applications. We have already seen a great interest in this API even before it is completed.
We are also working on automating replication between clusters in different regions.
As usual there is also a set of interesting product ideas on how to improve the RonDB core with even more flexibility in growing and shrinking, making use of SIMD operations to speed up various parts of RonDB and some thoughts on long-term development projects as well.
As usual a benchmark or two is in the works as well. These are further developments of the benchmark described on www.rondb.com where we show throughput and latency of YCSB both in normal operations as well as during recovery.