The beta release of Apache Cassandra 4.0 is finally here, it’s been two years in the making. We’ve had a preview release available to customers since March for testing. A wide range of improvements have been made.
The explicit goal of this release has been to be “the most stable major release ever” to accelerate adoption within the release cycle, which I blogged about in January. For this release a series of new testing frameworks were implemented focusing on stability, and performance, which have paid off handsomely. The feeling of the team at Instaclustr is that we have never been more confident about a release (and we’ve seen quite a few!)
This release integrates the async event-driven networking code from Netty for communication between nodes. I blogged about Netty in February but it’s worth reiterating what’s been achieved with this upgrade. It has enabled Cassandra 4.0 to have a single thread pool for all connections to other nodes instead of maintaining N threads per peer which was cramping performance by causing lots of context switching. It has also facilitated zero copy streaming for SStables which now goes x5 times faster than before.
This complete overhaul of the networking infrastructure has delivered some serious gains.
- Tail end latency has been reduced by 40%+ in P99s in initial testing
- Node recovery time has been vastly reduced
- Scaling large clusters is easier and faster
Auditing and Observability
Cassandra 4.0 introduces a powerful new set of enterprise class audit capabilities that I covered here in March. These help Cassandra operators meet their SOX and PCI requirements with a robust high level interface. Audit logging saves to the node, outside of the database, with configurable log rollover. Audit logs can be configured to attend to particular keyspaces, commands, or users and they can be inspected with the auditlogviewer utility.
Full Query Logging is also supported and the fqltool allows inspection of these logs.
Virtual tables, which I covered here in February, enable a series of metrics to be pulled from a node via CQL from read-only tables. This is a more elegant mechanism than JMX access as it avoids the additional configuration required. JMX access is not going anywhere soon, but this presents a really solid improvement to a number of metric monitoring tasks.
Our community is our most powerful feature in all our releases and I can’t think of a better validation of open source under the Apache Foundation community model than this release. I just want to take this opportunity to congratulate and thank everyone in the community who have taken both Cassandra and its release processes to the next level with this beta release.
As always, you can spin up a free trial of Cassandra on our platform. Even with the performance gains delivered in this release our popular Cassandra Data Modeling Guide to Best Practices is always worth a read to get the most out of Cassandra.