AWS has just announced the general availability of Aurora MySQL 8 compatibility (known as Aurora Version 3). This is long awaited addition to RDS MySQL 8 and provides many of the new features that can be found in the open-source MySQL 8 community version.
For those unfamiliar with Amazon Aurora my Understanding AWS RDS Aurora Capabilities presentation from Percona Live 2021 provides a great introduction of the benefits of this managed service.
There is a lot to digest and the Aurora User Guide provides details of the new features from the MySQL 8 community version, and of Aurora 3 new features, and feature differences or unsupported features. This AWS blog post also provides a general introduction.
It is very easy to spin up a new Aurora MySQL 3.01.0 cluster in an existing environment containing existing Aurora clusters. After defining new cluster and instance parameter groups for the
aurora-mysql8.0 family, or starting with the available default.aurora-mysql8.0 parameter groups, there are no other differences in aws rds create-db-cluster syntax, or using the AWS Console or Terraform syntax for example.
Before considering a migration of an existing Aurora cluster, there is a lot of information around parameter changes (including inclusive language functionality), and those related status and CloudWatch Metrics changes. Yes, looking at the 29 ‘Removed from Aurora MySQL version 3′, 30 ‘This parameter applies to Aurora MySQL version 3 and higher’ and presently ‘Currently not available in Aurora MySQL version 3′ LOAD|SELECT S3 capabilities is important. There are new reserved words to be aware of, you will need to note how to take advantage of roles within the Aurora permissions model.
Migrating an existing Aurora MySQL 2 cluster to Aurora 3 is a little more involved than specifying the snapshot-id. Seeing your restored Aurora 2 snapshot in an Aurora 3 cluster but with a status of incompatible-parameters is a good indication that more work is needed. While I will detail some of my experiences in a subsequent post, one helpful tip is found in those additional pages of the 5 rows of logs for your new cluster after all the error.log files, you will find an
upgrade-prechecks.log file. This contains an extensive list of checks and warnings performed for the upgrade. Skipping to the end of the JSON will give you an idea of your errorCount, warningCount and noticeCount.
Searching then for an object of “status”: “ERROR” will find the errorCount entries matching the count. Several other checks provide a “detectedProblems” section and a “level”: “Error” which would seem to be needed to be also corrected. There are a lot of checks between the INFORMATION_SCHEMA, InnoDB internal data dictionary and actual data/files on disk. You will also be presented with a nice long list of tables/columns using reserved words, as well as character set deprecations.
At a more technical glance of the documentation, there is a key change in how internal temporary tables are created, and how this differs from writer and reader instances. Benchmarking your application in different configurations will definitely be recommended.
Restoring an Aurora 2 cluster into Aurora 3 also took significantly more time; many hours; than a simple restore-db-cluster-from-snapshot you may be used to. While Terraform cluster creation timeouts need to be increased for global clusters, this time the default 1h30 timeout for an instance was also exceeded.
While different features will benefit different consumers of your Aurora database, one of the most anticipated is CTEs. From the operations perspective, as a managed service Aurora offers a subset of community features. One great feature that is now available in Aurora 3 is binary log filtering, a simple long-lived option in MySQL land that will help replacing more complex functionality.
This is a great and long awaited version release for Aurora.