(aws-rds): DatabaseCluster should default `auto_minor_version_upgrade` to `False` if engine version is pinned

This may not be truly a bug, but it is not an ideal setup.

When using aws_rds.DatabaseCluster, I had a failing over-the-top deploy yesterday because the Aurora PostgreSQL version had been upgraded outside of the CFN stack. I was unaware that this was a behaviour, especially as I had specifically targeted a version of PostgreSQL.

I eventually realised there is the flag auto_minor_version_upgrade which I’d not set, but I noticed that its default value is True.

What I think would be more user friendly is if this value defaults to False if engine has a version specified.

Reproduction Steps

from aws_cdk import aws_rds

# Specify an RDS Cluster with a Version set like so
aws_rds.DatabaseCluster(
    self,
    id="rds-cluster",
    engine=aws_rds.DatabaseClusterEngine.aurora_postgres(
        version=aws_rds.AuroraPostgresEngineVersion.VER_10_12
    ),
    ...
)
...

What did you expect to happen?

Deploying the stack without further changes to the DatabaseCluster construct would result in a successful deploy.

What actually happened?

The deploy failed with the following error message:

23/42 |4:47:39 PM | UPDATE_FAILED        | AWS::RDS::DBInstance                        | prod-downloader-rds/Instance2 (proddownloaderrdsInstance2BE42B470) The specified DB Instance is a member of a cluster. Modify the DB engine version for the DB Cluster using the ModifyDbCluster API (Service: AmazonRDS; Status Code: 400; Error Code: InvalidParameterCombination; Request ID: fe271c1d-16c8-4acf-928e-dd93574d7ac4; Proxy: null)
	createInstances (/tmp/jsii-kernel-0VxvEJ/node_modules/@aws-cdk/aws-rds/lib/cluster.js:360:26)
	\_ new DatabaseCluster (/tmp/jsii-kernel-0VxvEJ/node_modules/@aws-cdk/aws-rds/lib/cluster.js:201:34)
	\_ /tmp/tmpbizvy7oq/lib/program.js:8154:58
	\_ Kernel._wrapSandboxCode (/tmp/tmpbizvy7oq/lib/program.js:8582:24)
	\_ Kernel._create (/tmp/tmpbizvy7oq/lib/program.js:8154:34)
	\_ Kernel.create (/tmp/tmpbizvy7oq/lib/program.js:7895:29)
	\_ KernelHost.processRequest (/tmp/tmpbizvy7oq/lib/program.js:9479:36)
	\_ KernelHost.run (/tmp/tmpbizvy7oq/lib/program.js:9442:22)
	\_ Immediate._onImmediate (/tmp/tmpbizvy7oq/lib/program.js:9443:46)
	\_ processImmediate (internal/timers.js:461:21)

When I checked the RDS Console, I could indeed see that even though my Stack was deployed with 10_12 as the version, it had been upgraded from underneath me to 10_14, breaking my deployment.

Environment

• CDK CLI Version: 1.105.0 (build 4813992)
• Framework Version: 1.105.0
• Node.js Version: v12.22.1
• OS: Ubuntu 20.04.2
• Language (Version): Python 3.8.6

Other

Following on from a discussion with @skinny85 here https://cdk-dev.slack.com/archives/C018XT6REKT/p1625763866267400

---

This is 🐛 Bug Report

[nohack]

@moltar yeah, looks like auto upgrading minor version is not yet supported in cloud formation itself.

https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/aws-resource-rds-dbcluster.html

[moltar]

And to piggyback on this issue.

Is there a setting for this for Aurora Serverless as well?

I checked the constructs and the Serverless one has no such options at all.

Non-serverless:

https://github.com/aws/aws-cdk/blob/283ed02c64f827161edba3e11c3cead3b54b7ee9/packages/%40aws-cdk/aws-rds/lib/cluster.ts#L719

Serverless:

https://github.com/aws/aws-cdk/blob/283ed02c64f827161edba3e11c3cead3b54b7ee9/packages/%40aws-cdk/aws-rds/lib/serverless-cluster.ts#L440-L457