It’s a good practice to use CloudFormation’s DeletionPolicy
to protect stateful resources such as DynamoDB tables or RDS databases from accidental deletions. Such as when someone accidentally deletes a CloudFormation stack!
As I discussed previously [1], this is a much better way to guard against these accidental data losses than separating stateful and stateless resources into different stacks. After all, how we think about infrastructure vs. code needs to evolve in the serverless era [2]. That separation is no longer so clear-cut.
The problem with DeletionPolicy
But using DeletionPolicy
on all your stateful resources causes friction with another common practice in serverless development – the use of ephemeral (or temporary) environments [3].
If you set DeletionPolicy
to Retain
on all your stateful resources, then they will all linger when you delete the temporary environment.
In fact, you only need to use DeletionPolicy
in production. That’s the only environment where it’s necessary.
So, how do we do that?
The CDK solution
CDK gives you the full power of a general-purpose programming language (even though it’s often not needed [4]). So it’s trivial to do this in a CDK application:
- Pass the environment name as either a CloudFormation parameter or context variable.
- Set the
removalPolicy
inside anif
block like this:
if (environment === 'prod') {
myTable.removalPolicy = cdk.RemovalPolicy.RETAIN
}
When you have branch logic like this in your CDK code, you should add unit tests for them.
And instead of using magic strings like “prod”, you should also capture constants as enums.
Additionally, you can use CDK Aspects to select multiple resources and update their policy in one go. For more details, check out this article [5] by Wojciech Matuszewski.
Serverless / SAM / CloudFormation solution
For the Serverless Framework, SAM or CloudFormation, you can do this:
- Add an
Environment
CloudFormation parameter. (not needed for the Serverless Framework) - Add an
IsProd
condition. - Use the
!If
intrinsic function so theDeletionPolicy
attribute is set toRetain
when theEnvironment
is “prod”, else set it toDelete
.
Like this:
Parameters:
Environment:
Type: String
Default: dev
Description: Environment name, e.g. dev, test, prod
Conditions:
IsProd: !Equals [ !Ref Environment, "prod" ]
Resources:
MyTable:
Type: AWS::DynamoDB::Table
DeletionPolicy: !If [ "IsProd", "Retain", "Delete" ]
Properties:
...
For the Serverless Framework, you can skip step 1 because it has a built-in sls:stage
parameter. So the IsProd
condition becomes:
Conditions:
IsProd: !Equals [ "${sls:stage}", "prod" ]
DynamoDB DeletionProtectionEnabled attribute
I have used DynamoDB in the examples above. But it’s worth pointing out that DeletionPolicy
is actually not the most effective protection against data loss for DynamoDB.
For starters, it doesn’t guard against someone deleting a DynamoDB table in the console or programmatically through the AWS CLI or SDK.
Fortunately, DynamoDB also has the DeletionProtectionEnabled
[6] attribute. When enabled, it protects the table from accidental deletion by any user or process.
Additionally, DynamoDB offers other protection against data loss, such as point-in-time recovery and the ability to export data to S3.
Having said that, DeletionPolicy
is still a very useful tool for other services such as S3, EventBridge, SQS, SNS, etc.
Use RetainExceptOnCreate instead of Retain
Another thing to consider is that the new RetainExceptOnCreate
deletion policy (announced in July 2023) has superseded the Retain
deletion policy. It should be used as a stand-in replacement for the Retain
deletion policy these days.
Built it into your Platform
Instead of every team having to remember to do this with every project and every stateful resource, it might be better to make this a feature of your cloud platform.
Lars Jacobsson gave us an example of how that might look like, using a CloudFormation macro.
If you need a refresher on CloudFormation macros and what they can do, then please check out this article [7] by Alex DeBrie
Wrap up
Finally, I want to give a shout-out to Hala Al Aali for asking me about this in the Testing Serverless Architectures [8] course’s forum.
It’s a common problem that many students have come across. I hope this article has helped you. If you want to learn more about building serverless applications for the real world, check out my upcoming workshops [9].
Links
[1] This is why you should keep stateful and stateless resources together
[2] Are we getting infrastructure all wrong in the Serverless era?
[3] Serverless Ephemeral (Temporary) Environments Explained
[4] Are You Ready for This? Top 5 Earth-Shattering Pros and Cons of AWS CDK
[5] Enforcing compliance with AWS CDK Aspects
[6] DynamoDB’s DeletionProtectionEnabled attribute
[7] How and Why to use CloudFormation Macros
[8] Testing Serverless Architectures course
[9] Production-Ready Serverless workshop