Enabling DR for application

[raman231]

I want to have DR enabled for my application deployed on SB. And since we have serverless application, having an Active Active DR strategy would not greatly impact the costs.

Just wanted to brainstorm around what approach we can take to make it DR enabled. maybe most of SB can remain in primary region only and just the application is replicated across regions. CFN stacksets can be an option to have a multi-region deployment. @PoeppingT @brtrvn what are your thoughts on how we should approach this.

[raman231]

I was thinking to have just the application being deployed to 2 regions and rest of the SB can remain in single region since we want just the application to run even in case of disaster (doesn't make sense to onboard new tenants during disaster).

I changed some lines in OnboardingService -
CreateStackSetResponse cfnResponse = cfn.createStackSet(CreateStackSetRequest.builder()
.stackSetName(stackName)
//.disableRollback(false)
.capabilitiesWithStrings("CAPABILITY_NAMED_IAM", "CAPABILITY_AUTO_EXPAND")
//.notificationARNs(ONBOARDING_STACK_SNS)
.templateURL(templateUrl)
.parameters(templateParameters)
.tags(tags)
.build()
);
CreateStackInstancesRequest stackInstancesRequest = CreateStackInstancesRequest.builder()
.stackSetName(stackName)
.deploymentTargets(DeploymentTargets.builder().accounts("xxxx").build())
.regions("us-east-1", "us-west-1")
.build();
cfn.createStackInstances(stackInstancesRequest);

It creates the stackset and add multiple stack instances.
Few challenges I can see -

1. Doesn't find the SSM parameters in recovery region, hence that instance of the stack fails.
2. Some services already have multi-region support (like dynamodb tables), we would not want to create different dynamodb table in each region)
3. SSM parameters are regional... there are options to replicate them to other region writing some lambda functions and all (but some of those are specific to a deployment in a region)
4. Installing SB in each region would make tenant onboarding a challenge in itself
5. StackSets do not support notificationARN support, but there is some support for EventBridge integration. And since there would be 2 stack instances, how do we track success of each before changing tenant onboarding status.
6. we are using different cognito userpool per tenant for application. since cognito is also regional, it would not fit in the multi-region strategy. We need to have the service running in multiple regions. We can replicate user profiles across regions, but passwords don't sync, hence will be a challenge.

@PoeppingT @brtrvn I know there will be other challenges than the ones I listed above, can you please highlight what other challenges do you foresee or if you have a better approach or suggest a different DR strategy.