[aws-rds] Minimize downtime during DBCluster updates

[hixi-hyi]

Minimize downtime during DB Cluster updates

Current Status

The CfnDBInstance of DBCluster is currently loosely coupled.
That is, if there are multiple CfnDBInstances, Instance updates will occur at the same time because there is no dependency on them on Cloudformation.
Therefore, the cluster will not be available until the DBInstance update is complete.

Proposal

Adds a dependency to CfnDBInstance.
As a result, one by one, RollingUpdate will be performed, and the only downtime will be the timing of the primary switch.
In other words, when there are two instances, it will take only two failover times to update. (A)
If we can create Dependency dynamically, it will take only a one-time failover time to update. (B)

I think primary failover times are faster than Instance updates. So I think it would be useful to include this feature.
However, the update time for Stack and the maintenance time for offline updates will increase.

What do you think about this proposal?
I'd like to hear your opinion.

Proposal Solution (A)

aws-cdk/packages/@aws-cdk/aws-rds/lib/cluster.ts

Line 734 in d95af00

instance.node.addDependency(internetConnected);

Add instance.node.addDependency(previous_instance);

Proposal Solution (B)

I think we need to use aws-sdk to determine if the current Instance is primary or replica, but I haven't thought about it in detail.

• 👋 I may be able to implement this feature request
• ⚠️ This feature might incur a breaking change

---

This is a 🚀 Feature Request

[hixi-hyi]

p.s. I thought it might be a good idea to create an instanceUpdateBehavior argument and change the behavior accordingly. (ROLLING, BULK)

[skinny85]

Hey @hixi-hyi ,

thanks for opening the issue. This is a very interesting proposal. Pinging @jogold as well , for visibility.

p.s. I thought it might be a good idea to create an instanceUpdateBehavior argument and change the behavior accordingly. (ROLLING, BULK)

Where does this instanceUpdateBehavior live? On the Cluster itself, or somewhere else?

[hixi-hyi]

@skinny85

Where does this instanceUpdateBehavior live? On the Cluster itself, or somewhere else?

I think a better definition would be Define to

aws-cdk/packages/@aws-cdk/aws-rds/lib/cluster.ts

Line 22 in 4e03667

interface DatabaseClusterBaseProps {

The developer assigns this attribute when creating a DatabaseCluster.

[skinny85]

@hixi-hyi I think I see where you're going with this. So we would add a property to DatabaseClusterProps, called something like instanceUpdateBehavior, whose type would be an enum with 2 members, with names like BULK (the current behavior, and so the default) and ROLLING (update the instances one-by-one by adding dependencies between them)?

Did I understand your suggestion correctly?

[hixi-hyi]

@skinny85 Yes, You know exactly what I mean.

[skinny85]

I'm glad @hixi-hyi 🙂.

Any chance of opening us a PR implementing this? Should only require adding a property here (or perhaps InstanceProps are a better place for it...?), and adding the DependsOn somewhere here.

Here's our Contributing guide: https://github.com/aws/aws-cdk/blob/master/CONTRIBUTING.md.

Thanks,
Adam

[spanierm42]

@skinny85 I added a PR for this issue some weeks ago. How long does it commonly take for the CDK maintainers to provide feedback to it? Is there anything I can do to speed up the process?

[skinny85]

@mod-enter apologies for the bad experience! Unfortunately, I'm no longer with the CDK team, so I can't review your Pull Request.

Perhaps @TheRealAmazonKendra can help with this one?

[spanierm42]

No worries, thanks for helping me anyway :)

[github-actions]

⚠️COMMENT VISIBILITY WARNING⚠️

Comments on closed issues are hard for our team to see.
If you need more assistance, please either tag a team member or open a new issue that references this one.
If you wish to keep having a conversation with other community members under this issue feel free to do so.