Amazon DocumentDB Serverless is now available
I am Iwasa.
Amazon DocumentDB is a MongoDB compatible fully managed database.
There are "instance-based clusters" that manage instances and "Elastic clusters" that don't require instance management.
For instance-based clusters, you need to select instances, and until now you had to select and manage instance classes (such as db.r6g.xlarge) from memory-optimized instances, NVMe-backed instances, etc.
With this morning's update, you can now select "serverless" as a new instance class.
The concept is the same as Aurora Serverless v2, where you only specify the minimum and maximum capacity units, and leave the control to the cluster.
The capacity unit is called DCU (DocumentDB Capacity Unit), with 1 DCU consisting of 2 GiB of memory and corresponding CPU, which is the same concept as Aurora Serverless v2.
The official documentation is as follows, which closely follows the Aurora Serverless v2 page.
I tried it in the Tokyo region, so let me introduce it.
Creating a Serverless Instance
For the creation method, you can now select serverless in the DocumentDB instance-based cluster's instance configuration.
This cannot be specified in Elastic clusters that don't manage instances.
When you select serverless instance, you can select the number of instances as usual, and additionally specify a range for the upper and lower limits of DCU.
As mentioned earlier, 1 DCU is equivalent to 2 GiB, and can be specified in units of 0.5 DCU.
Other storage configurations and connection/authentication settings are the same as before. Like Aurora, you just specify the DCU for computing and leave the rest to be managed for you.
When specified during the creation of a new cluster, all instances were configured as serverless.
Below is what it looks like when connecting. It's the same as before.
### Adding serverless instances to a cluster configured with existing provisioned instances
Above, we specified serverless instances for a new cluster, but it's also possible to add serverless instances to clusters that have been configured with only traditional provisioned instances.
Earlier, we specified the DCU range when creating a cluster, but when adding to an existing cluster, you can specify the DCU range when adding a new serverless instance for the first time as follows.
Engine version 5.0.0 or higher
There is a prerequisite for use: when using serverless instances, you must specify engine version 5.0.0 or higher.
For version 4.0.0 or lower, I couldn't specify it as shown below.
When applying to existing instances, please note that you may need to upgrade the engine version first.
Zero scaling is not possible
A while ago, Aurora Serverless started supporting zero scaling which reduces DCU to 0[1], but as of today, DocumentDB Serverless does not support this. The minimum DCU is 0.5.
Mixing serverless and provisioned instances within a cluster
Similar to Aurora Serverless clusters, you can mix traditional provisioned instances and serverless instances within a cluster.
Below, I changed the replica instance to provisioned after creating a serverless cluster.
This can accommodate use cases such as making only the primary serverless.
Pricing
Serverless pricing has been added to the DocumentDB pricing page.
In the Tokyo region, I believe the minimum traditional instance was db.t4g.medium at $0.115 per hour.
Serverless starts from $0.0992 per hour per DCU, which means 0.5 DCU would be $0.0496 per hour.
Roughly speaking, while provisioned would cost from about $83/month, you can now use it from about $36/month.
Being able to use it at an even smaller and cheaper price than before seems good for development environments.
Conclusion
Today, since Amazon DocumentDB Serverless became available, I actually created one, added it to an existing cluster, and calculated the cost.
The usage image is the same as Aurora Serverless v2, and it seems good to adopt when workloads are unpredictable or when downtime during scale-up/down is not acceptable.
