Instance sizing

Instance sizing is the CPU and memory allocated to each running instance of an Akka service container, chosen from a fixed catalog of named instance types.

Each instance type is named g<generation>.<shape>.<size>, for example g1.general.small or g1.memory.large. The shape hints at the workload it suits:

  • general — a balanced CPU-to-memory ratio, the right default for most services.

  • memory — extra memory relative to CPU, for services with large in-memory entity state or high concurrent load.

  • compute — extra CPU relative to memory, for services doing heavier per-request processing with modest state.

Within a shape, size scales from x-small up to large.

The instance type catalog

Instance type Shape CPU Memory

g1.compute.x-small

compute

800m

1280Mi

g1.compute.small

compute

1600m

2560Mi

g1.general.x-small

general

400m

1280Mi

g1.general.small (platform default)

general

800m

2560Mi

g1.memory.x-small

memory

400m

1920Mi

g1.memory.small

memory

800m

3840Mi

g1.memory.medium

memory

1600m

6144Mi

g1.memory.large

memory

2000m

12288Mi

Every project has a platform default, applied to any service that does not request a specific instance type.

Viewing what your project may use

An organization can restrict which instance types its projects are allowed to deploy with, and set its own default in place of the platform default. Both are scoped: they can apply to an entire organization, or be narrowed to a region, a project, or a single service.

See what your project is allowed to use, and which type applies as the default, with:

akka projects instance-types list

Narrow the resolution to a specific region or service:

akka projects instance-types list --region us-east-1 --service shopping-cart

The output always marks which instance type is the default for the scope you queried.

Setting a service’s instance type

If you do not set an instance type, a service runs on its project’s default. To pin a service to a specific instance type instead, set it at deploy time:

akka service deploy shopping-cart shopping-cart:1.0 --instance-type g1.memory.small

The requested type must be one your organization allows; otherwise the deployment is rejected.

The same setting is available in a service descriptor, under resources.instanceType:

name: shopping-cart
service:
  image: my-registry/shopping-cart:1.0
  resources:
    instanceType: g1.memory.small
    runtime:
      mode: embedded

Apply the descriptor with akka service apply -f shopping-cart.yaml, as described in using service descriptors.

A change to a service’s instance type takes effect on its next deployment. Instances already running keep their current size until they are replaced by a rolling update.

Requesting a change to your allowlist or default

Changing which instance types a project, region, or service is allowed to use, or changing the scoped default, is not currently self-service. Contact the Akka Support Portal and specify:

  • The organization, project, region, or service the change applies to.

  • The instance type or types you need allowed.

  • Whether it should also become the default for that scope.

Cost implications

A larger instance type increases the CPU and memory reserved for every running instance of that service, and that cost multiplies across however many instances autoscaling keeps running at peak. Choosing g1.memory.large for a service that also scales to a high --max-instances compounds both factors.

Size services individually rather than raising a project’s default: give the services that need more headroom a larger instance type, and leave the rest on the default. Sizing choices are part of the cost forecast you agree with your Akka team during production readiness.

Related documentation