Skip to content

OCI compute templates

The packaged catalog at src/orb/providers/oci/config/oci_templates.json contains 25 examples covering 23 shapes. orb templates generate reads this single catalog in both source checkouts and installed wheels.

For the full handler contract and acquire/status/return workflow, see OCI Compute Instances.

Catalog examples and configured templates

The shipped catalog is a portable set of shape and sizing examples. It does not contain a customer's image, compartment, subnet, NSG, SSH public key, or bootstrap script. It requires deployment configuration before use; generating examples does not make every shape ready to launch.

Location Contents
Packaged oci_templates.json Shape, Flex sizing, capacity type, boot size, and explicit ARM/HPC configuration placeholders
Deployment config.json Your provider credentials/auth mode and deployment defaults: image, compartment, subnet, NSGs, SSH public key, and optional bootstrap/tags
orb templates list output Runtime templates after local defaults have been applied, including the actual resource IDs used for a launch

Seeing a subnet or image OCID in templates list does not mean it is hard coded in the shipped catalog. Inspect your generated template file to see the saved catalog, and provider.provider_defaults.oci.template_defaults in your local config.json to see shared deployment settings. A saved per-template value overrides the shared default. Edit OCI-specific settings in the generated catalog. Generic template create/update commands do not provide an OCI-specific authoring extension. Keep deployment files local.

<replace-...> values deliberately mark incomplete configuration. ARM templates must not inherit a shared x86 image, and RDMA requires its own compatible HPC image, cluster, and availability domain. Configure each template you intend to use and check image/shape compatibility. A shared image used successfully for E5 is not proof that it supports every Intel, DenseIO, or bare metal shape.

is_active: true means the template is enabled, not that OCI compatibility or capacity has been verified. created_at and updated_at describe the template object, not the image's release date or a successful deployment. The catalog limit max_instances: 1 is an example limit, not the shape's hardware capacity.

Bootstrap (machine_bootstrap, also accepted as user_data), public keys, and environment tags are deployment choices. The example configs do not install a canary bootstrap or assume a development/production environment tag. A manual test deployment can still have its own manual-test tag and test bootstrap; these appear in its resolved list output and are not production defaults.

Shape coverage

Processor / family VM shapes Bare metal shapes
AMD E5 Standard VM.Standard.E5.Flex BM.Standard.E5.192
AMD E6 Standard VM.Standard.E6.Flex, VM.Standard.E6.Ax.Flex BM.Standard.E6.256, BM.Standard.E6.Ax.192
Intel Standard VM.Standard3.Flex, VM.Standard4.Ax.Flex BM.Standard3.64, BM.Standard4.Ax.120
Intel Optimized VM.Optimized3.Flex BM.Optimized3.36 (existing RDMA example)
Ampere Standard VM.Standard.A1.Flex, VM.Standard.A2.Flex, VM.Standard.A4.Flex, VM.Standard.A4.Ax.Flex BM.Standard.A1.160, BM.Standard.A4.48, BM.Standard.A4.Ax.48
AMD E5 DenseIO VM.DenseIO.E5.Flex BM.DenseIO.E5.128
AMD E6 DenseIO VM.DenseIO.E6.Ax.Flex BM.DenseIO.E6.Ax.192

These are examples of supported template definitions, not a promise of regional capacity. Check your region, availability domain, image compatibility, limits, and access using OCI's ListShapes API before requesting workers. The exact E6 DenseIO shape includes .Ax; there is no VM.DenseIO.E6.Flex example. Current Intel and Ampere families have Standard examples; current DenseIO examples are AMD. See Oracle's shape reference.

Existing IDs remain valid: oci-vm-flex-ondemand-small, oci-vm-flex-preemptible-small, and oci-vm-flex-preemptible-fallback-medium use E6 Standard; oci-bm-standard-ondemand uses E5 Standard. New IDs follow oci-<shape-in-lowercase-with-hyphens-without-flex>-ondemand, for example oci-vm-standard-e5-ondemand and oci-vm-denseio-e6-ax-ondemand.

Flex sizing

Each VM example explicitly declares its CPU and memory. Standard and Optimized examples start with 1 OCPU / 8 GB; DenseIO examples start with 8 OCPUs / 96 GB. Clone a template, give it a unique template_id, and edit shape_config:

{
  "shape_config": {
    "ocpus": 4,
    "memoryInGBs": 32
  }
}

This is a sizing fragment for Standard/Optimized Flex, not a complete template. ORB passes these values to OCI's launch shapeConfig. OCI enforces the shape's CPU/memory limits and permitted ratios; values must be finite and positive. machine_types: {"VM.Standard.E5.Flex": 1} selects the shape; its weight 1 does not set the CPU count. Use one shape per OCI template, so the image, architecture, sizing, and scheduling identity are unambiguous. The OCI handler selects the first entry; the map is not a multi-shape fallback list.

DenseIO Flex uses fixed CPU/memory steps. For both shipped E5 and E6 Ax examples, the allowed pairs are 8/96, 16/192, 24/288, 32/384, 40/480, and 48/576 (OCPUs/GB). ORB rejects other pairs before provisioning. Local NVMe capacity follows the selected shape configuration; boot_volume_gbs sets the separate boot disk and does not resize or mount NVMe. Treat local NVMe as temporary worker storage and configure mounts in your own bootstrap. See Oracle's NVMe guidance.

Bare metal examples are fixed shapes. Do not add shape_config, ocpus, or memory_gbs; ORB rejects Flex overrides on bare metal.

Images and deployment settings

Keep compartment, subnet, NSGs, SSH public keys, and bootstrap settings in provider_defaults.oci.template_defaults, as described in the integration guide.

Use a compatible x86_64 image for AMD/Intel. Every Ampere example explicitly has image_id: "ocid1.image.oc1..<replace-arm64-image-ocid>"; replace this on each chosen template with a compatible ARM64 image before launching. The placeholder intentionally overrides a shared x86 image default. Your worker software and bootstrap must also support ARM64. OCPU-to-vCPU conversion differs between A1 and later Ampere families; sizing is always expressed in OCI OCPUs, not an assumed universal thread count. See Arm-based Compute.

The RDMA example keeps its explicit HPC image, availability domain, cluster ID, and agent settings. Follow Compute Clusters and RDMA. Other new examples launch independent on-demand workers using the same existing OCI handler. To use preemptible capacity on a compatible VM shape, clone an existing preemptible example and replace its shape, sizing, and image; fallback changes capacity type for that same shape, not processor architecture.

Pricing remains an estimate. Shapes without configured reference rates report unknown compute cost; supply pricing overrides for your applicable rates. Catalog coverage and offline tests do not establish a live launch on every shape.

Validate a selected template

After filling deployment defaults and the selected template's image/settings:

export ORB_CONFIG_DIR="$PWD/config"
orb --config "$ORB_CONFIG_DIR/config.json" templates generate
orb --config "$ORB_CONFIG_DIR/config.json" templates list
orb --config "$ORB_CONFIG_DIR/config.json" templates validate oci-vm-standard-e5-ondemand
orb --config "$ORB_CONFIG_DIR/config.json" machines request oci-vm-standard-e5-ondemand 1 --dry-run

Generation writes examples; edit your generated catalog afterward. Do not regenerate over deployment edits. For a real canary, use the acquire/status/return procedure in the integration guide and confirm cleanup independently in OCI.