OCI CLI ignores shape-config memory_in_gbs parameter for Flex shapes, always uses default memory allocation

[timmcfadden]

Bug Description:

Summary

The OCI CLI compute instance launch command completely ignores the memory_in_gbs parameter in --shape-config for Flex shapes (VM.Standard.E4.Flex,
VM.Standard3.Flex, VM.Standard.A1.Flex), always defaulting to the shape's default memory allocation regardless of the specified value.

Environment

• OCI CLI Version: 3.64.1 (Oracle-PythonCLI/3.64.1)
• Platform: Linux 6.12.0-102.36.5.2.el9uek.x86_64
• Region: us-ashburn-1 (iad)
• Shapes Tested: VM.Standard.E4.Flex, VM.Standard3.Flex, VM.Standard.A1.Flex

Expected Behavior

When launching an instance with --shape-config '{"ocpus": 2, "memory_in_gbs": 4}', the instance should be created with exactly 4 GB of memory.

Actual Behavior

The instance is created with the shape's default memory allocation:

• VM.Standard.E4.Flex: Always gets 32 GB (16 GB per OCPU default)
• VM.Standard3.Flex: Always gets 32 GB (16 GB per OCPU default)
• VM.Standard.A1.Flex: Always gets 12 GB (6 GB per OCPU default)

Steps to Reproduce

1. Launch instance with JSON inline:
   oci compute instance launch
   --compartment-id [COMPARTMENT_ID]
   --availability-domain [AD]
   --display-name "test-memory-4gb"
   --image-id [COMPATIBLE_IMAGE_ID]
   --subnet-id [SUBNET_ID]
   --shape VM.Standard.E4.Flex
   --shape-config '{"ocpus": 2, "memory_in_gbs": 4}'

2. Launch instance with JSON file:
   echo '{"ocpus": 2, "memory_in_gbs": 4}' > shape-config.json
   oci compute instance launch
   --shape VM.Standard.E4.Flex
   --shape-config file://shape-config.json
   [other parameters...]

3. Attempt resize on stopped instance:
   oci compute instance update
   --instance-id [INSTANCE_ID]
   --shape-config file://shape-config.json --force

4. Verify memory allocation:
   oci compute instance get
   --instance-id [INSTANCE_ID]
   --query 'data.shape_config.memory_in_gbs'

Result: All methods return default memory (32.0 GB for E4.Flex) instead of requested 4 GB.

Workaround

Using Terraform with the OCI provider correctly allocates the specified memory:
resource "oci_core_instance" "test" {
shape = "VM.Standard.E4.Flex"
shape_config {
ocpus = 2
memory_in_gbs = 4 # This works correctly
}
}

Impact

• High: Prevents accurate resource provisioning for migration and automation scenarios
• Cost: Forces over-allocation of memory resources leading to unnecessary costs
• Automation: Breaks infrastructure-as-code deployments that require specific memory configurations

Additional Evidence

• Shape constraints confirm memory_in_gbs range is valid (min: 1.0, max: 64.0 per OCPU for E4.Flex)
• Work requests complete successfully but memory allocation remains unchanged
• Issue persists across multiple regions and image types
• Same JSON configuration works correctly with Terraform OCI provider

This appears to be a CLI-specific issue where the shape-config memory parameter is not being properly passed to the underlying API, while the
Terraform provider correctly implements the same functionality.