feat(aws): add support for aws_sagemaker_endpoint_configuration

[muhammadahmed-01]

Objective:

Add support for AWS SageMaker Endpoint Configuration pricing. Fixes #2733.

Pricing details:

Overview

This PR adds comprehensive cost estimation for aws_sagemaker_endpoint_configuration resources, covering both serverless and provisioned instance-based inference endpoints.

Supported Components

Serverless Inference:

• Compute Duration: Charged per second based on memory allocation and inference duration

  • Usage parameter: monthly_inference_duration_secs

• Provisioned Concurrency - Warm: Charged for maintaining warm container instances

  • Usage parameter: monthly_provisioned_concurrency_usage_secs
  • Represents readiness cost for provisioned concurrency

• Provisioned Concurrency - Execution: Charged for actual inference execution on provisioned capacity

  • Usage parameter: monthly_provisioned_concurrency_inference_duration_secs
  • Scaled by variant memory allocation

• Data Transfer (In/Out): Per-GB charges for data processed through endpoints

  • Usage parameters: monthly_data_processed_in_gb, monthly_data_processed_out_gb
  • Separate billing for inbound and outbound traffic

Provisioned Instances:

• Instance Hours: Hourly instance charges for dedicated inference capacity

  • Usage parameter: instance_hrs (defaults to 730 hours/month)

• General Purpose Storage: Prorated storage for model artifacts and code

  • Usage parameter: storage_days (defaults to 30)
  • Formula: (storage_gb × storage_days) / 30

• Data Transfer (In/Out): Same as serverless variant

Shadow Variants

Shadow variants are billed identically to production variants, enabling cost estimation for A/B testing and gradual traffic migration scenarios.

Multiple Variants

Endpoint configurations with multiple variants accurately reflect combined costs across all variants.

Test Coverage

• Golden file testing with 5 endpoint configuration scenarios:
  • Instance-based endpoints with multiple variants
  • Serverless endpoints (basic and with provisioned concurrency)
  • Multiple serverless variants with different memory sizes
  • Shadow variant configuration

Usage File Example

aws_sagemaker_endpoint_configuration.my_serverless_config:
    instance_hrs: 100
    monthly_inference_duration_secs: 500000
    monthly_provisioned_concurrency_usage_secs: 200000
    monthly_provisioned_concurrency_inference_duration_secs: 100000
    monthly_data_processed_in_gb: 20
    monthly_data_processed_out_gb: 10
    storage_days: 15

Status:

• Generated the resource files
• Updated the internal/resources file
• Updated the internal/provider/terraform/.../resources file
• Added usage parameters to infracost-usage-example.yml
• Added test cases without usage-file
• Added test cases with usage-file
• Compared test case output to cloud cost calculator
• Created a PR to update "Supported Resources" in the docs

Issues:

None

[muhammadahmed-01]

Hi @aliscott,

I've opened this PR to provide a proper implementation for aws_sagemaker_endpoint_configuration.

I appreciate your feedback on the previous attempt. I had to step away for a while due to personal reasons, but I’ve now completed a total manual overhaul of the logic to ensure it meets Infracost’s quality standards.

A few points on the manual refinement in this version:

• Architectural Consistency: I used the lambda_function implementation as a structural reference to ensure consistency in how we handle usage-based compute.
• Deep-Dive Pricing Mapping: I manually mapped the logic for both Serverless (including Provisioned Concurrency) and Provisioned Instances. I specifically ensured that Shadow Variants are captured, as they contribute to the bottom line but are often overlooked in basic estimations.
• Calculator Validation: Every test case in the golden files has been manually cross-referenced against the AWS Sagemaker AI pricing page to ensure accuracy across various memory and instance configurations.
• Guideline Compliance: I have performed a thorough pass against the contributing.md and resource adding and resource mapping guidelines.

I'm back on this now and committed to seeing this through to completion. I'd appreciate a fresh look whenever you have a moment.

[muhammadahmed-01]

Hey @aliscott, did you get a chance to review this PR?