[ML] Add xpack.ml.model_graph_validation_enabled cluster setting

[edsavage]

Summary

Adds a dynamic node-scope cluster setting to control TorchScript model graph validation in the pytorch_inference native process.

PUT _cluster/settings
{
  "persistent": {
    "xpack.ml.model_graph_validation_enabled": false
  }
}

When set to false, the pytorch_inference process is launched with --skipModelValidation, bypassing the operation allowlist check. Default is true (validation enabled).

This provides an operator-accessible escape hatch for all deployment types (self-managed, Cloud, serverless) via the cluster settings API.

Changes

• MachineLearningField.java — defines xpack.ml.model_graph_validation_enabled (dynamic, node-scope)
• MachineLearning.java — registers the setting
• NativePyTorchProcessFactory.java — reads the setting, subscribes to dynamic updates, passes to builder
• PyTorchBuilder.java — adds --skipModelValidation to the command when validation is disabled
• PyTorchBuilderTests.java — updated existing tests + new testBuildWithValidationDisabled

Cross-repo dependency

• ml-cpp: elastic/ml-cpp#3013 adds the --skipModelValidation CLI flag to the pytorch_inference binary

Test plan

• PyTorchBuilderTests pass
• Integration: set xpack.ml.model_graph_validation_enabled: false, deploy a model, verify WARN log "Model graph validation SKIPPED"
• Integration: set back to true, deploy a model, verify validation runs normally

Depends on: elastic/ml-cpp#3013

Made with Cursor

[Copilot]

Pull request overview

Adds an operator-configurable ML cluster setting to control TorchScript model graph validation for the pytorch_inference native process, allowing operators to bypass the operation allowlist check by launching the process with --skipModelValidation.

Changes:

• Introduces xpack.ml.model_graph_validation_enabled as a dynamic node-scope cluster setting (default true).
• Wires the setting into NativePyTorchProcessFactory and passes it into PyTorchBuilder.
• Extends PyTorchBuilder to append --skipModelValidation when validation is disabled, with corresponding unit test updates.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/ml/MachineLearningField.java	Defines the new cluster setting and documents its intent/security implications.
x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/MachineLearning.java	Registers the new setting with the ML plugin.
x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/pytorch/process/NativePyTorchProcessFactory.java	Reads the setting and applies it to new pytorch_inference process launches; subscribes to dynamic updates.
x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/pytorch/process/PyTorchBuilder.java	Adds --skipModelValidation flag emission when validation is disabled.
x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/inference/pytorch/process/PyTorchBuilderTests.java	Updates existing expectations and adds coverage for the disabled-validation command-line.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)

[valeriy42]

LGTM in general. Just a few comments. It would be great to have it in 9.4 alongside the main change, but I am not sure whether this is still possible.

[valeriy42]

I think, you don't need this test. It essentially tests Elasticsearch settings infrastructure itself.

[valeriy42]

1. Yesterday was FF for 9.4. Is it still possible to add it?
2. Maybe calling it xpack.ml.trained_models.graph_validation_enabled would be more along the usual naming conventions.

[edsavage]

I think we may have just missed the boat for 9.4.0, but it will be there for the 9.4.1 patch release.