Will AWS OpenSearch Service eventually block setting plugins.ml_commons.only_run_on_ml_node to false?

Based on the available documentation, there is no indication that AWS plans to formally block the plugins.ml_commons.only_run_on_ml_node setting in future versions of Amazon OpenSearch Service. However, the documentation explicitly advises against disabling this setting (setting it to false) on production clusters, stating that the option to disable this safeguard is intended for facilitating development purposes only.

The guidance is clear that production clusters should be using connectors for ML functionality. While the setting isn't listed among the unsupported ML Commons settings (like plugins.ml_commons.allow_registering_model_via_url and plugins.ml_commons.allow_registering_model_via_local_file, which are explicitly not supported), the strong recommendation against disabling it in production suggests this approach carries inherent risks.

Regarding your specific concerns about whether this is a "reasonable medium-term workaround or a ticking time bomb" - the documentation doesn't provide information about future enforcement plans or whether the setting might be ignored or overridden in updates. However, the fact that AWS explicitly discourages this configuration in production environments suggests it's not intended as a supported production pattern, even if it's technically possible today.

For a more stable long-term solution aligned with AWS's recommended architecture, you would need to work with your organization to get approval for the supported connector-based approach using services like Amazon Bedrock or SageMaker, as these represent the officially supported path for ML functionality in production OpenSearch Service clusters.
Sources
Machine learning for Amazon OpenSearch Service - Amazon OpenSearch Service