What Does This Do

The first AI Guard evaluation can start a shaded OkHttp connection-pool worker while the ai_guard span is active. Thread instrumentation captures that context and keeps it active for the worker’s lifetime.

The evaluation span itself finishes correctly, but pool maintenance is not part of the request and should not retain its context. Smoke continuation diagnostics exposed this unintended propagation.

flowchart LR
  A[AI Guard evaluation] --> B[Synchronous HTTP call]
  B -->|Before: captures evaluation context| C[Long-lived pool worker]
  B -->|Fixed: async propagation disabled| D[Pool worker without request context]
  D --> E[Restore previous propagation setting]

  classDef broken fill:#fee2e2,stroke:#dc2626
  classDef healthy fill:#dcfce7,stroke:#16a34a
  class C broken
  class D,E healthy

Motivation

Disable async propagation only around AI Guard’s synchronous HTTP execution, restoring the previous setting in finally.

Additional Notes

Contributor Checklist

• Format the title according to the contribution guidelines
• Assign the type: and (comp: or inst:) labels in addition to any other useful labels
• Avoid using close, fix, or any linking keywords when referencing an issue
  Use solves instead, and assign the PR milestone to the issue
• Update the CODEOWNERS file on source file addition, migration, or deletion
• Update public documentation with any new configuration flags or behaviors
• Once approved, use merge queue to merge the PR

Jira ticket: [PROJ-IDENT]