Hypothesis:
If we move to Rust for the function-evaluator, our service will be more stable, mightier, and more efficient in resource consumption.
Acceptance Criteria
- We flip the production evaluators from Node to Rust
Goals & Success Metrics
- Rust evaluators are in production, replacing Node-based ones
- Monotonic memory increase is no longer observed in production function evaluators
- Other performance metrics (evaluation runtime, memory consumption, CPU usage, etc.) are AT LEAST as good as current Node implementation
This is a follow-on Epic to T343720: If we create a working evaluator service in Rust, we can measure its performance and stability characteristics and plan to productionize it in a subsequent quarter, the work scheduled for 2026Q1 (July–September 2025); provisionally, this may be scheduled for 2026Q2 (October–December 2025), but is not confirmed yet.