Feature: Google SRE Client-Side Adaptive Throttling

[cinar]

Description: Implement Google's mathematical approach to handling overload (detailed in the SRE Book). Instead of relying strictly on a circuit breaker that entirely opens or closes, adaptive throttling calculates a dynamic probability of dropping requests locally before they even hit the network, based on the server's recent success/failure ratio.

• Architecture & Implementation Requirements:

  • State Tracking: Maintain two counters (ideally within the sync.Pool or a lightweight ring buffer): requests (attempted calls) and accepts (successful calls).
  • The Math (Proportional Probability): Before initiating a request, calculate the rejection probability using Google's formula: max(0, (requests - K * accepts) / (requests + 1)) (Where K is a configurable multiplier, defaulting to 2).
  • Execution Logic: If P_{drop} > 0, generate a random float between 0.0 and 1.0. If the random number is less than P_{drop}, immediately return ErrThrottledLocally without making the network call.

• Acceptance Criteria:

  • Must be mathematically accurate and update counters atomically to avoid race conditions.
  • Must gracefully decay older stats so the client eventually resumes sending traffic when the server recovers.
  • Must implement clean code changes without over-refactoring the existing Do loop.

• Resources:

  • Google SRE Book: Handling Overload (Client-Side Throttling)