Dynamic Rate Limiting Solution Based on Sentinel Secondary Development

This article introduces a dynamic rate limiting solution based on Sentinel secondary development, covering what dynamic rate limiting is, why it is needed, and its implementation principles.

Background: With the growth of internet services and business, high-concurrency systems require rate limiting to control traffic and ensure system availability. The current solution uses VSentinel with QPS rate limiting and hot parameter rate limiting. However, for a game recommendation system with 600+ scenarios and 100+ scene calls per interface, existing solutions have problems: parameter-level rate limiting cannot achieve fine-grained control per scenario; interface-level rate limiting cannot guarantee core scenario availability; threshold adjustments require manual maintenance; and traffic changes in real-time without dynamic adjustment capability.

Dynamic Rate Limiting: Unlike fixed threshold configuration, dynamic rate limiting configures parameter priority levels and dynamically adjusts rate limiting thresholds based on parameter priority. When traffic exceeds the threshold, high-priority parameters are guaranteed to pass while low-priority parameters are blocked.

Sentinel Architecture: Sentinel is an open-source flow control component from Alibaba for distributed service architectures. It implements rate limiting through a responsibility chain pattern composed of ProcessorSlot objects, including NodeSelectorSlot, ClusterBuilderSlot, StatisticSlot for data statistics, and FlowSlot, ParamFlowSlot, DegradeSlot for rate limiting and degradation.

Sliding Window Counter Algorithm: The article explains the fixed counter algorithm and its traffic burst problem, then introduces the sliding window counter algorithm that divides time into finer granular intervals to solve the burst issue.

Implementation: The dynamic rate limiting implementation follows three steps: data statistics using sliding window to count parameter-level traffic; rule management using ZK for synchronization; and flow validation using parameter priority threshold calculation with subdivided windows plus dynamic prediction, plus double-check mechanism to prevent traffic bursts.

Architecture: Reuses Sentinel's responsibility chain + SPI architecture, packaged as an independent SDK without affecting the original Sentinel processing flow.