NOTE: The project has been migrated to slime-io/slime

---

• Overview
  • Background
  • Features
  • Function
  • Thinking
  • Architecture
  • Sample
  • Dependencies

Overview

中文

Background

In Service Mesh, in order to configure rate limit for a service, users have to face an unusually complex EnvoyFilter rate limit configuration. To solve this problem, this project introduces SmartLimiter, which can automatically convert user-submitted SmartLimiter into EnvoyFilter. Installation and Use

Features

1. easy to use, just submit SmartLimiter to achieve the purpose of service rate limit.
2. adaptive rate limit, dynamic triggering of rate limit rules according to metric.
3. Cover many scenarios, support global shared rate limit, global average rate limit, and single rate limit.

Function

1. single rate limit, each load of the service will have its own rate limit counter.
2. global shared rate limit, all loads of a service share a single rate limit counter.
3. global average rate limit, which distributes the rate limit counters equally among all loads. see function

Thinking

To get users out of the tedious EnvoyFilter configuration, we define an easy API using kubernetes CRD mechanism, the SmartLimiter resource within kubernetes. After a user submits a SmartLimiter to a kubernetes cluster, the SmartLimiter Controler generates an EnvoyFilter in conjunction with the SmartLimiter spec and service metrics.

Architecture

The main architecture of adaptive rate limit is divided into two parts, one part includes the logical transformation of SmartLimiter to EnvoyFilter, and the other part includes the acquisition of monitoring data within the cluster, including service metrics such as CPU, Memory, POD counts, etc., as detailed in architecture

Sample

When the total amount of cpu consumed by all loads of the reviews service is greater than 10, trigger a rate limit so that each load's port 9080 can only handle 10 requests per second, see example

apiVersion: microservice.slime.io/v1alpha2
kind: SmartLimiter
metadata:
  name: review
  namespace: default
spec:
  sets:
    _base:
      descriptor:
      - action:
          fill_interval:
            seconds: 1
          quota: "10"
          strategy: "single"
        condition: "{{._base.cpu.sum}}>10"
        target:
          port: 9080

Dependencies

1. In order to complete the adaptive function, we need to get the basic metrics of the service, so this service depends on prometheus, for details on how to build a simple prometheus, see prometheus
2. In order to complete the global shared rate limitation, we need a global counter, we introduced RLS, about RLS see RLS