壹、入围方案
Sentinel
阿里出品，Spring Cloud Alibaba限流组件，目前持续更新中
自带Dashboard，可以查看接口Qps等，并且可以动态修改各种规则
流量控制，直接限流、冷启动、排队
熔断降级，限制并发限制数和相应时间
系统负载保护，提供系统级别防护，限制总体CPU等
主要核心：资源，规则(流量控制规则、熔断降级规则、系统保护规则、来源访问控制规则 和 热点参数规则。)，和指标
文档非常清晰和详细，中文
支持动态规则(推模式和拉模式)
Hystrix
Netflix出品，Spring Cloud Netflix限流组件，已经停止新特性开发，只进行bug修复，最近更新为2018年，功能稳定
有简单的dashboard页面
以隔离和熔断为主的容错机制，超时或被熔断的调用将会快速失败，并可以提供 fallback 机制的初代熔断框架，异常统计基于滑动窗口
resilience4j
是一款轻量、简单，并且文档非常清晰、丰富的熔断工具。是Hystrix替代品，实现思路和Hystrix一致，目前持续更新中
需要自己对micrometer、prometheus以及Dropwizard metrics进行整合
CircuitBreaker 熔断
Bulkhead 隔离
RateLimiter QPS限制
Retry 重试
TimeLimiter 超时限制
Cache 缓存
自己实现(基于Guava)
基于Guava的令牌桶，可以轻松实现对QPS进行限流
贰、技术对比
叁、应用改造
3.1、sentinel
3.1.1、引入依赖
com.alibaba.cloud
spring-cloud-starter-alibaba-sentinel
2.0.3.RELEASE
3.1.2、改造接口或者service层
@SentinelResource(value = "allInfos",fallback = "errorReturn")
@Target({ElementType.METHOD, ElementType.TYPE})
@Retention(RetentionPolicy.RUNTIME)
@Inherited
public @interface SentinelResource {
//资源名称
String value() default "";
//流量方向
EntryType entryType() default EntryType.OUT;
//资源类型
int resourceType() default 0;
//异常处理方法
String blockHandler() default "";
//异常处理类
Class>[] blockHandlerClass() default {};
//熔断方法
String fallback() default "";
//默认熔断方法
String defaultFallback() default "";
//熔断类
Class>[] fallbackClass() default {};
//统计异常
Class extends Throwable>[] exceptionsToTrace() default {Throwable.class};
//忽略异常
Class extends Throwable>[] exceptionsToIgnore() default {};
}
@RequestMapping("/get")
@ResponseBody
@SentinelResource(value = "allInfos",fallback = "errorReturn")
public JsonResult allInfos(HttpServletRequest request, HttpServletResponse response, @RequestParam Integer num){
try {
if (num % 2 == 0) {
log.info("num % 2 == 0");
throw new BaseException("something bad with 2", 400);
}
return JsonResult.ok();
} catch (ProgramException e) {
log.info("error");
return JsonResult.error("error");
}
}3.1.3、针对接口配置熔断方法或者限流方法
默认过滤拦截所有Controller接口
/**
* 限流，参数需要和方法保持一致
* @param request
* @param response
* @param num
* @return
* @throws BlockException
*/
public JsonResult errorReturn(HttpServletRequest request, HttpServletResponse response, @RequestParam Integer num) throws BlockException {
return JsonResult.error("error 限流" + num );
}
/**
* 熔断，参数需要和方法保持一直，并且需要添加BlockException异常
* @param request
* @param response
* @param num
* @param b
* @return
* @throws BlockException
*/
public JsonResult errorReturn(HttpServletRequest request, HttpServletResponse response, @RequestParam Integer num,BlockException b) throws BlockException {
return JsonResult.error("error 熔断" + num );
}注意也可以不配置限流或者熔断方法。通过全局异常去捕获UndeclaredThrowableException或者BlockException避免大量的开发量
3.1.4、接入dashboard
spring:
cloud:
sentinel:
transport:
port: 8719
dashboard: localhost:8080
3.1.5、规则持久化和动态更新
接入配置中心如：zookeeper等等，并对规则采用推模式
3.2、hystrix
3.2.1、引入依赖
org.springframework.boot
spring-boot-starter-actuator
org.springframework.cloud
spring-cloud-starter-netflix-hystrix-dashboard
2.0.4.RELEASE
org.springframework.cloud
spring-cloud-starter-netflix-hystrix
2.0.4.RELEASE
3.2.2、改造接口
@HystrixCommand(fallbackMethod = "timeOutError")
@Target({ElementType.METHOD})
@Retention(RetentionPolicy.RUNTIME)
@Inherited
@Documented
public @interface HystrixCommand {
String groupKey() default "";
String commandKey() default "";
String threadPoolKey() default "";
String fallbackMethod() default "";
HystrixProperty[] commandProperties() default {};
HystrixProperty[] threadPoolProperties() default {};
Class extends Throwable>[] ignoreExceptions() default {};
ObservableExecutionMode observableExecutionMode() default ObservableExecutionMode.EAGER;
HystrixException[] raiseHystrixExceptions() default {};
String defaultFallback() default "";
}
@RequestMapping("/get")
@ResponseBody
@HystrixCommand(fallbackMethod = "fallbackMethod")
public JsonResult allInfos(HttpServletRequest request, HttpServletResponse response, @RequestParam Integer num){
try {
if (num % 3 == 0) {
log.info("num % 3 == 0");
throw new BaseException("something bad whitch 3", 400);
}
return JsonResult.ok();
} catch (ProgramException | InterruptedException exception) {
log.info("error");
return JsonResult.error("error");
}
}
3.2.3、针对接口配置熔断方法
/**
* 该方法是熔断回调方法，参数需要和接口保持一致
* @param request
* @param response
* @param num
* @return
*/
public JsonResult fallbackMethod(HttpServletRequest request, HttpServletResponse response, @RequestParam Integer num) {
response.setStatus(500);
log.info("发生了熔断！！");
return JsonResult.error("熔断");
}3.2.4、配置默认策略
hystrix:
command:
default:
execution:
isolation:
strategy: THREAD
thread:
# 线程超时15秒,调用Fallback方法
timeoutInMilliseconds: 15000
metrics:
rollingStats:
timeInMilliseconds: 15000
circuitBreaker:
# 10秒内出现3个以上请求(已临近阀值),并且出错率在50%以上,开启断路器.断开服务,调用Fallback方法
requestVolumeThreshold: 3
sleepWindowInMilliseconds: 10000
3.2.5、接入监控
曲线：用来记录2分钟内流量的相对变化，我们可以通过它来观察到流量的上升和下降趋势。
集群监控需要用到注册中心
3.3、resilience4j
3.3.1、引入依赖
dependency>
org.springframework.boot
spring-boot-starter-web
org.springframework.boot
spring-boot-starter-test
test
io.github.resilience4j
resilience4j-spring-boot2
1.6.1
io.github.resilience4j
resilience4j-bulkhead
1.6.1
io.github.resilience4j
resilience4j-ratelimiter
1.6.1
io.github.resilience4j
resilience4j-timelimiter
1.6.1
可以按需要引入：bulkhead，ratelimiter，timelimiter等
3.3.2、改造接口
@RequestMapping("/get")
@ResponseBody
//@TimeLimiter(name = "BulkheadA",fallbackMethod = "fallbackMethod")
@CircuitBreaker(name = "BulkheadA",fallbackMethod = "fallbackMethod")
@Bulkhead(name = "BulkheadA",fallbackMethod = "fallbackMethod")
public JsonResult allInfos(HttpServletRequest request, HttpServletResponse response, @RequestParam Integer num){
log.info("param----->" + num);
try {
//Thread.sleep(num);
if (num % 2 == 0) {
log.info("num % 2 == 0");
throw new BaseException("something bad with 2", 400);
}
if (num % 3 == 0) {
log.info("num % 3 == 0");
throw new BaseException("something bad whitch 3", 400);
}
if (num % 5 == 0) {
log.info("num % 5 == 0");
throw new ProgramException("something bad whitch 5", 400);
}
if (num % 7 == 0) {
log.info("num % 7 == 0");
int res = 1 / 0;
}
return JsonResult.ok();
} catch (BufferUnderflowException e) {
log.info("error");
return JsonResult.error("error");
}
}3.3.3、针对接口配置熔断方法
/**
* 需要参数一致，并且加上相应异常
* @param request
* @param response
* @param num
* @param exception
* @return
*/
public JsonResult fallbackMethod(HttpServletRequest request, HttpServletResponse response, @RequestParam Integer num, BulkheadFullException exception) {
return JsonResult.error("error 熔断" + num );
}3.3.4、配置规则
resilience4j.circuitbreaker:
instances:
backendA:
registerHealthIndicator: true
slidingWindowSize: 100
backendB:
registerHealthIndicator: true
slidingWindowSize: 10
permittedNumberOfCallsInHalfOpenState: 3
slidingWindowType: TIME_BASED
minimumNumberOfCalls: 20
waitDurationInOpenState: 50s
failureRateThreshold: 50
eventConsumerBufferSize: 10
recordFailurePredicate: io.github.robwin.exception.RecordFailurePredicate
resilience4j.retry:
instances:
backendA:
maxRetryAttempts: 3
waitDuration: 10s
enableExponentialBackoff: true
exponentialBackoffMultiplier: 2
retryExceptions:
- org.springframework.web.client.HttpServerErrorException
- java.io.IOException
ignoreExceptions:
- io.github.robwin.exception.BusinessException
backendB:
maxRetryAttempts: 3
waitDuration: 10s
retryExceptions:
- org.springframework.web.client.HttpServerErrorException
- java.io.IOException
ignoreExceptions:
- io.github.robwin.exception.BusinessException
resilience4j.bulkhead:
instances:
backendA:
maxConcurrentCalls: 10
backendB:
maxWaitDuration: 10ms
maxConcurrentCalls: 20
resilience4j.thread-pool-bulkhead:
instances:
backendC:
maxThreadPoolSize: 1
coreThreadPoolSize: 1
queueCapacity: 1
resilience4j.ratelimiter:
instances:
backendA:
limitForPeriod: 10
limitRefreshPeriod: 1s
timeoutDuration: 0
registerHealthIndicator: true
eventConsumerBufferSize: 100
backendB:
limitForPeriod: 6
limitRefreshPeriod: 500ms
timeoutDuration: 3s
resilience4j.timelimiter:
instances:
backendA:
timeoutDuration: 2s
cancelRunningFuture: true
backendB:
timeoutDuration: 1s
cancelRunningFuture: false配置的规则可以被代码覆盖
3.3.5、配置监控
如grafana等
肆、关注点
是否需要过滤部分异常
是否需要全局默认规则
可能需要引入其他中间件
k8s流量控制
规则存储和动态修改
接入改造代价
