集成监控Executor的线程池参数

实现方式采用spring的bean后置装载的接口

代码：

public class ExecutorMetricWapper implements BeanPostProcessor {       @Override     public Object postProcessAfterInitialization(Object bean, String beanName) throws BeansException {         if ((bean instanceof Executor)) {             ExecutorServiceMetrics.monitor(Metrics.globalRegistry, (Executor) bean, beanName, Collections.emptyList());             LOGGER.info("Executor beanName:{} Has been registered to monitor", beanName);             return bean;         }         return bean;     }       private static final Logger LOGGER = LoggerFactory.getLogger(ExecutorMetricWapper.class); }

 

效果：

只是公司内的grafana监控图

 

指标的含义：

spring自带的ThreadPoolTaskExecutor线程池（至于spring的Executor为什么参数没有自定实现的Executor线程池全，这问题是由于实现机制的原因）

指标	含义	推荐expr
executor_idle_seconds_count	任务闲置次数	执行器平均闲置秒数：increase(executor_idle_seconds_sum[5m])/increase(executor_idle_seconds_count[5m])
executor_idle_seconds_sum	任务耗时	同上
executor_idle_seconds_max	任务空闲秒数（最大值）	--
executor_execution_seconds_count	任务执行次数	执行器平均执行秒数:increase(executor_execution_seconds_sum[5m])/increase(executor_execution_seconds_count[5m])
executor_execution_seconds_sum	任务耗时	同上
executor_execution_seconds_max	任务执行秒数（最大值）	--

自定义的Executor线程池

指标	含义	推荐expr
executor_completed_tasks_total	完成的任务	完成的任务/分钟：increase(executor_completed_tasks_total[1m])
executor_idle_seconds_count	执行器闲置次数	执行器平均闲置秒数：increase(executor_idle_seconds_sum[5m])/increase(executor_idle_seconds_count[5m])
executor_idle_seconds_sum	执行器耗时	同上
executor_idle_seconds_max	执行器空闲秒数（最大值）	--
executor_active_threads	执行器活跃的线程数	--
executor_seconds_count	执行器执行次数	执行器平均执行秒数:increase(executor_seconds_sum[5m])/increase(executor_seconds_count[5m])
executor_seconds_sum	执行器耗时	同上
executor_seconds_max	执行器执行秒数（最大值）	--
executor_queued_tasks	排队的任务	队列占用率：100*(1-executor_queue_remaining_tasks/ (executor_queue_remaining_tasks + executor_queued_tasks ))
executor_queue_remaining_tasks	排队的剩余任务	同上
executor_pool_size_threads	执行器的线程池大小	--
executor_rejected_tasks_total	拒绝的任务	拒绝的任务/分钟：increase(executor_rejected_tasks_total[1m])
executor_scheduled_cron_total	执行器预定的cron	--
executor_scheduled_once_total	执行器预定一次大小	--
executor_scheduled_repetitively_total	执行器重复执行次数	--

参照：INFRA|springboot2.X统一配置面板的grpc统计模块

 

其他的监控也是类似，甚至可以自己来写实现

从这个问题主要是引出监控的事情，要是想要对一个线程池进行matrix监控，那么当前比较low的实现方法如下

 

@Service public class ThreadPoolMonitor implements InitializingBean {       @Resource(name = "consumerExecutor")     ThreadPoolTaskExecutor consumerExecutor;     @Resource     ConnectionPool connectionPool;       private static final Iterable<Tag> consumerExecutorTAG = Collections.singletonList(Tag.of("thread.pool.name", "consumerExecutor"));     private static final Iterable<Tag> okHttpExecutorTAG = Collections.singletonList(Tag.of("thread.pool.name", "okHttpExecutor"));     private final ScheduledExecutorService scheduledExecutor = Executors.newSingleThreadScheduledExecutor();       private Runnable monitor = () -> {         //这里需要捕获异常,尽管实际上不会产生异常,但是必须预防异常导致调度线程池线程失效的问题         try {             Metrics.gauge("thread.pool.core.size", consumerExecutorTAG, consumerExecutor, ThreadPoolTaskExecutor::getCorePoolSize);             Metrics.gauge("thread.pool.largest.size", consumerExecutorTAG, consumerExecutor, e -> e.getThreadPoolExecutor().getLargestPoolSize());             Metrics.gauge("thread.pool.max.size", consumerExecutorTAG, consumerExecutor, ThreadPoolTaskExecutor::getMaxPoolSize);             Metrics.gauge("thread.pool.active.size", consumerExecutorTAG, consumerExecutor, ThreadPoolTaskExecutor::getActiveCount);             Metrics.gauge("thread.pool.thread.count", consumerExecutorTAG, consumerExecutor, ThreadPoolTaskExecutor::getPoolSize);             Metrics.gauge("thread.pool.queue.size", consumerExecutorTAG, consumerExecutor, e -> e.getThreadPoolExecutor().getQueue().size());                 Metrics.gauge("thread.pool.connectionCount", okHttpExecutorTAG, connectionPool, ConnectionPool::connectionCount);             Metrics.gauge("thread.pool.idleConnectionCount", okHttpExecutorTAG, connectionPool, ConnectionPool::idleConnectionCount);         } catch (Exception e) {             //ignore         }     };       @Override     public void afterPropertiesSet() throws Exception {         // 每5秒执行一次         scheduledExecutor.scheduleWithFixedDelay(monitor, 0, 5, TimeUnit.SECONDS);     }   }