緩存也許是程序員心中最熟悉的性能優(yōu)化手段之一庸诱，在舊文中微服務(wù)緩存漫談之Guava Cache 和 Redis 集群的構(gòu)建和監(jiān)控中分別介紹了最常用的本地內(nèi)存的 Guava Cache 和遠(yuǎn)程的 Redis Cache. 這里我們重點(diǎn)聊聊緩存的度量扼鞋。

緩存的常見問題

對于緩存珍昨，我們關(guān)心這幾個(gè)問題：

Cache hit ratio 緩存的命中率
Cache key size 緩存的鍵值數(shù)量
Cache resource usage 緩存的資源使用率
Cache loading performance 緩存的加載性能
Cache capacity 緩存的容量
Cache lifetime 緩存的生命周期

Cache 不可能無限增長, 不可能永遠(yuǎn)有效, 所以對于 Cache 的清除策略和失效策略要細(xì)細(xì)考量.
對于放在 Cache 中的數(shù)據(jù)也最好是讀寫比較高的, 即讀得多, 寫得少, 不會(huì)頻繁地更新.

緩存不是萬能藥，緩存使用不當(dāng)會(huì)生成緩存穿透，擊穿和雪崩瘦黑，先簡單解釋一下這幾個(gè)概念

穿透
某條記錄壓根不存在感混，所以在緩存中找不到竹伸，每次都需要到數(shù)據(jù)庫中讀取泥栖，但是結(jié)果還是找不到。

常用的應(yīng)對方法是布隆過濾器（它的特點(diǎn)是）或者反向緩存（在緩存中保存這條記錄勋篓，標(biāo)識它是不存在的）

擊穿
某條記錄過期被移除了吧享，恰好大量相關(guān)的查詢請求這條記錄，導(dǎo)致瞬時(shí)間大量請求繞過緩存訪問數(shù)據(jù)庫

常用的應(yīng)對方法是將從數(shù)據(jù)庫加載數(shù)據(jù)的操作加鎖譬嚣，這樣就不會(huì)有很多訪問請求繞過緩存钢颂。
或者干脆不設(shè)置過期時(shí)間，而是用一個(gè)后臺job 定時(shí)刷新緩存拜银，外部的請求總能從緩存中讀到數(shù)據(jù)

3.雪崩
多條記錄在多個(gè)服務(wù)器上的緩存同時(shí)過期失效甸陌，導(dǎo)致瞬時(shí)間大量請求繞過緩存訪問數(shù)據(jù)庫，這個(gè)比擊穿更嚴(yán)重盐股。

常用的應(yīng)對方法是多個(gè)服務(wù)器上的多條記錄設(shè)置不同的失效時(shí)間，可以用個(gè)隨機(jī)值作為零頭耻卡，將大量的并發(fā)請求從某個(gè)時(shí)間點(diǎn)分布到一個(gè)時(shí)間段中

緩存的度量

緩存的命中率疯汁，加載性能等等都是我們關(guān)心的重點(diǎn)，例如：

性能 Performance: Cache 加載的延遲 latency
吞吐量Throughput: 每秒請求次數(shù) CPS（Call Per Second）
命中率：sucess_ratio = hitCount / (hitCount + missCount)
資源使用量: 使用了多少內(nèi)存
資源飽和度 saturation: 由于容量限制被移出cache 的記錄數(shù)卵酪，緩存滿了無法增加的記錄數(shù)

注：飽和度是資源負(fù)載超出其處理能力的地方幌蚊。

以 Guava Cache 為例，它的緩存統(tǒng)計(jì)信息根據(jù)以下規(guī)則遞增：

當(dāng)緩存查找遇到現(xiàn)有緩存條目時(shí)溃卡，hitCount會(huì)增加溢豆。
當(dāng)緩存查找第一次遇到丟失的緩存條目時(shí)，將加載一個(gè)新條目瘸羡。
成功加載條目后漩仙，missCount和loadSuccessCount會(huì)增加，并將總加載時(shí)間（以納秒為單位）添加到totalLoadTime中。
在加載條目時(shí)引發(fā)異常時(shí)队他，missCount和loadExceptionCount會(huì)增加卷仑，并且總加載時(shí)間（以納秒為單位）將添加到totalLoadTime中。
遇到仍在加載的缺少高速緩存條目的高速緩存查找將等待加載完成（無論是否成功）麸折，然后遞增missCount锡凝。
從緩存中逐出條目時(shí)，evictionCount會(huì)增加垢啼。
當(dāng)緩存條目無效或手動(dòng)刪除時(shí)窜锯，不會(huì)修改任何統(tǒng)計(jì)信息。
在緩存的asMap視圖上調(diào)用的操作不會(huì)修改任何統(tǒng)計(jì)信息芭析。

我們在寫代碼時(shí)可以調(diào)用它的 recordStats 來記錄這些度量數(shù)據(jù)

 @Bean
    public LoadingCache<String, CityWeather> cityWeatherCache() {
        LoadingCache<String, CityWeather> cache = CacheBuilder.newBuilder()
                .recordStats()
                .maximumSize(1000)
                .expireAfterWrite(60, TimeUnit.MINUTES)
                .build(weatherCacheLoader());

        recordCacheMetrics("cityWeatherCache", cache);
        return cache;
    }

    public void recordCacheMetrics(String cacheName, Cache cache) {
        MetricRegistry metricRegistry = metricRegistry();
        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "hitCount"), () -> () -> cache.stats().hitCount());
        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "hitRate"), () -> () -> cache.stats().hitRate());

        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "missCount"), () -> () -> cache.stats().missCount());
        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "missRate"), () -> () -> cache.stats().missRate());

        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "requestCount"), () -> () -> cache.stats().requestCount());

        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "loadCount"), () -> () -> cache.stats().loadCount());
        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "loadSuccessCount"), () -> () -> cache.stats().loadSuccessCount());
        metricRegistry.gauge(generateMetricsKeyForCache(cacheName, "loadExceptionCount"), () -> () -> cache.stats().loadExceptionCount());
    }

 public String generateMetricsKeyForCache(String cacheName, String keyName) {
        String metricKey = MetricRegistry.name("cache", cacheName, keyName);
        log.info("metric key generated for cache: {}", metricKey);
        return metricKey;
    }

對基于 Spring boot 的應(yīng)用程序锚扎，我們可以用 Micrometer 這個(gè)軟件庫來暴露度量指標(biāo)。
監(jiān)控軟件百花齊放放刨，類似于 SLF4j 作為一個(gè)外觀模式的應(yīng)用把 log4j, logback 這些庫的異同封裝起來工秩， MicroMeter 也把對各種監(jiān)控和度量技術(shù)棧的細(xì)節(jié)封裝起來，這樣應(yīng)用程序的開發(fā)者可以把精力放到應(yīng)用本身的度量上面进统。

Micrometer provides a simple facade over the instrumentation clients for the most popular monitoring systems, allowing you to instrument your JVM-based application code without vendor lock-in. Think SLF4J, but for application metrics! Application metrics recorded by Micrometer are intended to be used to observe, alert, and react to the current/recent operational state of your environment.

它的應(yīng)用也很簡單助币，在 pom.xml 中加入下面的依賴項(xiàng)

<dependency>
    <groupId>org.springframework.boot</groupId>
    <artifactId>spring-boot-starter-actuator</artifactId>
</dependency>

<dependency>
    <groupId>io.micrometer</groupId>
    <artifactId>micrometer-core</artifactId>
</dependency>

<dependency>
    <groupId>io.micrometer</groupId>
    <artifactId>micrometer-registry-prometheus</artifactId>
</dependency>

WeatherCacheConfig 配置如下

package com.github.walterfan.hellocache;


import com.codahale.metrics.MetricRegistry;
import com.google.common.cache.Cache;
import com.google.common.cache.CacheBuilder;
import com.google.common.cache.LoadingCache;

import io.micrometer.core.instrument.Gauge;
import io.micrometer.core.instrument.MeterRegistry;
import io.micrometer.core.instrument.composite.CompositeMeterRegistry;
import io.micrometer.core.instrument.simple.SimpleMeterRegistry;
import io.micrometer.prometheus.PrometheusConfig;
import io.micrometer.prometheus.PrometheusMeterRegistry;
import lombok.extern.slf4j.Slf4j;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.ComponentScan;
import org.springframework.context.annotation.Configuration;
import org.springframework.context.annotation.EnableAspectJAutoProxy;
import org.springframework.context.annotation.Lazy;
import org.springframework.core.env.Environment;

import org.springframework.http.MediaType;
import org.springframework.http.converter.HttpMessageConverter;
import org.springframework.http.converter.json.MappingJackson2HttpMessageConverter;
import org.springframework.web.client.RestTemplate;

import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.concurrent.TimeUnit;

/**
 * Created by yafan on 14/10/2017.
 */
@EnableAspectJAutoProxy
@ComponentScan
@Configuration
@Slf4j
public class WeatherCacheConfig {//implements EnvironmentAware

    @Autowired
    private Environment environment;

    @Bean
    public WeatherCacheLoader weatherCacheLoader() {
        return new WeatherCacheLoader();
    }

    @Bean
    public RestTemplate restTemplate() {

        final RestTemplate restTemplate = new RestTemplate();

        List<HttpMessageConverter<?>> messageConverters = new ArrayList<>();
        MappingJackson2HttpMessageConverter converter = new MappingJackson2HttpMessageConverter();
        converter.setSupportedMediaTypes(Collections.singletonList(MediaType.ALL));
        messageConverters.add(converter);
        restTemplate.setMessageConverters(messageConverters);

        return restTemplate;
    }

    @Bean
    public String appToken() {
        return this.environment.getProperty("BAIDU_AK");
    }


    @Bean
    public LoadingCache<String, CityWeather> cityWeatherCache() {
        LoadingCache<String, CityWeather> cache = CacheBuilder.newBuilder()
                .recordStats()
                .maximumSize(1000)
                .expireAfterWrite(60, TimeUnit.MINUTES)
                .build(weatherCacheLoader());

        recordCacheMetrics("cityWeatherCache", cache);

        recordCacheMeters("cityWeatherCache", cache);

        return cache;
    }

    public void recordCacheMetrics(String cacheName, Cache cache) {

        MetricRegistry metricRegistry = metricRegistry();

        metricRegistry.gauge(makeMetricsKeyName(cacheName, "hitCount"), () -> () -> cache.stats().hitCount());
        metricRegistry.gauge(makeMetricsKeyName(cacheName, "hitRate"), () -> () -> cache.stats().hitRate());

        metricRegistry.gauge(makeMetricsKeyName(cacheName, "missCount"), () -> () -> cache.stats().missCount());
        metricRegistry.gauge(makeMetricsKeyName(cacheName, "missRate"), () -> () -> cache.stats().missRate());

        metricRegistry.gauge(makeMetricsKeyName(cacheName, "requestCount"), () -> () -> cache.stats().requestCount());

        metricRegistry.gauge(makeMetricsKeyName(cacheName, "loadCount"), () -> () -> cache.stats().loadCount());
        metricRegistry.gauge(makeMetricsKeyName(cacheName, "loadSuccessCount"), () -> () -> cache.stats().loadSuccessCount());
        metricRegistry.gauge(makeMetricsKeyName(cacheName, "loadExceptionCount"), () -> () -> cache.stats().loadExceptionCount());
    }

    public void recordCacheMeters(String cacheName, Cache cache) {
        MeterRegistry meterRegistry = meterRegistry();
        Gauge.builder(makeMetricsKeyName(cacheName, "hitCount"), () -> cache.stats().hitCount()).register(meterRegistry);
        Gauge.builder(makeMetricsKeyName(cacheName, "hitRate"), () -> cache.stats().hitRate()).register(meterRegistry);

        Gauge.builder(makeMetricsKeyName(cacheName, "missCount"), () -> cache.stats().missCount()).register(meterRegistry);
        Gauge.builder(makeMetricsKeyName(cacheName, "missRate"), () -> cache.stats().missRate()).register(meterRegistry);

        Gauge.builder(makeMetricsKeyName(cacheName, "requestCount"), () -> cache.stats().requestCount()).register(meterRegistry);

        Gauge.builder(makeMetricsKeyName(cacheName, "loadCount"), () -> cache.stats().loadCount()).register(meterRegistry);
        Gauge.builder(makeMetricsKeyName(cacheName, "loadSuccessCount"), () -> cache.stats().loadSuccessCount()).register(meterRegistry);
        Gauge.builder(makeMetricsKeyName(cacheName, "loadExceptionCount"), () -> cache.stats().loadExceptionCount()).register(meterRegistry);

    }

    public String makeMetricsKeyName(String cacheName, String keyName) {
        String metricKey = MetricRegistry.name("cache", cacheName, keyName);
        log.info("metric key generated for cache: {}", metricKey);
        return metricKey;
    }

    @Bean
    public DurationTimerAspect durationTimerAspect() {
        return new DurationTimerAspect();
    }

    @Bean
    @Lazy
    public MetricRegistry metricRegistry() {
        return new MetricRegistry();
    }

    @Bean
    public MeterRegistry meterRegistry() {
        CompositeMeterRegistry compositeRegistry = new CompositeMeterRegistry();

        SimpleMeterRegistry simpleMeter = new SimpleMeterRegistry();
        PrometheusMeterRegistry prometheusMeterRegistry = new PrometheusMeterRegistry(PrometheusConfig.DEFAULT);

        compositeRegistry.add(simpleMeter);
        compositeRegistry.add(prometheusMeterRegistry);
        return compositeRegistry;
    }
}

這樣打開 http://localhost:8080/actuator/metrics 可以看到如下度量指標(biāo)

{
names: [
"cache.cityWeatherCache.hitCount",
"cache.cityWeatherCache.hitRate",
"cache.cityWeatherCache.loadCount",
"cache.cityWeatherCache.loadExceptionCount",
"cache.cityWeatherCache.loadSuccessCount",
"cache.cityWeatherCache.missCount",
"cache.cityWeatherCache.missRate",
"cache.cityWeatherCache.requestCount",
"http.server.requests",
"jvm.buffer.count",
"jvm.buffer.memory.used",
"jvm.buffer.total.capacity",
"jvm.classes.loaded",
"jvm.classes.unloaded",
"jvm.gc.live.data.size",
"jvm.gc.max.data.size",
"jvm.gc.memory.allocated",
"jvm.gc.memory.promoted",
"jvm.gc.pause",
"jvm.memory.committed",
"jvm.memory.max",
"jvm.memory.used",
"jvm.threads.daemon",
"jvm.threads.live",
"jvm.threads.peak",
"jvm.threads.states",
"logback.events",
"process.cpu.usage",
"process.files.max",
"process.files.open",
"process.start.time",
"process.uptime",
"system.cpu.count",
"system.cpu.usage",
"system.load.average.1m",
"tomcat.sessions.active.current",
"tomcat.sessions.active.max",
"tomcat.sessions.alive.max",
"tomcat.sessions.created",
"tomcat.sessions.expired",
"tomcat.sessions.rejected"
]
}

詳情可打開 http://localhost:8080/actuator/metrics/cache.cityWeatherCache.hitCount

{
  name: "cache.cityWeatherCache.hitCount",
  description: null,
  baseUnit: null,
  measurements: [
  {
    statistic: "VALUE",
    value: 3
  }
  ],
  availableTags: [ ]
}

微服務(wù)緩存的使用度量

微服務(wù)緩存的使用度量

緩存的常見問題

緩存的度量