前言
我们在上篇文章中讲解了Exporter的实现,Exporter最终是需要定时器来调度的,那么本文就来分析一下其实现.同时,定时器是需要配置的,因此抽象出TriggerProperties.其类图如下:
解析
TriggerProperties
TriggerProperties–> 定时器配置的抽象基类.其声明了如下字段:
// 2次export之间的间隔时间
private Long delayMillis;
// 是否该metric 可以导出
private boolean enabled = true;
// 是否关闭任何可优化的但是未导出的指标值
private Boolean sendLatest;
/** * List of patterns for metric names to include. */
private String[] includes;
/** * List of patterns for metric names to exclude. Applied after the includes. */
private String[] excludes;SpecificTriggerProperties
SpecificTriggerProperties–> 继承自TriggerProperties.其声明了如下字段:
// 指定满足该名字(或者正则)的bean 的id
private String[] names;MetricExportProperties
MetricExportProperties 继承自TriggerProperties.
声明了如下字段:
// key--> MetricWriter bean的id,value --> SpecificTriggerProperties private Map<String, SpecificTriggerProperties> triggers = new LinkedHashMap<String, SpecificTriggerProperties>(); private Aggregate aggregate = new Aggregate(); private Redis redis = new Redis(); private Statsd statsd = new Statsd();其中:
Aggregate–> Aggregate的配置类,声明了如下字段:
// repository的全局前缀.应该在jvm中是全局唯一的,但是大多数这个是很有用的,如果其形如"a.b",a-->逻辑 b-->物理进程. private String prefix = ""; // 指明该aggregator 该对repository 中的什么类型的key 进行操作 private String keyPattern = "";Redis–> redis的配置类,声明了如下字段:
// redis repository 的前缀,应该全局唯一 private String prefix = "spring.metrics"; // redis repository export 的key,全局唯一 private String key = "keys.spring.metrics";Statsd –> Statsd的配置类,声明了如下字段:
// statsd的主机名 private String host; // statsd的端口号 private int port = 8125; // 向statsd暴露metrics时添加的前缀s private String prefix;
由于该类中的setUpDefaults被@PostConstruct注解,因此该方法会在MetricExportProperties初始化完毕之后调用.代码如下:
@PostConstruct public void setUpDefaults() { TriggerProperties defaults = this; // 1. 遍历triggers, for (Entry<String, SpecificTriggerProperties> entry : this.triggers.entrySet()) { String key = entry.getKey(); SpecificTriggerProperties value = entry.getValue(); // 1.1 如果SpecificTriggerProperties 没有指定名字,则将其赋值为MetricWriter bean的id if (value.getNames() == null || value.getNames().length == 0) { value.setNames(new String[] { key }); } } if (defaults.isSendLatest() == null) { defaults.setSendLatest(true); } if (defaults.getDelayMillis() == null) { defaults.setDelayMillis(5000); } for (TriggerProperties value : this.triggers.values()) { if (value.isSendLatest() == null) { value.setSendLatest(defaults.isSendLatest()); } if (value.getDelayMillis() == null) { value.setDelayMillis(defaults.getDelayMillis()); } } }由于该类声明了@ConfigurationProperties(prefix = “spring.metrics.export”)注解,因此可以通过如下进行配置:
spring.metrics.export.aggregate.key-pattern= # Pattern that tells the aggregator what to do with the keys from the source repository. spring.metrics.export.aggregate.prefix= # Prefix for global repository if active. spring.metrics.export.delay-millis=5000 # Delay in milliseconds between export ticks. Metrics are exported to external sources on a schedule with this delay. spring.metrics.export.enabled=true # Flag to enable metric export (assuming a MetricWriter is available). spring.metrics.export.excludes= # List of patterns for metric names to exclude. Applied after the includes. spring.metrics.export.includes= # List of patterns for metric names to include. spring.metrics.export.redis.key=keys.spring.metrics # Key for redis repository export (if active). spring.metrics.export.redis.prefix=spring.metrics # Prefix for redis repository if active. spring.metrics.export.send-latest= # Flag to switch off any available optimizations based on not exporting unchanged metric values. spring.metrics.export.statsd.host= # Host of a statsd server to receive exported metrics. spring.metrics.export.statsd.port=8125 # Port of a statsd server to receive exported metrics. spring.metrics.export.statsd.prefix= # Prefix for statsd exported metrics. spring.metrics.export.triggers.*= # Specific trigger properties per MetricWriter bean name.
MetricExporters
MetricExporters–> 实现了SchedulingConfigurer, Closeable接口.
字段,构造器如下:
private MetricReader reader; private Map<String, GaugeWriter> writers = new HashMap<String, GaugeWriter>(); // 配置文件 private final MetricExportProperties properties; // key--> Exporter bean 的id, value --> Exporter private final Map<String, Exporter> exporters = new HashMap<String, Exporter>(); // MetricCopyExporter 所对应的 Exporter bean 的id(closeables中持有的是writers的key) private final Set<String> closeables = new HashSet<String>(); public MetricExporters(MetricExportProperties properties) { this.properties = properties; }方法实现如下:
configureTasks–> SchedulingConfigurer 中声明的接口方法.代码如下:
public void configureTasks(ScheduledTaskRegistrar taskRegistrar) { // 1. 遍历exporters for (Entry<String, Exporter> entry : this.exporters.entrySet()) { String name = entry.getKey(); Exporter exporter = entry.getValue(); // 1.1 获得属于该Exporter的TriggerProperties TriggerProperties trigger = this.properties.findTrigger(name); if (trigger != null) { // 1.2 如果存在TriggerProperties,则实例化ExportRunner ExportRunner runner = new ExportRunner(exporter); // 实例化IntervalTask,其内部会执行ExportRunner的run方法.在实例化后延迟DelayMillis后执行,之后每隔DelayMillis后执行 IntervalTask task = new IntervalTask(runner, trigger.getDelayMillis(), trigger.getDelayMillis()); // 1.3 添加到ScheduledTaskRegistrar中 taskRegistrar.addFixedDelayTask(task); } } // 2. 遍历writers for (Entry<String, GaugeWriter> entry : this.writers.entrySet()) { String name = entry.getKey(); GaugeWriter writer = entry.getValue(); // 2.1 获得属于该Exporter的TriggerProperties TriggerProperties trigger = this.properties.findTrigger(name); if (trigger != null) { // 2.2 根据GaugeWriter 创建MetricCopyExporter MetricCopyExporter exporter = getExporter(writer, trigger); // 2.3 添加到exporters，closeables 中 this.exporters.put(name, exporter); this.closeables.add(name); // 2.4 实例化ExportRunner,IntervalTask,添加到ScheduledTaskRegistrar中 ExportRunner runner = new ExportRunner(exporter); IntervalTask task = new IntervalTask(runner, trigger.getDelayMillis(), trigger.getDelayMillis()); taskRegistrar.addFixedDelayTask(task); } } }遍历exporters
- 获得属于该Exporter的TriggerProperties
- 如果存在TriggerProperties,则实例化ExportRunner
- 实例化IntervalTask,其内部会执行ExportRunner的run方法.在实例化后延迟DelayMillis后执行,之后每隔DelayMillis后执行
- 添加到ScheduledTaskRegistrar中
遍历writers
- 获得属于该Exporter的TriggerProperties
根据GaugeWriter 创建MetricCopyExporter.代码如下:
private MetricCopyExporter getExporter(GaugeWriter writer, TriggerProperties trigger) { MetricCopyExporter exporter = new MetricCopyExporter(this.reader, writer); exporter.setIncludes(trigger.getIncludes()); exporter.setExcludes(trigger.getExcludes()); exporter.setSendLatest(trigger.isSendLatest()); return exporter; }- 添加到exporters，closeables 中
- 实例化ExportRunner,IntervalTask,添加到ScheduledTaskRegistrar中
ExportRunner实现了Runnable接口,代码如下:
private static class ExportRunner implements Runnable { private final Exporter exporter; ExportRunner(Exporter exporter) { this.exporter = exporter; } @Override public void run() { this.exporter.export(); } }close,代码如下:
public void close() throws IOException { for (String name : this.closeables) { Exporter exporter = this.exporters.get(name); if (exporter instanceof Closeable) { ((Closeable) exporter).close(); } } }- 遍历其持有的closeables(closeables中持有的是writers的key)
- 依次获得对应的Exporter,如果其实现了Closeable接口,则调用其close方法(此时实际获得的是MetricCopyExporter,其实现了Closeable)
MetricExporters 自动装配
MetricExporters的自动配置类为MetricExportAutoConfiguration.
该类声明了如下注解:
@Configuration @EnableScheduling @ConditionalOnProperty(value = "spring.metrics.export.enabled", matchIfMissing = true) @EnableConfigurationProperties- @Configuration –> 配置类
- @EnableScheduling –> 引入了SchedulingConfiguration,在该类中注册了ScheduledAnnotationBeanPostProcessor,开启了对@Scheduled注解的处理
- @ConditionalOnProperty(value = “spring.metrics.export.enabled”, matchIfMissing = true) –> 当配置有spring.metrics.export.enabled=true时生效,如果没有配置的话,则默认生效
- @EnableConfigurationProperties –> 引入了EnableConfigurationPropertiesImportSelector,最终执行其selectImports,由于此时我们没有配置value值,因此会返回ConfigurationPropertiesBindingPostProcessorRegistrar.该类会注册ConfigurationBeanFactoryMetaData(关于这个,我们之前已经讲过很多次了)
MetricExportAutoConfiguration中有2个配置类:
MetricExportPropertiesConfiguration,代码如下:
@Configuration protected static class MetricExportPropertiesConfiguration { // 默认为application.随机值 如果配置了spring.application.name 则为spring.application.name.随机值 @Value("${spring.application.name:application}.${random.value:0000}") private String prefix = ""; private String aggregateKeyPattern = "k.d"; @Bean(name = "spring.metrics.export-org.springframework.boot.actuate.metrics.export.MetricExportProperties") @ConditionalOnMissingBean public MetricExportProperties metricExportProperties() { // 1. 实例化MetricExportProperties MetricExportProperties export = new MetricExportProperties(); // 2. 设置Redis的前缀 export.getRedis().setPrefix("spring.metrics" + (this.prefix.length() > 0 ? "." : "") + this.prefix); // 3. 设置Aggregate的前缀 export.getAggregate().setPrefix(this.prefix); // 4. 设置Aggregate的KeyPattern为k.d export.getAggregate().setKeyPattern(this.aggregateKeyPattern); return export; } }当BeanFactory中不存在MetricExportProperties类型的bean时,会执行metricExportProperties方法,进行注册–>id为spring.metrics.export-org.springframework.boot.actuate.metrics.export.MetricExportProperties
StatsdConfiguration,代码如下:
@Configuration static class StatsdConfiguration { @Bean @ExportMetricWriter @ConditionalOnMissingBean @ConditionalOnProperty(prefix = "spring.metrics.export.statsd", name = "host") public StatsdMetricWriter statsdMetricWriter(MetricExportProperties properties) { MetricExportProperties.Statsd statsdProperties = properties.getStatsd(); return new StatsdMetricWriter(statsdProperties.getPrefix(), statsdProperties.getHost(), statsdProperties.getPort()); } }当满足如下条件进行注册:
- @ConditionalOnMissingBean –> BeanFactory中不存在StatsdMetricWriter类型的bean时生效
- @ConditionalOnProperty(prefix = “spring.metrics.export.statsd”, name = “host”) –> 配置有spring.metrics.export.statsd.host 时生效
个人觉得,此处的代码有问题,当我们配置有spring.metrics.export.statsd.host,确没有在项目中加入java-statsd-client的依赖,此处就会抛出java.lang.ClassNotFoundException: com.timgroup.statsd.StatsDClientErrorHandler
解决方案:在该方法上加上@ConditionalOnClass(name=”com.timgroup.statsd.NonBlockingStatsDClient”)
该bug在spring-boot 2.0.0.M5中fix.在2.0.0.M4 中不支持StatsdMetricWriter.
该类的字段,构造器如下:
// 默认情况,没有自动装配 private final MetricsEndpointMetricReader endpointReader; // 此处获得的是被@ExportMetricReader注解的MetricReader private final List<MetricReader> readers; // 此处获得的是被@ExportMetricWriter注解的GaugeWriter,key--> GaugeWriter的id,value --> GaugeWriter private final Map<String, GaugeWriter> writers; // key --> Exporter bean 的id,value --> Exporter private final Map<String, Exporter> exporters; public MetricExportAutoConfiguration(MetricExportProperties properties, ObjectProvider<MetricsEndpointMetricReader> endpointReader, @ExportMetricReader ObjectProvider<List<MetricReader>> readers, @ExportMetricWriter ObjectProvider<Map<String, GaugeWriter>> writers, ObjectProvider<Map<String, Exporter>> exporters) { this.endpointReader = endpointReader.getIfAvailable(); this.readers = readers.getIfAvailable(); this.writers = writers.getIfAvailable(); this.exporters = exporters.getIfAvailable(); }其字段的说明如下:
- MetricsEndpointMetricReader endpointReader –> 默认情况,没有自动装配
List readers –> 此处获得的是被@ExportMetricReader注解的MetricReader
- 在jdk1.8及以上的环境中并且没有加入metrics-core依赖时,获得的是BufferMetricReader
- 在jdk1.8以前的环境中并且BaanFactory中不存在id为actuatorMetricRepository的bean时,LegacyMetricRepositoryConfiguration会生效.因此此时获得的是InMemoryMetricRepository,InMemoryMultiMetricRepository
Map writers–> 此处获得的是被@ExportMetricWriter注解的GaugeWriter,key–> GaugeWriter的id,value –> GaugeWriter:
在加入spring-boot-starter-integration,并且注册了id为metricsChannel,类型为metricsChannel的bean之后,MetricsChannelAutoConfiguration配置类就会生效,因此会获得MessageChannelMetricWriter
exporters –> key –> Exporter bean 的id,value –> Exporter.默认情况下，此时获得的是空集合
在该类中,声明了1个@Bean方法,如下:
@Bean @ConditionalOnMissingBean(name = "metricWritersMetricExporter") public SchedulingConfigurer metricWritersMetricExporter( MetricExportProperties properties) { Map<String, GaugeWriter> writers = new HashMap<String, GaugeWriter>(); // 1. 如果endpointReader 没有配置并且readers 不为空,则实例化CompositeMetricReader MetricReader reader = this.endpointReader; if (reader == null && !CollectionUtils.isEmpty(this.readers)) { reader = new CompositeMetricReader( this.readers.toArray(new MetricReader[this.readers.size()])); } // 2. 如果reader 等于null并且exporters 为空,则返回NoOpSchedulingConfigurer if (reader == null && CollectionUtils.isEmpty(this.exporters)) { return new NoOpSchedulingConfigurer(); } // 3. 实例化MetricExporters MetricExporters exporters = new MetricExporters(properties); // 3.1 如果reader存在的话 if (reader != null) { // 如果GaugeWriter不回空,则加入到writers中 if (!CollectionUtils.isEmpty(this.writers)) { writers.putAll(this.writers); } // 3.2 赋值到MetricExporters中 exporters.setReader(reader); exporters.setWriters(writers); } // 3.3 设置exporters exporters.setExporters(this.exporters == null ? Collections.<String, Exporter>emptyMap() : this.exporters); return exporters; }该方法的逻辑如下:
如果endpointReader 没有配置并且readers 不为空,则实例化CompositeMetricReader. 默认情况下会执行此步.
如果reader 等于null并且exporters 为空,则返回NoOpSchedulingConfigurer,此时会在加入metrics-core依赖时生效.默认不执行此步骤
实例化MetricExporters
- 如果reader存在的话
- 如果GaugeWriter不回空,则加入到writers中,此步骤默认不执行,需要加入spring-boot-starter-integration依赖
- 赋值到MetricExporters中
- 设置exporters. 默认exporters等于null,则此时传入的是空集合
因此:
- 默认情况下,MetricExporters中不会配置定时任务
加入spring-boot-starter-integration,并且注册了id为metricsChannel之后,会注册1个定时任务,此时会将BufferMetricReader中的信息写到名为metricsChannel的Channel中.该步骤要想成功,需要如下几步:
在pom文件加入如下依赖:
</dependency> <dependency> <groupId>org.springframework.boot</groupId> <artifactId>spring-boot-starter-integration</artifactId> </dependency> <dependency> <groupId>org.springframework.integration</groupId> <artifactId>spring-integration-stream</artifactId> <version>4.3.12.RELEASE</version> </dependency>在配置类加入如下配置:
@Bean @ExportMetricWriter public MessageChannel metricsChannel() { return new DirectChannel(); } @Bean public IntegrationFlow test() { return IntegrationFlows.from("metricsChannel") .handle(CharacterStreamWritingMessageHandler.stdout()) .get(); }在controller中加入如下代码:
@Autowired private CounterService counterService; @RequestMapping("/test-counter") public String testCounter() { counterService.increment("test-counter.count"); return "操作成功"; }此时访问http://127.0.0.1:8080/test-counter后,会发现控制台打印如下日志:
Metric [name=gauge.response.test-counter, value=32.0, timestamp=Wed Jan 31 17:36:36 CST 2018]Metric [name=gauge.response.star-star.favicon.ico, value=14.0, timestamp=Wed Jan 31 17:36:36 CST 2018] Metric [name=counter.status.200.star-star.favicon.ico, value=1, timestamp=Wed Jan 31 17:36:36 CST 2018]Metric [name=counter.test-counter.count, value=1, timestamp=Wed Jan 31 17:36:36 CST 2018] Metric [name=counter.status.200.test-counter, value=1, timestamp=Wed Jan 31 17:36:36 CST 2018]
