zoukankan      html  css  js  c++  java
  • Springboot Actuator之七:actuator 中原生endpoint源码解析1

    看actuator项目的包结构,如下:

    本文中的介绍Endpoints。

    Endpoints(端点)介绍

    Endpoints 是 Actuator 的核心部分,它用来监视应用程序及交互,spring-boot-actuator中已经内置了非常多的Endpoints(health、info、beans、httptrace、shutdown等等),同时也允许我们扩展自己的端点。

    Endpoints 分成两类:原生端点用户自定义端点:

    1. 原生端点是在应用程序里提供的众多 restful api 接口,通过它们可以监控应用程序运行时的内部状况。原生端点又可以分成三类:
      • 应用配置类:可以查看应用在运行期间的静态信息:例如自动配置信息、加载的spring bean信息、yml文件配置信息、环境信息、请求映射信息;
      • 度量指标类:主要是运行期间的动态信息,例如堆栈、请求连、一些健康指标、metrics信息等;
      • 操作控制类:主要是指shutdown,用户可以发送一个请求将应用的监控功能关闭。
    2. 自定义端点主要是指扩展性,用户可以根据自己的实际应用,定义一些比较关心的指标,在运行期进行监控。

    我们这里详细说明org.springframework.boot.actuate.endpoint中原生端点的实现.通过如下几个维度来进行分析:

    • xxxEndpoint的作用
    • xxxEndpoint的字段,构造器
    • xxxEndpoint核心方法invoke 实现分析
    • xxxEndpoint如何进行配置
    • xxxEndpoint如何自动化装配

    在org.springframework.boot.actuate.endpoint 中还有2个子包-jmx(可通过jmx协议访问),mvc(通过spring mvc 暴露,可通过接口进行访问,在下篇文章进行分析).这里我们不关注这些,这看org.springframework.boot.actuate.endpoint 包下的类,类比较多,先看个类图吧,如下:

    Endpoint接口:org.springframework.boot.actuate.endpoint.Endpoint.java

    Endpoint接口:一个端点可以用于暴露(系统信息、操作入口等)信息。通常暴露方式是通过spring mvc的,如继承AbstractEndpoint的方式实现自己的endpoint。

    public interface Endpoint<T> {
    
        // 端点的逻辑标识(字母、数字和下划线('_') 组成)
        String getId();
    
        // 端点是否启用
        boolean isEnabled();
    
        // 端点是否输出敏感数据
        boolean isSensitive();
    
        // 调用端点,并返回调用结果
        T invoke();
    }

     其中泛型参数T为暴露的数据类型.方法的作用已经注释。

    AbstractEndpoint抽象类

    Endpoint的一个抽象子类:AbstractEndpoint(Endpoint接口实现的抽象基类),该类实现了EnvironmentAware,因此, AbstractEndpoint也就持有了Environment。

    1、AbstractEndpoint 有如下属性:
    // 匹配包括下划线的任何单词字符。类似但不等价于“[A-Za-z0-9_]”
    private static final Pattern ID_PATTERN = Pattern.compile("\w+");
    // 通过EnvironmentAware接口注入
    private Environment environment;
    // 端点标识符
    private String id;
    // 是否默认敏感
    private final boolean sensitiveDefault;
    // 标识该端点是否暴露敏感信息
    private Boolean sensitive;
    // 是否端点可用
    private Boolean enabled;
    2、AbstractEndpoint方法

    AbstractEndpoint方法实现了Endpoint接口中的getId, isEnabled, isSensitive,其中, getId只需返回AbstractEndpoint中的id属性即可,我们分别来看下其他方法的实现:

    2.1、isEnabled,代码如下:

    @Override
    public boolean isEnabled() {
            return EndpointProperties.isEnabled(this.environment, this.enabled);
    }
    @ConfigurationProperties(prefix = "endpoints")
    public class EndpointProperties {
    
        private static final String ENDPOINTS_ENABLED_PROPERTY = "endpoints.enabled";
    
        private static final String ENDPOINTS_SENSITIVE_PROPERTY = "endpoints.sensitive";
    
        public static boolean isEnabled(Environment environment, Boolean enabled) {
            //1、如果AbstractEndpoint#enabled属性有值,则使用AbstractEndpoint的配置
            if (enabled != null) {
                return enabled;
            }
            //如果Environment 不等于null 并且Environment 配置有endpoints.enabled的属性,则返回其配置的值
            if (environment != null
                    && environment.containsProperty(ENDPOINTS_ENABLED_PROPERTY)) {
                return environment.getProperty(ENDPOINTS_ENABLED_PROPERTY, Boolean.class);
            }
            //3、如果1和2没有值,则返回默认值true
            return true;
        }

    2.2、isSensitive和isEnabled实现差不多,如下:

        @Override
        public boolean isSensitive() {
            return EndpointProperties.isSensitive(this.environment, this.sensitive,
                    this.sensitiveDefault);
        }
    @ConfigurationProperties(prefix = "endpoints")
    public class EndpointProperties {
    
        private static final String ENDPOINTS_ENABLED_PROPERTY = "endpoints.enabled";
    
        private static final String ENDPOINTS_SENSITIVE_PROPERTY = "endpoints.sensitive";
    
        public static boolean isSensitive(Environment environment, Boolean sensitive,
                boolean sensitiveDefault) {
            //1、如果abstractEndpoint的sensitive有值,则使用这个配置
            if (sensitive != null) {
                return sensitive;
            }
            //2、如果environment 不等于null 并且 environment中配置有endpoints.sensitive的属性,则返回其配置值
            if (environment != null
                    && environment.containsProperty(ENDPOINTS_SENSITIVE_PROPERTY)) {
                return environment.getProperty(ENDPOINTS_SENSITIVE_PROPERTY, Boolean.class);
            }
            //3、返回指定的默认值(默认为false)
            return sensitiveDefault;
        }

    EnvironmentEndpoint

    AbstractEndpoint的实现类之EnvironmentEndpoint--敏感数据

    1、构造函数

    @ConfigurationProperties(prefix = "endpoints.env")
    public class EnvironmentEndpoint extends AbstractEndpoint<Map<String, Object>> {
        public EnvironmentEndpoint() {
            super("env");
        }
            //调用AbstractEndpoint的构造函数
        public AbstractEndpoint(String id) {
            this(id, true);
        }

    最终,设置id为env,标识为敏感数据。

    2、实现的invoke(),代码如下:

    public Map<String, Object> invoke() {
        // 1. 先定义空的map返回值
        Map<String, Object> result = new LinkedHashMap<String, Object>();
        // 2. 将spring boot 中激活的profile 放入result中,key --> profile
        result.put("profiles", getEnvironment().getActiveProfiles());
        // 3. 获得PlaceholderSanitizingPropertyResolver --> 处理占位符,处理敏感数据
        PropertyResolver resolver = getResolver();
        // 4. 遍历environment 配置的PropertySource,依次处理之
        for (Entry<String, PropertySource<?>> entry : getPropertySourcesAsMap()
                .entrySet()) {
            PropertySource<?> source = entry.getValue();
            String sourceName = entry.getKey();
            if (source instanceof EnumerablePropertySource) {
                // 4.1 只针对EnumerablePropertySource 类型的PropertySource 进行处理--> 依次将属性添加到properties中,
                // 如果属性值为string,则在添加前进行占位符,数据脱敏的处理
                EnumerablePropertySource<?> enumerable = (EnumerablePropertySource<?>) source;
                Map<String, Object> properties = new LinkedHashMap<String, Object>();
                for (String name : enumerable.getPropertyNames()) {
                    Object property = source.getProperty(name);
                    Object resolved = property instanceof String
                            ? resolver.resolvePlaceholders((String) property) : property;
                    //调用Sanitizer类进行脱敏
                    properties.put(name, sanitize(name, resolved));
                }
                // 4.2 后置处理,该方法的实现是直接返回原始值,可以通过覆写的方式进行扩展
                properties = postProcessSourceProperties(sourceName, properties);
                if (properties != null) {
                    // 4.3 如果不为空,则添加到result中
                    result.put(sourceName, properties);
                }
            }
        }
        return result;
    }

    处理占位符,处理敏感数据:PlaceholderSanitizingPropertyResolver.java是EnvironmentEndpoint的内部类

    public PropertyResolver getResolver() {
    // 1. 实例化PlaceholderSanitizingPropertyResolver --> 处理占位符,处理敏感数据
    PlaceholderSanitizingPropertyResolver resolver = new PlaceholderSanitizingPropertyResolver(
            getPropertySources(), this.sanitizer);
    // 2. 设置ignoreUnresolvableNestedPlaceholders 为true
    resolver.setIgnoreUnresolvableNestedPlaceholders(true);
    return resolver;
    }

    PlaceholderSanitizingPropertyResolver继承了PropertySourcesPropertyResolver,这样就能对占位符进行处理了,又因为其内部持有Sanitizer(用于敏感数据脱敏),复写了getPropertyAsRawString,这样就能处理占位符,敏感数据了.代码如下:

            @Override
            protected String getPropertyAsRawString(String key) {
                String value = super.getPropertyAsRawString(key);
                return (String) this.sanitizer.sanitize(key, value);
            }

    3、EnvironmentEndpoint的属性配置,由于EnvironmentEndpoint被@ConfigurationProperties(prefix = “endpoints.env”)注解,因此可通过如下配置进行个性化配置:

    endpoints.env.id=env  
    endpoints.env.sensitive=true  
    endpoints.env.enabled=true

    同时,又因为其声明了如下方法:

    public void setKeysToSanitize(String... keysToSanitize) {
        this.sanitizer.setKeysToSanitize(keysToSanitize);
    }

    因此可以通过endpoints.env.keys-to-sanitize=xx,xx 来配置对指定的数据进行脱敏。脱敏配置

     4、EnvironmentEndpoint的自动化装配

    EnvironmentEndpoint的自动化装配是在EndpointAutoConfiguration中,代码如下:

    @Configuration
    @AutoConfigureAfter({ FlywayAutoConfiguration.class, LiquibaseAutoConfiguration.class })
    @EnableConfigurationProperties(EndpointProperties.class)
    public class EndpointAutoConfiguration {
        @Bean
        @ConditionalOnMissingBean
        public EnvironmentEndpoint environmentEndpoint() {
            return new EnvironmentEndpoint();
        }
    • @Bean注解:注册1个id为environmentEndpoint,类型为EnvironmentEndpoint的bean
    • @ConditionalOnMissingBean注解:当beanFactory中不存在EnvironmentEndpoint类型的bean时注册
    InfoEndpoint

    AbstractEndpoint的实现类之InfoEndpoint,用于暴露应用信息。

    其字段和构造器如下:

    private final List<InfoContributor> infoContributors;
    public InfoEndpoint(List<InfoContributor> infoContributors) {
        super("info", false);
        Assert.notNull(infoContributors, "Info contributors must not be null");
        this.infoContributors = infoContributors;
    }

    其内部持有了BeanFactory中所有InfoContributor类型的bean,其通过构造器注入。

    2、invoke 实现如下:

    public Map<String, Object> invoke() {
            Info.Builder builder = new Info.Builder();
            for (InfoContributor contributor : this.infoContributors) {
                contributor.contribute(builder);
            }
            Info build = builder.build();
            return build.getDetails();
    }

    通过遍历其内部的持有infoContributors,因此调用其contribute将info的数据添加到Info.Builder中,最后通过Info.Builder构建出Info,返回Info持有的details(建造者模式). Info中的details为Map.

    InfoContributor接口用于向Info$Builder添加信息,关于这部分的内容,我们后续文章有分析.这里就不在赘述了.

    3、InfoEndpoint的属性配置

    @ConfigurationProperties(prefix = "endpoints.info")
    public class InfoEndpoint extends AbstractEndpoint<Map<String, Object>> {

    因此可通过如下进行配置:

    endpoints.info.id=info  
    endpoints.info.sensitive=true  
    endpoints.info.enabled=true

    4、InfoEndpoint的自动化装配–>在EndpointAutoConfiguration中,代码如下:

        @Bean
        @ConditionalOnMissingBean
        public InfoEndpoint infoEndpoint() throws Exception {
            return new InfoEndpoint(this.infoContributors == null
                    ? Collections.<InfoContributor>emptyList() : this.infoContributors);
        }

    和EnvironmentEndpoint一样。

    RequestMappingEndpoint:

    AbstractEndpoint的实现类之RequestMappingEndpoint,由于RequestMappingEndpoint同时也实现了ApplicationContextAware接口,因此,在初始化该类时会注入applicationContext。这个类的作用是打印Spring MVC 映射信息。

    1、构造函数

        public RequestMappingEndpoint() {
            super("mappings");
        }

    因此, RequestMappingEndpoint的id为 mappings,默认为敏感

    2、invoke 实现如下:

    public Map<String, Object> invoke() {
        Map<String, Object> result = new LinkedHashMap<String, Object>();
        // 1. 从handlerMappings中获取HandlerMapping,默认情况下handlerMappings是不存在数据的
        extractHandlerMappings(this.handlerMappings, result);
        // 2. 从applicationContext中获取AbstractUrlHandlerMapping类型的bean,依次将其注册的handler 添加进去.
        extractHandlerMappings(this.applicationContext, result);
        // 3. 从methodMappings中获取HandlerMapping,默认情况下methodMappings是不存在数据的
        extractMethodMappings(this.methodMappings, result);
        // 3. 从applicationContext中获取AbstractUrlHandlerMapping类型的bean,依次获得其持有的HandlerMethods,进行处理.
        extractMethodMappings(this.applicationContext, result);
        return result;
    }

    从applicationContext中获取

        protected void extractHandlerMappings(ApplicationContext applicationContext,
                Map<String, Object> result) {
            if (applicationContext != null) {
                Map<String, AbstractUrlHandlerMapping> mappings = applicationContext
                        .getBeansOfType(AbstractUrlHandlerMapping.class);
                for (Entry<String, AbstractUrlHandlerMapping> mapping : mappings.entrySet()) {
                    Map<String, Object> handlers = getHandlerMap(mapping.getValue());
                    for (Entry<String, Object> handler : handlers.entrySet()) {
                        result.put(handler.getKey(),
                                Collections.singletonMap("bean", mapping.getKey()));
                    }
                }
            }
        }

    获得AbstractUrlHandlerMapping类型的bean,此时有4个:
    beanNameHandlerMapping=org.springframework.web.servlet.handler.BeanNameUrlHandlerMapping
    resourceHandlerMapping=org.springframework.web.servlet.handler.SimpleUrlHandlerMapping
    faviconHandlerMapping=org.springframework.web.servlet.handler.SimpleUrlHandlerMapping
    welcomePageHandlerMapping=org.springframework.boot.autoconfigure.web.WebMvcAutoConfiguration$WelcomePageHandlerMapping

    依次遍历mappings:
    获得AbstractUrlHandlerMapping中注册的handler,key–> path,value–>handler
    依次遍历handlerss,存入结果集中,存入的key–>AbstractUrlHandlerMapping的id,value={bean=AbstractUrlHandlerMapping中注册的handler的路径}

    从methodMappings中获取HandlerMapping,默认情况下methodMappings是不存在数据的

    从applicationContext中获取AbstractUrlHandlerMapping类型的bean,依次获得其持有的HandlerMethods,进行处理.代码如下:

        protected void extractMethodMappings(
                Collection<AbstractHandlerMethodMapping<?>> methodMappings,
                Map<String, Object> result) {
            for (AbstractHandlerMethodMapping<?> mapping : methodMappings) {
                Map<?, HandlerMethod> methods = mapping.getHandlerMethods();
                for (Map.Entry<?, HandlerMethod> entry : methods.entrySet()) {
                    result.put(String.valueOf(entry.getKey()), Collections
                            .singletonMap("method", String.valueOf(entry.getValue())));
                }
            }
        }

    获得AbstractUrlHandlerMapping类型的bean
    依次遍历AbstractUrlHandlerMapping中注册的handler,添加至结果集中,key–> Handler 映射路径 ,value = {bean = AbstractHandlerMethodMapping的id,method=HandlerMethod}

    3、RequestMappingEndpoint的配置

    @ConfigurationProperties(prefix = "endpoints.mappings")
    public class RequestMappingEndpoint extends AbstractEndpoint<Map<String, Object>>
            implements ApplicationContextAware {

    从类上的配置注解,可知:

    endpoints.mappings.enabled= # Enable the endpoint.
    endpoints.mappings.id= # Endpoint identifier.
    endpoints.mappings.sensitive= # Mark if the endpoint exposes sensitive information.

    4、RequestMappingEndpoint的自动装配

        @Configuration
        @ConditionalOnClass(AbstractHandlerMethodMapping.class)
        protected static class RequestMappingEndpointConfiguration {
    
            @Bean
            @ConditionalOnMissingBean
            public RequestMappingEndpoint requestMappingEndpoint() {
                RequestMappingEndpoint endpoint = new RequestMappingEndpoint();
                return endpoint;
            }
    
        }

    当满足如下两个条件时创建requestMappingEndpoint,即注册1个id为requestMappingEndpoint,类型为RequestMappingEndpoint的bean:

    @ConditionalOnClass(AbstractHandlerMethodMapping.class) –> 在beanFactory中存在AbstractHandlerMethodMapping类型的bean时生效
    @ConditionalOnMissingBean–>在beanFactory中不存在RequestMappingEndpoint类型的bean时生效

    DumpEndpoint

    AbstractEndpoint的实现类之DumpEndpoint,这个类的作用是打印线程信息,为敏感

    1、构造函数

        public DumpEndpoint() {
            super("dump");
        }

    2、invoke()方法实现

        @Override
        public List<ThreadInfo> invoke() {
            return Arrays
                    .asList(ManagementFactory.getThreadMXBean().dumpAllThreads(true, true));
        }

    调用了ThreadMXBean的dumpAllThreads来返回所有活动线程的线程信息,并带有堆栈跟踪和同步信息。 当此方法返回时,返回数组中包含的一些线程可能已经终止。其中两个参数指的意义如下:

    第1个–>如果为 true,则转储所有锁定的监视器。
    第2个–>如果为 true,则转储所有锁定的可拥有同步器。

     ThreadMXBean怎么读取Thread信息见《JMX学习一

    3、DumpEndpoint的配置

    @ConfigurationProperties(prefix = "endpoints.dump")
    public class DumpEndpoint extends AbstractEndpoint<List<ThreadInfo>> {

    可知:

    endpoints.dump.enabled= # Enable the endpoint.
    endpoints.dump.id= # Endpoint identifier.
    endpoints.dump.sensitive= # Mark if the endpoint exposes sensitive information.

    4、DumpEndpoint自动化装配:

    @Bean
    @ConditionalOnMissingBean
    public DumpEndpoint dumpEndpoint() {
        return new DumpEndpoint();
    }

    FlywayEndpoint

    AbstractEndpoint的实现类之FlywayEndpoint,Flyway是一款开源的数据库版本管理工具。

    ShutdownEndpoint

    AbstractEndpoint的实现类之ShutdownEndpoint,作用是关闭应用。这个类继承了ApplicationAware,得到applicationContext,关闭spring容器调用applicationContext.close()方法。

    1、构造函数

        public ShutdownEndpoint() {
            super("shutdown", true, false);
        }

    id为shutdown,为敏感,关闭应用的endpoin默认是关闭的,不启用的。

    2、invoke()方法实现

        @Override
        public Map<String, Object> invoke() {
                    //如果context为null,直接返回
            if (this.context == null) {
                return NO_CONTEXT_MESSAGE;
            }
                    //context不为null,先返回,再启动一个线程,在该线程中,先sleep 5秒后,然后调用了ShutdownEndpoint中持有的context的close方法进行关闭.
            try {
                return SHUTDOWN_MESSAGE;
            }
            finally {
                Thread thread = new Thread(new Runnable() {
                    @Override
                    public void run() {
                        try {
                            Thread.sleep(500L);
                        }
                        catch (InterruptedException ex) {
                            Thread.currentThread().interrupt();
                        }
                        ShutdownEndpoint.this.context.close();
                    }
                });
                thread.setContextClassLoader(getClass().getClassLoader());
                thread.start();
            }
        }

    在返回后,还希望做点啥,用try--finally。

    3、ShutdownEndpoint的配置

    @ConfigurationProperties(prefix = "endpoints.shutdown")
    public class ShutdownEndpoint extends AbstractEndpoint<Map<String, Object>>
            implements ApplicationContextAware {

    可知配置有:

    endpoints.shutdown.enabled= # Enable the endpoint.
    endpoints.shutdown.id= # Endpoint identifier.
    endpoints.shutdown.sensitive= # Mark if the endpoint exposes sensitive information.

    4、ShutdownEndpoint自动化装配:

    @Bean
    @ConditionalOnMissingBean
    public ShutdownEndpoint shutdownEndpoint() {
        return new ShutdownEndpoint();
    }

    AutoConfigurationReportEndpoint

    AbstractEndpoint的实现类之AutoConfigurationReportEndpoint,作用是暴露ConditionEvaluationReport.。

    1、构造函数

        public AutoConfigurationReportEndpoint() {
            super("autoconfig");
        }

    id为autoconfig,为敏感

    2、invoke()方法实现

        @Override
        public Report invoke() {
            return new Report(this.autoConfigurationReport);
        }

    Report为其内部类

        @JsonPropertyOrder({ "positiveMatches", "negativeMatches", "exclusions" })
        @JsonInclude(Include.NON_EMPTY)
        public static class Report {
    • @JsonPropertyOrder–>作用在类上,被用来指明当序列化时需要对属性做排序,它有2个属性:一个是alphabetic:布尔类型,表示是否采用字母拼音顺序排序,默认是为false,即不排序
    • @JsonInclude–> Report中的属性值为空集合则不进行展示

    Report 有如下字段:

    // 匹配的
    private final MultiValueMap<String, MessageAndCondition> positiveMatches;
    
    // 不匹配的
    private final Map<String, MessageAndConditions> negativeMatches;
    
    // 去除的
    private final List<String> exclusions;
    
    // 一般为null
    private final Report parent;

    其中MessageAndCondition封装了ConditionAndOutcome中的condition,message以进行更好的展示(json友好).其类上声明了如下注解:

    @JsonPropertyOrder({ "condition", "message" })

    因此在进行输出的时候,先输出condition,再输出message.

    构造器如下:

            public MessageAndCondition(ConditionAndOutcome conditionAndOutcome) {
                Condition condition = conditionAndOutcome.getCondition();
                ConditionOutcome outcome = conditionAndOutcome.getOutcome();
                this.condition = ClassUtils.getShortName(condition.getClass());
                if (StringUtils.hasLength(outcome.getMessage())) {
                    this.message = outcome.getMessage();
                }
                else {
                    this.message = (outcome.isMatch() ? "matched" : "did not match");
                }
            }

    赋值condition为ConditionAndOutcome中的Condition的短类名.
    赋值message:如果ConditionAndOutcome中的Message有值则直接赋值,否则,如果对应的Condition匹配,则赋值为matched,否则赋值为did not match。

    回到report,构造器如下:

    public Report(ConditionEvaluationReport report) {
    this.positiveMatches = new LinkedMultiValueMap<String, MessageAndCondition>();
    this.negativeMatches = new LinkedHashMap<String, MessageAndConditions>();
    // 1. 通过report#getExclusions 获得不进行加载的bean
    this.exclusions = report.getExclusions();
    // 2. 
    for (Map.Entry<String, ConditionAndOutcomes> entry : report
    .getConditionAndOutcomesBySource().entrySet()) {
    // 2.1 如果该配置生效条件都匹配,则加入到positiveMatches,否则,加入到negativeMatches
    if (entry.getValue().isFullMatch()) {
        add(this.positiveMatches, entry.getKey(), entry.getValue());
    }
    else {
        add(this.negativeMatches, entry.getKey(), entry.getValue());
    }
    }
    // 3. 如果report存在父report,则进行初始化Report 赋值为当前类的parent 属性
    boolean hasParent = report.getParent() != null;
    this.parent = (hasParent ? new Report(report.getParent()) : null);
    }

    通过ConditionEvaluationReport#getExclusions 获得不进行加载的bean,赋值为exclusions
    调用ConditionEvaluationReport#getConditionAndOutcomesBySource 获得ConditionEvaluationReport中持有匹配信息,返回的map中,key–> 匹配类名,ConditionAndOutcomes–> 匹配结果.

    依次遍历第2步的返回值–>如果该配置生效条件都匹配,则加入到positiveMatches,否则,加入到negativeMatches.其中add 代码如下:

            private void add(MultiValueMap<String, MessageAndCondition> map, String source,
                    ConditionAndOutcomes conditionAndOutcomes) {
                String name = ClassUtils.getShortName(source);
                for (ConditionAndOutcome conditionAndOutcome : conditionAndOutcomes) {
                    map.add(name, new MessageAndCondition(conditionAndOutcome));
                }
            }

    因此positiveMatches,negativeMatches 中的key为配置类的简单类名.
    如果report存在父report,则进行初始化Report 赋值为当前类的parent 属性.一般来说,是不存在父report的

    属性配置(因为有@ConfigurationProperties(prefix = “endpoints.autoconfig”) 注解):

    endpoints.autoconfig.enabled= # Enable the endpoint.
    endpoints.autoconfig.id= # Endpoint identifier.
    endpoints.autoconfig.sensitive= # Mark if the endpoint exposes sensitive information.

    自动装配:

    同样还是在EndpointAutoConfiguration中,代码如下:

    @Bean
    @ConditionalOnBean(ConditionEvaluationReport.class)
    @ConditionalOnMissingBean(search = SearchStrategy.CURRENT)
    public AutoConfigurationReportEndpoint autoConfigurationReportEndpoint() {
        return new AutoConfigurationReportEndpoint();
    }

    @Bean –> 注册1个id为autoConfigurationReportEndpoint,类型为AutoConfigurationReportEndpoint的bean
    @ConditionalOnBean(ConditionEvaluationReport.class)–>beanFactory中存在ConditionEvaluationReport类型的bean时生效
    @ConditionalOnMissingBean(search = SearchStrategy.CURRENT)–> 在当前上下文中不存在AutoConfigurationReportEndpoint类型的bean时生效

    ConditionEvaluationReport

    AutoConfigurationReportEndpoint 是通过ConditionEvaluationReport 来进行暴露信息.

    ConditionEvaluationReport 字段如下:

    private static final String BEAN_NAME = "autoConfigurationReport";
    
    // 如果一个配置类中内部配置类不匹配,则在其外部类的所对应的ConditionAndOutcomes中添加1个AncestorsMatchedCondition
    private static final AncestorsMatchedCondition ANCESTOR_CONDITION = new AncestorsMatchedCondition();
    
    // key-->配置类类名,ConditionAndOutcomes-->匹配条件结果的封装
    private final SortedMap<String, ConditionAndOutcomes> outcomes = new TreeMap<String, ConditionAndOutcomes>();
    
    // 是否添加AncestorsMatchedCondition,默认为false
    private boolean addedAncestorOutcomes;
    
    // 父ConditionEvaluationReport,一般为null
    private ConditionEvaluationReport parent;
    
    // 去除加载的配置
    private List<String> exclusions = Collections.emptyList();
    
    // 在ConditionEvaluationReportAutoConfigurationImportListener#onAutoConfigurationImportEvent 添加,用于保存还没有
    // 执行判断的class
    private Set<String> unconditionalClasses = new HashSet<String>();

    ConditionEvaluationReport 实例化过程如下:

    在SpringApplication#run中会调用AbstractApplicationContext#refresh,在refresh会调用PostProcessorRegistrationDelegate#invokeBeanFactoryPostProcessors,在该方法中最终会调用到AutoConfigurationImportSelector#selectImports方法.在该方法中会调用fireAutoConfigurationImportEvents,代码如下:AutoConfigurationImportSelector.fireAutoConfigurationImportEvents()

        private void fireAutoConfigurationImportEvents(List<String> configurations,
                Set<String> exclusions) {
            List<AutoConfigurationImportListener> listeners = getAutoConfigurationImportListeners();
            if (!listeners.isEmpty()) {
                AutoConfigurationImportEvent event = new AutoConfigurationImportEvent(this,
                        configurations, exclusions);
                for (AutoConfigurationImportListener listener : listeners) {
                    invokeAwareMethods(listener);
                    listener.onAutoConfigurationImportEvent(event);
                }
            }
        }

    会加载/META-INF/spring.factories 中配置的org.springframework.boot.autoconfigure.AutoConfigurationImportListener,实例化后,依次调用其onAutoConfigurationImportEvent 方法. spring.factories 配置如下:

    org.springframework.boot.autoconfigure.AutoConfigurationImportListener=
    org.springframework.boot.autoconfigure.condition.ConditionEvaluationReportAutoConfigurationImportListener

    因此此处会调用ConditionEvaluationReportAutoConfigurationImportListener#onAutoConfigurationImportEvent.代码如下:

        @Override
        public void onAutoConfigurationImportEvent(AutoConfigurationImportEvent event) {
            if (this.beanFactory != null) {
                ConditionEvaluationReport report = ConditionEvaluationReport
                        .get(this.beanFactory);
                report.recordEvaluationCandidates(event.getCandidateConfigurations());
                report.recordExclusions(event.getExclusions());
            }
        }

    A、实例化ConditionEvaluationReport,代码如下:

    public static ConditionEvaluationReport get(
    ConfigurableListableBeanFactory beanFactory) {
        synchronized (beanFactory) {
            ConditionEvaluationReport report;
            // 1. 如果当前beanFactory包含autoConfigurationReport定义的话,就从beanFactory中获取,
            if (beanFactory.containsSingleton(BEAN_NAME)) {
                report = beanFactory.getBean(BEAN_NAME,         ConditionEvaluationReport.class);
            }
            else {
                // 否则就实例化一个,然后进行注册
                report = new ConditionEvaluationReport();
                beanFactory.registerSingleton(BEAN_NAME, report);
            }
            // 2. 如果存在父容器的话,就从父容器中获取。
            locateParent(beanFactory.getParentBeanFactory(), report);
            return report;
        }
    }

    如果当前beanFactory包含autoConfigurationReport定义的话,就从beanFactory中获取,否则就实例化一个,然后进行注册

    如果存在父容器的话,就从父容器中获取,并将其赋值为当前context中获得的ConditionEvaluationReport的父ConditionEvaluationReport.代码如下:

        private static void locateParent(BeanFactory beanFactory,
                ConditionEvaluationReport report) {
            if (beanFactory != null && report.parent == null
                    && beanFactory.containsBean(BEAN_NAME)) {
                report.parent = beanFactory.getBean(BEAN_NAME,
                        ConditionEvaluationReport.class);
            }
        }

    一般都是null,不会执行的。

    B、设置unconditionalClasses为event#getCandidateConfigurations的返回值。
    C、设置exclusions为event#getExclusions的返回值。

    在AutoConfigurationReportEndpoint中是通过Report来进行暴露信息的,而在其构造器中,调用了ConditionEvaluationReport#getConditionAndOutcomesBySource方法,代码如下:

    public Map<String, ConditionAndOutcomes> getConditionAndOutcomesBySource() {
        if (!this.addedAncestorOutcomes) {
            // 1. 如果addedAncestorOutcomes 设为false,则依次遍历outcomes,如果一个配置类中内部配置类不匹配,则在其外部类的所对应的ConditionAndOutcomes中添加1个AncestorsMatchedCondition
            for (Map.Entry<String, ConditionAndOutcomes> entry : this.outcomes.entrySet()) {
                if (!entry.getValue().isFullMatch()) {
                    addNoMatchOutcomeToAncestors(entry.getKey());
                }
            }
            this.addedAncestorOutcomes = true;
        }
        return Collections.unmodifiableMap(this.outcomes);
    }

    如果addedAncestorOutcomes 设为false,则依次遍历outcomes,如果一个配置类中内部配置类不匹配,则在其外部类的所对应的ConditionAndOutcomes中添加1个AncestorsMatchedCondition

    返回outcomes.

    问题来了, outcomes 中的数据是如何添加的?

    答案: 有2处.

    还是在AutoConfigurationImportSelector#selectImports中,会调用其filter方法.代码如下

    private List<String> filter(List<String> configurations,
    AutoConfigurationMetadata autoConfigurationMetadata) {
        long startTime = System.nanoTime();
        String[] candidates = configurations.toArray(new     String[configurations.size()]);
        boolean[] skip = new boolean[candidates.length];
        boolean skipped = false;
        // 1. 获取META-INFspring.factories/中配置的org.springframework.boot.autoconfigure.AutoConfigurationImportFilter,.OnClassCondition.依次进行遍历之
        // 此时获得的是org.springframework.boot.autoconfigure.condition
        for (AutoConfigurationImportFilter filter : getAutoConfigurationImportFilters()) {
            // 1.1 进行属性注入
            invokeAwareMethods(filter);
            // 1.2 调用AutoConfigurationImportFilter#match 进行判断,依次遍历其返回值,如果返回的是false,则说明该配置为跳过,并将skipped设置为true
            // 获得AutoConfigurationMetadata中配置的            ConditionalOnClass,如果不会空,则依次遍历之,看是否在当前类路径下存在
            // 如果不匹配的话,则调用            ConditionEvaluationReport.#ecordConditionEvaluation 进行记录
            // 由于此时AutoConfigurationMetadata 什么都没有配置,因此此步骤相当于空操作,最终会在第2步返回
            boolean[] match = filter.match(candidates, autoConfigurationMetadata);
            for (int i = 0; i < match.length; i++) {
                if (!match[i]) {
                    skip[i] = true;
                    skipped = true;
                }
            }
        }
        // 2. 如果skipped 等于false,则直接返回configurations,说明没有配置是需要跳过的
        if (!skipped) {
            return configurations;
        }
        // 3. 依次遍历candidates,如果该配置是不进行跳过的,则添加至result中进行返回
        List<String> result = new ArrayList<String>(candidates.length);
        for (int i = 0; i < candidates.length; i++) {
            if (!skip[i]) {
                result.add(candidates[i]);
            }
        }
        return new ArrayList<String>(result);
    }

    获取META-INFspring.factories/中配置的org.springframework.boot.autoconfigure.AutoConfigurationImportFilter,.OnClassCondition.依次进行遍历之.此时获得的是org.springframework.boot.autoconfigure.condition
    进行属性注入
    调用AutoConfigurationImportFilter#match 进行判断,依次遍历其返回值,如果返回的是false,则说明该配置为跳过,并将skipped设置为true.由于此时调用的是OnClassCondition,其判断逻辑为获得AutoConfigurationMetadata中配置的ConditionalOnClass,如果不会空,则依次遍历之,看是否在当前类路径下存在.如果不匹配的话,则调用ConditionEvaluationReport#ecordConditionEvaluation 进行记录.由于此时AutoConfigurationMetadata 什么都没有配置,因此此步骤相当于空操作,最终会在第2步返回

    如果skipped 等于false,则直接返回configurations,说明没有配置是需要跳过的
    依次遍历candidates,如果该配置是不进行跳过的,则添加至result中进行返回

    在ConfigurationClassParser#processConfigurationClass进行解析加载配置类时,会调用ConditionEvaluator#shouldSkip,在该方法中,会因此遍历配置类配置的@Conditional所对应的处理类.此时,如果处理类是SpringBootCondition的子类的话,就会调用ConditionEvaluationReport进行记录匹配结果. 代码如下:

    private void recordEvaluation(ConditionContext context, String classOrMethodName,
    ConditionOutcome outcome) {
        if (context.getBeanFactory() != null) {
            ConditionEvaluationReport.get(context.getBeanFactory())
    .recordConditionEvaluation(classOrMethodName, this, outcome);
        }
    }

    LiquibaseEndpoint

    默认不生效,这里就不进行分析了
    BeansEndpoint

    BeansEndpoint的作用 –> 暴露关于beans的json视图.如果Environment 中设置了spring.liveBeansView.mbeanDomain,则所有spring 上下文的bean都会展示.否则只会展示当前的上下文中的bean.默认是没有配置的

    BeansEndpoint,构造器如下:

        public BeansEndpoint() {
            super("beans");
        }

    id为beans,默认为敏感

    BeansEndpoint的字段如下:

    // 继承自LiveBeansView,用于生成json格式的数据
    private final HierarchyAwareLiveBeansView liveBeansView = new HierarchyAwareLiveBeansView();
    
    // json 解析器对象
    private final JsonParser parser = JsonParserFactory.getJsonParser();

    由于BeansEndpoint实现了ApplicationContextAware接口,因此当前初始化时,会调用其setApplicationContext方法,代码如下:

        @Override
        public void setApplicationContext(ApplicationContext context) throws BeansException {
            if (context.getEnvironment()
                    .getProperty(LiveBeansView.MBEAN_DOMAIN_PROPERTY_NAME) == null) {
                this.liveBeansView.setLeafContext(context);
            }
        }

    如果没有设置spring.liveBeansView.mbeanDomain的属性,则将HierarchyAwareLiveBeansView中的leafContext设置为传入的ApplicationContext(通常是当前应用所对应的上下文)

    invoke实现如下:

        @Override
        public List<Object> invoke() {
            return this.parser.parseList(this.liveBeansView.getSnapshotAsJson());
        }

    调用HierarchyAwareLiveBeansView#getSnapshotAsJson 生成json串.代码如下:

    public String getSnapshotAsJson() {
    
        if (this.leafContext == null) {
            return super.getSnapshotAsJson();
        }
        // 2. 将leafContext的整个继承关系都添加到contexts中,即:如果给定的leafContext 存在父context,则一直递归的添加至contexts
        // 直至顶级容器,然后调用LiveBeansView#generateJson 来生成json串
        return generateJson(getContextHierarchy());
    }

    如果leafContext 等于null,则调用LiveBeansView#getSnapshotAsJson来生成json串.一般都会执行第2步

    将leafContext的整个继承关系都添加到contexts中,即:如果给定的leafContext 存在父context,则一直递归的添加至contexts.代码如下:

            private Set<ConfigurableApplicationContext> getContextHierarchy() {
                Set<ConfigurableApplicationContext> contexts = new LinkedHashSet<ConfigurableApplicationContext>();
                ApplicationContext context = this.leafContext;
                while (context != null) {
                    contexts.add(asConfigurableContext(context));
                    context = context.getParent();
                }
                return contexts;
            }

    将leafContext的整个继承关系都添加到contexts中,即:如果给定的leafContext 存在父context,则一直递归的添加至contexts直至顶级容器。

    调用LiveBeansView#generateJson 来生成json串.在该方法中没有使用第3方的json解析库的目的是为了避免对其进行依赖.返回的格式为数组格式.代码如下:

    由于BeansEndpoint声明了@ConfigurationProperties(prefix = “endpoints.beans”),因此可以通过如下属性来配置:

    endpoints.beans.id
    endpoints.beans.sensitive
    endpoints.beans.enabled

    自动化配置:

    beansEndpoint 是在EndpointAutoConfiguration 中进行配置的,代码如下:

    @Bean
    @ConditionalOnMissingBean
    public BeansEndpoint beansEndpoint() {
        return new BeansEndpoint();
    }

    @Bean –> 注册1个id为 beansEndpoint,类型为BeansEndpoint的bean
    @ConditionalOnMissingBean–> 当beanFactory中不存在BeansEndpoint类型的bean时生效

    ConfigurationPropertiesReportEndpoint

    作用:–>暴露被@ConfigurationProperties 注解的bean的属性.为了保护数据,将敏感数据进行了脱敏

    字段:

    private static final String CONFIGURATION_PROPERTIES_FILTER_ID = "configurationPropertiesFilter";
    
    // 数据脱敏
    private final Sanitizer sanitizer = new Sanitizer();
    
    // 通过实现ApplicationContextAware 进行自动注入
    private ApplicationContext context;
    
    public ConfigurationPropertiesReportEndpoint() {
        super("configprops");
    }

    invoke 实现:

    public Map<String, Object> invoke() {
    return extract(this.context);
    }

    调用:

    protected Map<String, Object> extract(ApplicationContext context) {
        // Serialize beans into map structure and sanitize values
        ObjectMapper mapper = new ObjectMapper();
        // 1. 配置ObjectMapper的属性
        configureObjectMapper(mapper);
        // 2. 抽取数据
        return extract(context, mapper);
    }

    对ObjectMapper 进行配置:

    代码如下:

        protected void configureObjectMapper(ObjectMapper mapper) {
            mapper.configure(SerializationFeature.FAIL_ON_EMPTY_BEANS, false);
            mapper.configure(SerializationFeature.WRITE_NULL_MAP_VALUES, false);
            applyConfigurationPropertiesFilter(mapper);
            applySerializationModifier(mapper);
        }

    SerializationFeature.FAIL_ON_EMPTY_BEANS –> 设置当对于给定的类型无法访问时如何处理,true–> 抛出异常,false–> 返回null
    SerializationFeature.WRITE_NULL_MAP_VALUES –> 设置当map对于的value为null时如何处理,treu–>序列化,false–> 跳过

    applyConfigurationPropertiesFilter 代码如下:

        private void applyConfigurationPropertiesFilter(ObjectMapper mapper) {
            mapper.setAnnotationIntrospector(
                    new ConfigurationPropertiesAnnotationIntrospector());
            mapper.setFilterProvider(new SimpleFilterProvider()
                    .setDefaultFilter(new ConfigurationPropertiesPropertyFilter()));
        }

    设置AnnotationIntrospector为ConfigurationPropertiesAnnotationIntrospector.该类的作用是:获得@JsonFilter注解的值,如果获取不到,则返回configurationPropertiesFilter.代码如下:

    private static class ConfigurationPropertiesAnnotationIntrospector
    extends JacksonAnnotationIntrospector {
        @Override
        public Object findFilterId(Annotated a) {
            // 1. 获得@JsonFilter注解的值,如果获取不到,则返回    configurationPropertiesFilter
            Object id = super.findFilterId(a);
            if (id == null) {
                id = CONFIGURATION_PROPERTIES_FILTER_ID;
            }
            return id;
        }
    }

    设置默认的过滤器为ConfigurationPropertiesPropertyFilter,该类的作用是进行如下规则的过滤:
    类名以$$开头的被过滤掉
    自我引用的字段被过滤掉
    当在序列化时抛出异常时过滤掉

    代码如下:

    public void serializeAsField(Object pojo, JsonGenerator jgen,
    SerializerProvider provider, PropertyWriter writer) throws Exception {
    if (writer instanceof BeanPropertyWriter) {
    try {
    // 1. 自我引用的字段被过滤掉
    if (pojo == ((BeanPropertyWriter) writer).get(pojo)) {
    if (logger.isDebugEnabled()) {
    logger.debug("Skipping '" + writer.getFullName() + "' on '"
    + pojo.getClass().getName()
    + "' as it is self-referential");
    }
    return;
    }
    }
    catch (Exception ex) {
    // 2. 当在序列化时抛出异常时过滤掉
    if (logger.isDebugEnabled()) {
    logger.debug("Skipping '" + writer.getFullName() + "' on '"
    + pojo.getClass().getName() + "' as an exception "
    + "was thrown when retrieving its value", ex);
    }
    return;
    }
    }
    // 3. 序列化字段
    super.serializeAsField(pojo, jgen, provider, writer);
    }

    设置序列化工厂中的方法序列化器为GenericSerializerModifier.代码如下:

    public List<BeanPropertyWriter> changeProperties(SerializationConfig config,
    BeanDescription beanDesc, List<BeanPropertyWriter> beanProperties) {
    List<BeanPropertyWriter> result = new ArrayList<BeanPropertyWriter>();
    for (BeanPropertyWriter writer : beanProperties) {
    boolean readable = isReadable(beanDesc, writer);
    if (readable) {
    result.add(writer);
    }
    }
    return result;
    }

    依次遍历beanProperties,如果可读的话,则添加至result中.可读的判断逻辑如下:
    根据bean 的类型和属性的类型 找出对应的set方法
    如果set方法 存在或者是同一包下的类或者是map或者集合的子类,则返回true

    调用extract方法抽取数据.代码如下:

    private Map<String, Object> extract(ApplicationContext context, ObjectMapper mapper) {
    Map<String, Object> result = new HashMap<String, Object>();
    // 1. 获得beanFactory中ConfigurationBeanFactoryMetaData类型的bean
    ConfigurationBeanFactoryMetaData beanFactoryMetaData = getBeanFactoryMetaData(
    context);
    // 2. 获得被@ConfigurationProperties注解的bean,key--> bean的id,value --> bean
    Map<String, Object> beans = getConfigurationPropertiesBeans(context,
    beanFactoryMetaData);
    // 3. 依次遍历beans
    for (Map.Entry<String, Object> entry : beans.entrySet()) {
    String beanName = entry.getKey();
    Object bean = entry.getValue();
    Map<String, Object> root = new HashMap<String, Object>();
    String prefix = extractPrefix(context, beanFactoryMetaData, beanName, bean);
    // 3.1 获得@ConfigurationProperties注解的前缀,添加至root中,key-->prefix,value-->@ConfigurationProperties注解的前缀
    root.put("prefix", prefix);
    // 3.2 
    root.put("properties", sanitize(prefix, safeSerialize(mapper, bean, prefix)));
    // 3.3 添加至result中,key--> bean id,value-->map{prefix=xx,properties=xx}
    result.put(beanName, root);
    }
    // 4. 如果ApplicationContext存在父容器的,则递归调用extract提取数据,key为parent.
    if (context.getParent() != null) {
    result.put("parent", extract(context.getParent(), mapper));
    }
    return result;
    }

    获得beanFactory中ConfigurationBeanFactoryMetaData类型的bean
    获得被@ConfigurationProperties注解的bean,key–> bean的id,value –> bean

    依次遍历beans
    获得@ConfigurationProperties注解的前缀,添加至root中,key–>prefix,value–>@ConfigurationProperties注解的前缀

    对数据进行脱敏.代码如下:

    private Map<String, Object> sanitize(String prefix, Map<String, Object> map) {
    for (Map.Entry<String, Object> entry : map.entrySet()) {
    String key = entry.getKey();
    String qualifiedKey = (prefix.isEmpty() ? prefix : prefix + ".") + key;
    Object value = entry.getValue();
    if (value instanceof Map) {
    // 1. 如果对应的属性值为Map,List 则递归调用sanitize 进行数据脱敏
    map.put(key, sanitize(qualifiedKey, (Map<String, Object>) value));
    }
    else if (value instanceof List) {
    map.put(key, sanitize(qualifiedKey, (List<Object>) value));
    }
    else {
    // 2. 如果属性名包含password", "secret", "key", "token", ".*credentials.*", "vcap_services,则将其替换为******
    value = this.sanitizer.sanitize(key, value);
    value = this.sanitizer.sanitize(qualifiedKey, value);
    map.put(key, value);
    }
    }
    return map;
    }

    如果对应的属性值为Map,List 则递归调用sanitize 进行数据脱敏
    否则如果属性名包含password, secret, key, token, .*credentials.*, vcap_services,则将其替换为**
    添加至result中,key–> bean id,value–>map{prefix=xx,properties=xx}

    如果ApplicationContext存在父容器的,则递归调用extract提取数据,key为parent.

    由于ConfigurationPropertiesReportEndpoint 被@ConfigurationProperties(prefix = “endpoints.configprops”)注解,因此可通过如下属性配置:

    endpoints.configprops.id=configprops 
    endpoints.configprops.sensitive=true 
    endpoints.configprops.enabled=true 
    endpoints.configprops.keys-to-sanitize=password,secret

    之所以可以通过 endpoints.configprops.keys-to-sanitize 进行配置,是因为ConfigurationPropertiesReportEndpoint声明了如下方法:

    public void setKeysToSanitize(String... keysToSanitize) {
    this.sanitizer.setKeysToSanitize(keysToSanitize);
    }

    自动化配置(在EndpointAutoConfiguration中配置):

    代码如下:

    @Bean
    @ConditionalOnMissingBean
    public ConfigurationPropertiesReportEndpoint configurationPropertiesReportEndpoint() {
    return new ConfigurationPropertiesReportEndpoint();
    }

    @Bean –> 注册1个id为configurationPropertiesReportEndpoint,类型为ConfigurationPropertiesReportEndpoint的Bean
    @ConditionalOnMissingBean –> beanFactory中不存在ConfigurationPropertiesReportEndpoint类型的bean时生效

    TraceEndpoint

    作用:–> 该端点用来返回基本的HTTP跟踪信息。默认情况下,跟踪信息的存储采用org.springframework.boot.actuate.trace.InMemoryTraceRepository实现的内存方式,始终保留最近的100条请求记录.其中,返回的Trace定义如下:

    public final class Trace {
    
    // 时间戳
    private final Date timestamp;
    
    // 保存用于分析上下文信息,例如HTTP头
    private final Map<String, Object> info;
    
    public Trace(Date timestamp, Map<String, Object> info) {
    super();
    Assert.notNull(timestamp, "Timestamp must not be null");
    Assert.notNull(info, "Info must not be null");
    this.timestamp = timestamp;
    this.info = info;
    }
    
    public Date getTimestamp() {
    return this.timestamp;
    }
    
    public Map<String, Object> getInfo() {
    return this.info;
    }
    }

    字段:

    private final TraceRepository repository;

    这里我们有必要说明一下TraceRepository:

    TraceRepository 是用来保存Trace的. 接口定义如下:

    public interface TraceRepository {

    // 返回保存的Trace
    List<Trace> findAll();

    // 进行添加
    void add(Map<String, Object> traceInfo);
    }

    TraceRepository 只要1个实现–>InMemoryTraceRepository.其字段如下:

    // 容量为100
    private int capacity = 100;
    // 是否倒序展示,默认为false
    private boolean reverse = true;
    // 容器,用于保存Trace
    private final List<Trace> traces = new LinkedList<Trace>();

    findAll只需简单的返回保存的traces即可.实现如下:

    public List<Trace> findAll() {
    synchronized (this.traces) {
    return Collections.unmodifiableList(new ArrayList<Trace>(this.traces));
    }
    }

    add,代码如下:

    public void add(Map<String, Object> map) {
    // 1. 实例化Trace,时间戳为当前时间
    Trace trace = new Trace(new Date(), map);
    synchronized (this.traces) {
    // 2. 如果traces中的容量大于等于了阈值,则进行删除.如果reverse等于true 则删除最后1个,否则,删除第1个
    while (this.traces.size() >= this.capacity) {
    this.traces.remove(this.reverse ? this.capacity - 1 : 0);
    }
    // 3. 进行添加,如果reverse等于true 则添加至第1个,否则,添加至最后
    if (this.reverse) {
    this.traces.add(0, trace);
    }
    else {
    this.traces.add(trace);
    }
    }
    }

    实例化Trace,时间戳为当前时间
    如果traces中的容量大于等于了阈值,则进行删除.如果reverse等于true 则删除最后1个,否则,删除第1个
    进行添加,如果reverse等于true 则添加至第1个,否则,添加至最后

    TraceRepository 是在何处配置的呢?

    在TraceRepositoryAutoConfiguration中,代码如下:

    @ConditionalOnMissingBean(TraceRepository.class)
    @Bean
    public InMemoryTraceRepository traceRepository() {
    return new InMemoryTraceRepository();
    }

    @Bean –> 注册1个id为traceRepository,类型为InMemoryTraceRepository的bean
    @ConditionalOnMissingBean(TraceRepository.class)–> 当beanFactory中不存在TraceRepository类型的bean时生效

    invoke 实现:

    @Override
    public List<Trace> invoke() {
    return this.repository.findAll();
    }

    只需调用TraceRepository# findAll,返回保存的Trace即可.

    属性配置,因此其被@ConfigurationProperties(prefix = “endpoints.trace”)注解,因此可以通过如下属性进行配置:

    endpoints.trace.id=trace
    endpoints.trace.sensitive=true
    endpoints.trace.enabled=true

    自动装配:

    在EndpointAutoConfiguration中进行了装配,代码如下:

    @Bean
    @ConditionalOnMissingBean
    public TraceEndpoint traceEndpoint() {
    return new TraceEndpoint(this.traceRepository == null
    ? new InMemoryTraceRepository() : this.traceRepository);
    }

    @Bean–> 注册1个id为traceEndpoint,类型为TraceEndpoint的bean
    @ConditionalOnMissingBean–> 当beanFactory中不存在TraceEndpoint类型的bean时生效

    这里提1个问题,TraceEndpoint 是通过TraceRepository获取Trace,那么TraceRepository中的Trace是如何保存的呢?

    答案:

    通过WebRequestTraceFilter(Filter)来实现. WebRequestTraceFilter实现了Ordered接口,指定了其在过滤器链中的顺序,代码如下:

    private int order = Ordered.LOWEST_PRECEDENCE - 10;

    public int getOrder() {
    return this.order;
    }

    WebRequestTraceFilter中的字段如下:

    private static final Log logger = LogFactory.getLog(WebRequestTraceFilter.class);

    // debug时使用.如果启用的话,并且log的trace级别可用的话,则打印请求头信息,默认为false
    private boolean dumpRequests = false;

    // Not LOWEST_PRECEDENCE, but near the end, so it has a good chance of catching all
    // enriched headers, but users can add stuff after this if they want to
    private int order = Ordered.LOWEST_PRECEDENCE - 10;

    private final TraceRepository repository;

    private ErrorAttributes errorAttributes;

    private final TraceProperties properties;

    其中TraceProperties的定义如下:

    @ConfigurationProperties(prefix = "management.trace")
    public class TraceProperties {

    private static final Set<Include> DEFAULT_INCLUDES;

    static {
    Set<Include> defaultIncludes = new LinkedHashSet<Include>();
    defaultIncludes.add(Include.REQUEST_HEADERS);
    defaultIncludes.add(Include.RESPONSE_HEADERS);
    defaultIncludes.add(Include.COOKIES);
    defaultIncludes.add(Include.ERRORS);
    defaultIncludes.add(Include.TIME_TAKEN);
    DEFAULT_INCLUDES = Collections.unmodifiableSet(defaultIncludes);
    }


    private Set<Include> include = new HashSet<Include>(DEFAULT_INCLUDES);

    public Set<Include> getInclude() {
    return this.include;
    }

    public void setInclude(Set<Include> include) {
    this.include = include;
    }
    }

    默认配置的是Include.REQUEST_HEADERS, Include.RESPONSE_HEADERS, Include.COOKIES ,Include.ERRORS, Include.TIME_TAKEN. 可以通过如下进行配置

    management.trace.include=REQUEST_HEADERS,RESPONSE_HEADERS
    1

    可选值为org.springframework.boot.actuate.trace.TraceProperties.Include.这里就不在贴出了

    WebRequestTraceFilter 继承了OncePerRequestFilter,因此只需实现doFilterInternal即可,代码如下:

    protected void doFilterInternal(HttpServletRequest request,
    HttpServletResponse response, FilterChain filterChain)
    throws ServletException, IOException {
    long startTime = System.nanoTime();
    // 1. 获得trace
    Map<String, Object> trace = getTrace(request);
    // 2. 打印日志-->如果log的trace级别可用的话并且dumpRequests等于true,则打印请求头信息,默认为false
    logTrace(request, trace);
    int status = HttpStatus.INTERNAL_SERVER_ERROR.value();
    try {
    // 3. 继续过滤器链的过滤,最后获得响应状态
    filterChain.doFilter(request, response);
    status = response.getStatus();
    }
    finally {
    // 4. 添加Http请求耗时统计-->如果TraceProperties中配置了Include#TIME_TAKEN(默认配置了),则添加到trace中,key为timeTaken,value-->当前时间-开始时间的毫秒值
    addTimeTaken(trace, startTime);
    // 5. 添加响应头信息
    enhanceTrace(trace, status == response.getStatus() ? response
    : new CustomStatusResponseWrapper(response, status));
    // 6. 添加至TraceRepository 中
    this.repository.add(trace);
    }
    }

    记录开始时间

    获得trace.代码如下:

    protected Map<String, Object> getTrace(HttpServletRequest request) {
    // 1. 获得HttpSession
    HttpSession session = request.getSession(false);
    // 2. 获得javax.servlet.error.exception,所对应的异常--> 当spring mvc 出现异常时,会加入到javax.servlet.error.exception中
    Throwable exception = (Throwable) request
    .getAttribute("javax.servlet.error.exception");
    // 3. 获得Principal,如果返回null,说明没有该请求没有进行验证
    Principal userPrincipal = request.getUserPrincipal();
    Map<String, Object> trace = new LinkedHashMap<String, Object>();
    Map<String, Object> headers = new LinkedHashMap<String, Object>();
    // 4. 添加请求方法,请求路径,请求头到trace中
    trace.put("method", request.getMethod());
    trace.put("path", request.getRequestURI());
    trace.put("headers", headers);
    if (isIncluded(Include.REQUEST_HEADERS)) {
    headers.put("request", getRequestHeaders(request));
    }
    // 省略掉默认不执行的代码....
    // 5. 如果有异常并且errorAttributes不等于null,则记录error
    if (isIncluded(Include.ERRORS) && exception != null
    && this.errorAttributes != null) {
    trace.put("error", this.errorAttributes
    .getErrorAttributes(new ServletRequestAttributes(request), true));
    }
    return trace;
    }

    继续过滤器链的过滤,最后获得响应状态

    添加Http请求耗时统计–>如果TraceProperties中配置了Include#TIME_TAKEN(默认配置了),则添加到trace中,key为timeTaken,value–>当前时间-开始时间的毫秒值.代码如下:

    private void addTimeTaken(Map<String, Object> trace, long startTime) {
    long timeTaken = System.nanoTime() - startTime;
    add(trace, Include.TIME_TAKEN, "timeTaken",
    "" + TimeUnit.NANOSECONDS.toMillis(timeTaken));
    }

    添加响应头信息.代码如下:

    protected void enhanceTrace(Map<String, Object> trace, HttpServletResponse response) {
    if (isIncluded(Include.RESPONSE_HEADERS)) {
    Map<String, Object> headers = (Map<String, Object>) trace.get("headers");
    headers.put("response", getResponseHeaders(response));
    }
    }

    getResponseHeaders 代码如下:

    private Map<String, String> getResponseHeaders(HttpServletResponse response) {
    Map<String, String> headers = new LinkedHashMap<String, String>();
    // 1. 依次遍历响应头,添加至headers 中,key--> 响应头名,value-->响应头对应的值
    for (String header : response.getHeaderNames()) {
    String value = response.getHeader(header);
    headers.put(header, value);
    }
    // 2. 如果TraceProperties中没有配置Include#COOKIES,则在headers中删除key为Set-Cookie的值.默认是配置了的,因此不会删除
    if (!isIncluded(Include.COOKIES)) {
    headers.remove("Set-Cookie");
    }
    // 3. 向headers 中添加 key--> status,value-->响应状态码
    headers.put("status", "" + response.getStatus());
    return headers;
    }

    遍历响应头,添加至headers 中,key–> 响应头名,value–>响应头对应的值
    如果TraceProperties中没有配置Include#COOKIES,则在headers中删除key为Set-Cookie的值.默认是配置了的,因此不会删除
    向headers 中添加 key–> status,value–>响应状态码
    添加至TraceRepository中

    WebRequestTraceFilter 是如何配置的呢?

    答案: 在TraceWebFilterAutoConfiguration中.代码如下:

    @Bean
    @ConditionalOnMissingBean
    public WebRequestTraceFilter webRequestLoggingFilter(BeanFactory beanFactory) {
    WebRequestTraceFilter filter = new WebRequestTraceFilter(this.traceRepository,
    this.traceProperties);
    if (this.errorAttributes != null) {
    filter.setErrorAttributes(this.errorAttributes);
    }
    return filter;
    }

    @Bean–> 注册1个id为webRequestLoggingFilter,类型为WebRequestTraceFilter的bean
    @ConditionalOnMissingBean–> 当beanFactory中不存在WebRequestTraceFilter类型的bean时生效

    同时由于TraceWebFilterAutoConfiguration声明了如下注解:

    @Configuration
    @ConditionalOnClass({ Servlet.class, DispatcherServlet.class, ServletRegistration.class })
    @AutoConfigureAfter(TraceRepositoryAutoConfiguration.class)
    @ConditionalOnProperty(prefix = "endpoints.trace.filter", name = "enabled", matchIfMissing = true)
    @EnableConfigurationProperties(TraceProperties.class)

    @Configuration–> 配置类
    @ConditionalOnClass({ Servlet.class, DispatcherServlet.class, ServletRegistration.class })–> 在当前的类路径下存在Servlet.class, DispatcherServlet.class, ServletRegistration.class时生效
    @AutoConfigureAfter(TraceRepositoryAutoConfiguration.class)–> 在TraceRepositoryAutoConfiguration之后进行自动装配,这样就可以自动注入TraceRepository
    @ConditionalOnProperty(prefix = “endpoints.trace.filter”, name = “enabled”, matchIfMissing = true)–> 当配置有endpoints.trace.filter.enabled 等于true时生效,如果没有配置,默认生效
    @EnableConfigurationProperties(TraceProperties.class)–> 可以通过management.trace.include 进行配置.

    转:https://blog.csdn.net/qq_26000415/article/details/79060258

  • 相关阅读:
    ResponsiveSlides.js最轻量级的幻灯片插件
    jQuery.extend 函数详解
    HQL查询语言的使用介绍
    使用Hibernate命名查询
    Jquery.KinSlideshow图片轮播插件
    关于文件下载自己忽略的文件中文命名。。。
    DB2中的ROW_NUMBER() OVER()函数用法
    css ul li 水平布局问题
    彻底弄懂css中单位px和em,rem的区别 转的自己看
    taglib 自定义标签
  • 原文地址:https://www.cnblogs.com/duanxz/p/11302110.html
Copyright © 2011-2022 走看看