Oinone远程调用链路源码分析

前提

源码分析版本是 5.1.x版本

概要

在服务启动时,获取注解REMOTE的函数,通过dubbo的泛化调用发布。在调用函数时,通过dubbo泛化调用获取结果。

注册服务者

  1. 在spring 启动方法installOrLoad中初始化
  2. 寻找定义REMOTE的方法
  3. 组装dubbo的服务配置
  4. 组装服务对象实现引用,内容如下,用于注册
    • 调用前置处理
      • 放信息到SessionApi
      • 函数调用链追踪,放到本地TransmittableThreadLocal
      • 从redis中获取到的数据进行反序列化并存在到本地的线程里
      • Trace信息,放一份在sessionApi中 和ThreadLocal
    • 调用函数执行
    • 返回数据转成特定格式
  5. 通过线程组调用dubbo的ServiceConfig.export 服务发布

时序图

Oinone远程调用链路源码分析
注册

源码分析

根据条件判断,确定向dubbo进行服务发布
RemoteServiceLoader

public void publishService(List<FunctionDefinition> functionList,Map<String,Runnable> isPublished) {
        // 因为泛化接口只能控制到namespace,控制粒度不能到fun级别,这里进行去重处理
        Map<String, Function> genericNamespaceMap = new HashMap<>();
        for (FunctionDefinition functionDefinition : functionList) {
            Function function = new Function(functionDefinition)

            try {
               //定义REMOTE, 才给予远程调用
                if (FunctionOpenEnum.REMOTE.in(function.getOpen()) && !ClassUtils.isInterface(function.getClazz())) {
                    genericNamespaceMap.putIfAbsent(RegistryUtils.getRegistryInterface(function), function);
                }
            } catch (PamirsException e) {
            }
        }
        // 发布远程服务
        for (String namespace : genericNamespaceMap.keySet()) {
            Function function = genericNamespaceMap.get(namespace);
            if(isPublished.get(RegistryUtils.getRegistryInterface(function)) == null){
                // 发布,注册远程函数服务,底层使用dubbo的泛化调用
                Runnable registryTask = () -> remoteRegistry.registryService(function);
                isPublished.put(RegistryUtils.getRegistryInterface(function),registryTask);
            }else{

            }
        }
    }

构造ServiceConfig方法,设置成泛化调用,进行发布export()
DefaultRemoteRegistryComponent

     public void registryGenericService(String interfaceName, List<MethodConfig> methods,
                                       String group, String version, Integer timeout, Integer retries) {
        ....
        try {
            ServiceConfig<GenericService> service = new ServiceConfig<>();
            // 服务接口名
            service.setInterface(interfaceName);
            // 服务对象实现引用
            service.setRef(genericService(interfaceName));
            if (null != methods) {
                service.setMethods(methods);
            }
            // 声明为泛化接口
            service.setGeneric(Boolean.TRUE.toString());
            // 基础元数据
            constructService(group, version, timeout, retries, service);
            service.export();
        } catch (Exception e) {
           .....
        }
    }

// 服务对象实现引用
private GenericService genericService(String interfaceName) {
        return (method, parameterTypes, args) -> {
            PamirsSession.clear();
            Function function = Objects.requireNonNull(PamirsSession.getContext()).getFunction(RegistryUtils.getFunctionNamespace(method), RegistryUtils.getFunctionFun(method));
            if (log.isDebugEnabled()) {
                log.debug("interfaceName: " + interfaceName + ", isDataManage: " + function.isDataManager());
            }
            try {
                //前置处理:服务提供者,对请求参数进行对象化拆解,并对请求携带的上下文进行处理
                // 放信息到SessionApi
                // 函数调用链追踪,放到本地TransmittableThreadLocal
                // CommonMetaDataCacheApi.computeMetaData() 从redis中获取到的数据进行反序列化并存在到本地的TL中
                // DataAuditApi.computeDataAuditSession() Trace信息,放一份在sessionApi中 和ThreadLocal中
                Object[] args1 = Spider.getDefaultExtension(RemoteRequestArgApi.class).providerHandle(function.getNamespace(), function.getFun(), args, function.getArguments());

                Object result = FunEngine.get().exclude(ScriptType.REMOTE).run(function, args1);

                //后置处理:服务提供者,对结果进行对象化封装、携带请求上下文进行处理
                return Spider.getDefaultExtension(RemoteResponseApi.class).providerHandle(function, method, result);
            } catch (Throwable e) {
                return Spider.getDefaultExtension(RemoteResponseApi.class).providerExceptionHandle(function, method, e);
            } finally {
                PamirsSession.clear();
            }
        };
    }

注册消费者

  1. 函数处理调用
  2. 注册服务消费者
    • 从ReferenceConfigCache获取泛化
  3. 调用dubbo泛化调用接口
  4. 获取返回信息
    • 获取用户id,放入PamirsSession
    • 如果开启debug模式
      • 存入DEBUG_THREAD_LOCAL本地线程
    • 返回格式
      • IWrapper
      • Pagination
      • Result

时序图

Oinone远程调用链路源码分析

源码分析

泛化调用dobbo接口,并解析返回对象
RemoteComputer

public Object compute(Function function, Object... args) {

        .....
        List<Arg> functionArguments = function.getArguments();
        String methodName = RegistryUtils.getGenericServiceMethodName(function);
        String[] argTypes = FunctionUtils.fetchArgTypes(functionArguments);
        //前置处理:服务消费者,对请求参数进行对象化封装、携带请求上下文进行处理
        Object[] arguments = getRemoteRequestApi().consumerHandle(function.getNamespace(),function.getFun(), args, functionArguments);
        // 泛化调用
        Object result = invoke(function, methodName, argTypes, arguments);
        .....
        //后置处理:服务消费者,对返回结果进行对象化拆解,并对结果携带的上下文进行处理
        // 数据转成IWrapper/Pagination/Result
        return getRemoteResponseApi().consumerHandle(function, result);
}

// 配置请求信息,通过$invoke 实际调用
private Object invoke(Function function, String methodName, String[] argTypes, Object[] arguments) {
        Object result;
        String configCdOwnSign = getSessionFillOwnSignApi().getConfigCdOwnSign();
        if (StringUtils.isBlank(configCdOwnSign)) {
.....
        } else {
            try {
            // 获取服务,由于是泛化调用,所以获取的一定是GenericService类型
                GenericService remoteClient = CommonApiFactory.getApi(RemoteRegistry.class).registryOriginConsumer(function);
         // 第一个参数是需要调用的方法名
         // 第二个参数是需要调用的方法的参数类型数组,为String数组,里面存入参数的全类名。
         // 第三个参数是需要调用的方法的参数数组,为Object数组,里面存入需要的参数
                result = remoteClient.$invoke(methodName, argTypes, arguments);
            } catch (RpcException e) {
                ....            }
        }
        return result;
    }

从缓存中获取泛化
DefaultRemoteRegistryComponent

    public GenericService registryGenericConsumer(String interfaceName, List<MethodConfig> methods,
                                                  String group, String version, Integer timeout, Integer retries) {
        ....
        // 创建服务引用配置
        ReferenceConfig<GenericService> reference = new ReferenceConfig<>();
        reference.setInterface(interfaceName);
        // 设置为泛化调用
        reference.setGeneric(Boolean.TRUE.toString());
        if (null != methods) {
            reference.setMethods(methods);
        }
        constructReference(group, version, timeout, retries, reference);
        return ReferenceConfigCache.getCache().get(reference);
    }

名词解释

泛化调用是指在调用方没有服务方提供的API(SDK)的情况下,对服务方进行调用,并且可以正常拿到调用结果
泛化调用(客户端泛化)
实现泛化实现(服务端泛化)

Oinone社区 作者:oinone原创文章,如若转载,请注明出处:https://doc.oinone.top/backend/17027.html

访问Oinone官网:https://www.oinone.top获取数式Oinone低代码应用平台体验

Like (0)
oinone's avataroinone
Previous 2024年9月3日 pm12:53
Next 2024年9月5日 pm8:13

相关推荐

  • Oinone连接外部数据源方案

    场景描述 在实际业务场景中,有是有这样的需求:链接外部数据进行数据的获取;通常的做法:1、【推荐】通过集成平台的数据连接器,链接外部数据源进行数据操作;2、项目代码中链接数据源,即通过程序代码操作外部数据源的数据; 本篇文章只介绍通过程序代码操作外部数据源的方式. 整体方案 Oinone管理外部数据源,即yml中配置外部数据源; 后端通过Mapper的方式进行数据操作(增/删/查/改); 调用Mapper接口的时候,指定到外部数据源; 详细步骤 1、数据源配置(application.yml), 与正常的数据源配置一样 out_ds_name(外部数据源别名): driverClassName: com.mysql.cj.jdbc.Driver type: com.alibaba.druid.pool.DruidDataSource # local环境配置调整 url: jdbc:mysql://ip(host):端口/数据库Schema?useSSL=false&allowPublicKeyRetrieval=true&useServerPrepStmts=true&cachePrepStmts=true&useUnicode=true&characterEncoding=utf8&serverTimezone=Asia/Shanghai&autoReconnect=true&allowMultiQueries=true username: 用户名 password: 命名 initialSize: 5 maxActive: 200 minIdle: 5 maxWait: 60000 timeBetweenEvictionRunsMillis: 60000 testWhileIdle: true testOnBorrow: false testOnReturn: false poolPreparedStatements: true asyncInit: true 2、外部数据源其他配置外部数据源限制创建表结构的执行,可以通过配置指定【不创建DB,不创建数据表】 persistence: global: auto-create-database: true auto-create-table: true ds: out_ds_name(外部数据源别名): # 不创建DB auto-create-database: false # 不创建数据表 auto-create-table: false 3、后端写Mapper SQL Mapper跟使用原生mybaits/mybaits-plus写法一样,无特殊限制; Mapper和SQL写到一起,或者分开两个文件都可以 4、Mapper被Service或者Action调用1)启动的Application中@MapperScan需要扫描到对应的包。2)调用是与普通bean一样(即调用方式跟传统的方式样),唯一的区别就是加上DsHintApi,即指定Mapper所使用的数据源。 @Autowired private ScheduleItemMapper scheduleItemMapper; public saveData(Object data) { ScheduleQuery scheduleQuery = new ScheduleQuery(); //scheduleQuery.setActionName(); try (DsHintApi dsHint = DsHintApi.use(“外部数据源名称”)) { List<ScheduleItem> scheduleItems = scheduleItemMapper.selectListForSerial(scheduleQuery); // 具体业务逻辑 } } 其他参考:如何自定义sql语句:https://doc.oinone.top/backend/4759.html

    2024年5月17日
    2.3K00
  • Oinone请求路由源码分析

    通过源码分析,从页面发起请求,如果通过graphQL传输到具体action的链路,并且在这之间做了哪些隐式处理分析源码版本5.1.x 请求流程大致如下: 拦截所有指定的请求 组装成graphQL请求信息 调用graphQL执行 通过hook拦截先执行 RsqlDecodeHook:rsql解密 UserHook: 获取用户信息, 通过cookies获取用户ID,再查表获取用户信息,放到本地Local线程里 RoleHook: 角色Hook FunctionPermissionHook: 函数权限Hook ,跳过权限拦截的实现放在这一层,对应的配置 pamirs: auth: fun-filter: – namespace: user.PamirsUserTransient fun: login #登录 – namespace: top.PetShop fun: action DataPermissionHook: 数据权限hook PlaceHolderHook:占位符转化替换hook RsqlParseHook: 解释Rsql hook SingletonModelUpdateHookBefore 执行post具体内容 通过hook拦截后执行 QueryPageHook4TreeAfter: 树形Parent查询优化 FieldPermissionHook: 字段权限Hook UserQueryPageHookAfter UserQueryOneHookAfter 封装执行结果信息返回 时序图 核心源码解析 拦截所有指定的请求 /pamirs/模块名RequestController @RequestMapping( value = "/pamirs/{moduleName:^[a-zA-Z][a-zA-Z0-9_]+[a-zA-Z0-9]$}", method = RequestMethod.POST ) public String pamirsPost(@PathVariable("moduleName") String moduleName, @RequestBody PamirsClientRequestParam gql, HttpServletRequest request, HttpServletResponse response) { } DefaultRequestExecutor 构建graph请求信息,并调用graph请求 () -> execute(GraphQL::execute, param), param private <T> T execute(BiFunction<GraphQL, ExecutionInput, T> executor, PamirsRequestParam param) { // 获取GraphQL请求信息,包含grapsh schema GraphQL graphQL = buildGraphQL(param); … ExecutionInput executionInput = ExecutionInput.newExecutionInput() .query(param.getQuery()) .variables(param.getVariables().getVariables()) .dataLoaderRegistry(Spider.getDefaultExtension(DataLoaderRegistryApi.class).dataLoader()) .build(); … // 调用 GraphQL的方法execute 执行 T result = executor.apply(graphQL, executionInput); … return result; } QueryAndMutationBinder 绑定graphQL读取写入操作 public static DataFetcher<?> dataFetcher(Function function, ModelConfig modelConfig) { if (isAsync()) { if (FunctionTypeEnum.QUERY.in(function.getType())) { return AsyncDataFetcher.async(dataFetchingEnvironment -> dataFetcherAction(function, modelConfig, dataFetchingEnvironment), ExecutorServiceApi.getExecutorService()); } else { return dataFetchingEnvironment -> dataFetcherAction(function, modelConfig, dataFetchingEnvironment); } } else { return dataFetchingEnvironment -> dataFetcherAction(function, modelConfig, dataFetchingEnvironment); } } private static Object dataFetcherAction(Function function, ModelConfig modelConfig, DataFetchingEnvironment environment) { try { SessionExtendUtils.tagMainRequest(); // 使用共享的请求和响应对象 return Spider.getDefaultExtension(ActionBinderApi.class) .action(modelConfig,…

    2024年8月21日
    6.1K02
  • 工作流-流程代办等页面自定义

    1. 审批/填写节点的视图页面 在界面设计器中创建对应模型的表单视图,可根据业务场景需要自定义所需流程待办的审批页面 2. 在审批/填写节点中选择刚创建的视图 在工作流待办数据权限可在节点数据权限中可对字段设置查看、编辑、隐藏

    2024年5月14日
    2.5K00
  • 环境运行时Jar版本控制

    环境运行时Jar版本控制 前景 为了避免基于低代码定义产生的元数据错乱。因此产生了运行时Jar版本检查功能。 现象 如果当前运行时依赖的Ja版本低于已安装版本,启动时会有如下类似信息提示: 解决 按照提示升级依赖Jar版本 通过启动参数 -PgoBack=true 强制覆盖安装当前运行时版本 java -jar 方式 java -jar xxx.jar -PgoBack=true [其他参数] mvn spring-boot run 方式 mvn clean compile spring-boot:run -Dspring-boot.run.arguments=”-PgoBack=true [其他参数]”

    2025年3月10日
    1.6K00
  • 树形下拉组件数据查询覆写

    一、背景与问题 Oinone 的树形下拉组件(TreeSelect)和树视图(Tree)在查询数据时,最终调用的是树节点所配置模型的数据管理器函数: 树接口 调用的数据管理器函数 场景 fetchAll queryListByWrapper 一次性全量加载 fetchChildren queryPage 按层级懒加载(默认) queryKeywords4Tree queryListByWrapper 关键字搜索 reverselyQuery queryListByWrapper 已选数据回填 官方文档原文:「所有树查询接口都会调用模型对应的数据管理器函数,可以通过重写的方式修改其查询结果。」 直接覆写的风险 如果直接在原模型(如 business.PamirsDepartment)上覆写 queryPage: // ⚠️ 危险写法:会影响所有使用 PamirsDepartment.queryPage 的场景 @Component @Model.model(PamirsDepartment.MODEL_MODEL) public class DeptQueryAction { @Function.Advanced(type = FunctionTypeEnum.QUERY, category = FunctionCategoryEnum.QUERY_PAGE) @Function.fun(FunctionConstants.queryPage) @Function(openLevel = {FunctionOpenEnum.LOCAL, FunctionOpenEnum.REMOTE, FunctionOpenEnum.API}) public Pagination<PamirsDepartment> queryPage(Pagination<PamirsDepartment> page, IWrapper<PamirsDepartment> queryWrapper) { // 追加过滤条件… return new PamirsDepartment().queryPage(page, queryWrapper); } } 该覆写会全局生效——不仅是树形下拉,所有调用 PamirsDepartment 的 queryPage 的地方(组织架构页面、人员管理列表、权限分配、框架内部关联查询等)都会受到影响,容易引发难以排查的副作用。 二、推荐方案:代理模型(PROXY)隔离 核心思路 创建一个继承原模型的代理模型(ModelTypeEnum.PROXY),拥有独立的模型命名空间 在代理模型中重新声明自关联字段(如 parent),将关联目标指向代理模型自身 在代理模型的 Action 中覆写数据管理器函数,追加树形场景专属的过滤逻辑 树形下拉的视图 DSL 中,<node> 的 model 指向代理模型 代理模型与原模型共享同一张数据表,但函数注册在各自独立的命名空间下,互不干扰。 框架官方示例 Oinone 框架自身就采用了这一模式:ResourceRegionProxyModel 继承 ResourceRegion,用于地区页面的代理展示,其扩展点 ResourceRegionProxyModelExtPoint 中的 queryPageAfter 只影响代理模型自身的查询。 三、完整实现步骤(以 PamirsDepartment 为例) 场景假设 业务需求:某个表单页面的「部门」树形下拉只需要展示当前公司(companyCode == 当前用户公司)且启用状态(dataStatus == ENABLED)的部门,而系统中其他使用 PamirsDepartment 的地方(组织架构、权限管理等)不受影响。 步骤 1:定义代理模型 package com.example.demo.model; import pro.shushi.pamirs.meta.annotation.Field; import pro.shushi.pamirs.meta.annotation.Model; import pro.shushi.pamirs.meta.enmu.ModelTypeEnum; import pro.shushi.pamirs.business.api.model.PamirsDepartment; import java.util.List; /** * 部门树形下拉专用代理模型 * <p> * 继承 PamirsDepartment,共享同一张数据表, * 但拥有独立的函数命名空间,覆写 queryPage 不影响原模型。 */ @Model.model(DeptTreeProxy.MODEL_MODEL) @Model.Advanced(type = ModelTypeEnum.PROXY) @Model(displayName = "部门(树选择)", labelFields = "name") public class DeptTreeProxy extends PamirsDepartment { public static final String MODEL_MODEL = "demo.DeptTreeProxy"; /** * 【必须重新声明】上级部门(自关联字段) * <p> * 基类 PamirsDepartment 中 parent 的类型是 PamirsDepartment, * 代理模型必须将其覆盖为指向自身(DeptTreeProxy), * 否则设计器/树组件无法在代理模型命名空间内构成完整的自关联关系, * 导致树形下拉无法正确展开层级。 */ @Field.many2one @Field.Relation(relationFields = {"parentCode"}, referenceFields = {"code"}) @Field(displayName = "上级部门")…

    后端 2026年7月27日
    6500

Leave a Reply

Please Login to Comment