1、喜马拉雅数据计算平台XQL2017.11Outline XQL总览 系统演进过程 周边产品 经验总结 未来展望XQL总览 研发背景 总体架构 使用人群与场景XQL总览 研发背景XQL总览 研发背景?XQL总览 研发背景?CoreSQLXQL总览 总体架构?XQL总览 总体架构 memory:4T spark Task:200w+XQL job:4000+dataSource:hdfs、hive、hbase、es、kafka、mysql、pg fileFormat:parquet、orc、csv、json、xmlXQL总览 使用人群与场景系统演进过程 v1版本 v2版本 v3版本系统演进过程 v
2、1版本架构?v1?系统演进过程 v1版本通信逻辑clientSingle Spark Application on YarnSpark Driver JVMMaster共享SparkContextSpark Executor JVMSpark Executor JVMSpark Executor JVMSpark Executor JVMHttp serverLocalProxySpark RpcDownloadHandlerWorker隔离SparkSessionHDFSWorker启动注册clientSingle Spark Application on YarnSpark Driver
3、JVMMaster共享SparkContextSpark Executor JVMSpark Executor JVMSpark Executor JVMSpark Executor JVMHttp serverLocalProxySpark RpcDownloadHandlerWorker隔离SparkSessionHDFSWorker启动注册用户提交clientSingle Spark Application on YarnSpark Driver JVMMaster共享SparkContextSpark Executor JVMSpark Executor JVMSpark Execut
4、or JVMSpark Executor JVMHttp serverLocalProxySpark RpcDownloadHandlerWorker隔离SparkSessionHDFSWorker启动注册用户提交请求可用地址clientSingle Spark Application on YarnSpark Driver JVMMaster共享SparkContextSpark Executor JVMSpark Executor JVMSpark Executor JVMSpark Executor JVMHttp serverLocalProxySpark RpcDownloadHan
5、dlerWorker隔离SparkSessionHDFSWorker启动注册用户提交请求可用地址返回异步IDclientSingle Spark Application on YarnSpark Driver JVMMaster共享SparkContextSpark Executor JVMSpark Executor JVMSpark Executor JVMSpark Executor JVMHttp serverLocalProxySpark RpcDownloadHandlerWorker隔离SparkSessionHDFSWorker启动注册用户提交请求可用地址返回异步ID分配执行c
6、lientSingle Spark Application on YarnSpark Driver JVMMaster共享SparkContextSpark Executor JVMSpark Executor JVMSpark Executor JVMSpark Executor JVMHttp serverLocalProxySpark RpcDownloadHandlerWorker隔离SparkSessionHDFSWorker启动注册用户提交请求可用地址返回异步ID分配执行写HDFSclientSingle Spark Application on YarnSpark Driver