当前位置:首页 > 报告详情

SPARK CONNECT 中的依赖关系管理:简单、隔离、强大.pdf

上传人: 张** 编号:167573 2024-06-15 33页 1.03MB

1、2024 Databricks Inc.All rights reservedAkhil Gudesa,Hyukjin Kwon Akhil Gudesa,Hyukjin Kwon-R&D DatabricksR&D DatabricksJune 2024June 20241Dependency Management Dependency Management in Spark Connect:in Spark Connect:Simple,Isolated,Simple,Isolated,PowerfulPowerful2024 Databricks Inc.All rights reser

2、ved2024 Databricks Inc.All rights reserved2How can Dependencies Help You?How can Dependencies Help You?Custom ETLCustom ETLBring your own,or 3rd party libraries in your data transformation pipelinesUse OSS AI librariesUse OSS AI librariesHave the freedom to experiment with the wide variety of existi

3、ng ML open source libraries2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved3How can Dependencies Help You?How can Dependencies Help You?Python PackagesJARsCustom LibrariesPython PackagesJARsCustom LibrariesDataDataTransformed DataTransformed DataUser Defined FunctionUse

4、r Defined Function2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved4Dependencies in“Classic”SparkDependencies in“Classic”SparkCluster ScopedCluster ScopedEnvironment/dependencies are shared between all usersInterInter-User ConflictsUser ConflictsUsers may require differe

5、nt versions of the same dependencyStaticStaticUpdating previously-set dependencies requires a Driver restartShared DependenciesShared Dependencies2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reservedWhat is Spark Connect?What is Spark Connect?52024 Databricks Inc.All rights

6、reserved2024 Databricks Inc.All rights reserved6DEPENDENCY DEPENDENCY ISOLATIONISOLATIONvia Spark Connectvia Spark Connect2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved7Dependencies in Spark ConnectDependencies in Spark ConnectSpark Connect Spark Connect ClientClientS

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文主要介绍了Databricks中Spark Connect的依赖管理功能。Spark Connect允许在数据转换管道中使用自定义或第三方库,如OSS AI库,并提供了实验和使用各种ML开源库的自由。文章详细讨论了依赖管理在经典Spark中的问题,如集群范围内环境/依赖项共享导致的用户间冲突,以及更新先前设置的依赖项需要驱动器重启的静态问题。而Spark Connect通过客户端会话范围的环境/依赖项,提供了依赖隔离,解决了这些问题。此外,文章还展示了如何通过Spark Connect的AddArtifact API上传依赖项,并提供了Python和Scala的代码模板示例。最后,文章提到了即将到来的Spark 4.0的新特性和Databricks Connect的优势。
"如何简化Spark中的依赖管理?" "如何在Spark Connect中使用自定义ETL库?" "Spark Connect如何实现依赖隔离和动态环境管理?"
客服
商务合作
小程序
服务号
折叠