[ https://issues.apache.org/jira/browse/SPARK-3561?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14137326#comment-14137326 ]
Oleg Zhurakousky commented on SPARK-3561: ----------------------------------------- Patrick, thanks for following up. Indeed Spark does provide first-class extensibility mechanism at many different levels (shuffle, rdd, readers/writers, etc.), however, we believe it is missing a crucial one and that is the "execution context”. And while SparkContext itself could easily be extended or mixed in with a custom trait to achieve such customization, it is less then ideal extension mechanism, since it would require code modification every time user wants to swap an execution environment (e.g., from “local” in testing to “yarn” in prod). And in fact Spark already supports an externally configurable model where the target execution environment is managed through “master" URL. However, the _nature_, _implementation_ and most importantly _customization_ of these environments are internal to Spark. {code} master match { case "yarn-client" => case mesosUrl @ MESOS_REGEX(_) => . . . } {code} Further more, any additional integration and/or customization work that may come in the future would require modification to the above _case_ statement which I am also sure you’d agree is less then ideal integration style, since it would require a new release of Spark every time new _case_ statement is added. So essentially what we’re proposing is to formalize what has always been supported by Spark to an externally configurable model so customization around _*native functionality*_ of the target execution environment could be handled in a flexible and pluggable way. So in this model we are simply proposing a variation of the "chain of responsibility pattern” where DAG execution could be delegated to an _execution context_ with no change to end user programs or semantics. Based on our investigation we’ve identified 4 core operations which you can see in _JobExecutionContext_. Two of them provide access to source RDD creation thus allowing customization of data _sourcing_ (custom readers, direct block access etc.). One for _broadcast_ to integrate with broadcast capabilities provided natively. And last but not least is the main _execution delegate_ for the job - “runJob”. And while I am sure there will be more questions, I hope the above response clarifies the overall intention of this proposal > Native Hadoop/YARN integration for batch/ETL workloads > ------------------------------------------------------ > > Key: SPARK-3561 > URL: https://issues.apache.org/jira/browse/SPARK-3561 > Project: Spark > Issue Type: New Feature > Components: core > Affects Versions: 1.1.0 > Reporter: Oleg Zhurakousky > Labels: features > Fix For: 1.2.0 > > Attachments: SPARK-3561.pdf > > > Currently Spark provides integration with external resource-managers such as > Apache Hadoop YARN, Mesos etc. Specifically in the context of YARN, the > current architecture of Spark-on-YARN can be enhanced to provide > significantly better utilization of cluster resources for large scale, batch > and/or ETL applications when run alongside other applications (Spark and > others) and services in YARN. > Proposal: > The proposed approach would introduce a pluggable JobExecutionContext (trait) > - a gateway and a delegate to Hadoop execution environment - as a non-public > api (@DeveloperAPI) not exposed to end users of Spark. > The trait will define 4 only operations: > * hadoopFile > * newAPIHadoopFile > * broadcast > * runJob > Each method directly maps to the corresponding methods in current version of > SparkContext. JobExecutionContext implementation will be accessed by > SparkContext via master URL as > "execution-context:foo.bar.MyJobExecutionContext" with default implementation > containing the existing code from SparkContext, thus allowing current > (corresponding) methods of SparkContext to delegate to such implementation. > An integrator will now have an option to provide custom implementation of > DefaultExecutionContext by either implementing it from scratch or extending > form DefaultExecutionContext. > Please see the attached design doc for more details. > Pull Request will be posted shortly as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org