Is it possible in JavaSparkContext ?

JavaSparkContext jsc = new JavaSparkContext(conf);
JavaRDD<String>lines = jsc.textFile(args[0]);

If yes , does its programmer's responsibilty to first calculate splits
locations and then instantiate spark context with preferred locations?

How does its achieved in MR2 with yarn, there is Application Master
specifies split locations to ResourceManager before acquiring the node
managers ?



On Mon, Jun 1, 2015 at 7:24 AM, bit1...@163.com <bit1...@163.com> wrote:

> Take a look at the following SparkContext constructor variant that tries
> to honor the data locality in YARN mode.
>
>   /**
> * :: DeveloperApi ::
> * Alternative constructor for setting preferred locations where Spark will
> create executors.
> *
> * @param preferredNodeLocationData used in YARN mode to select nodes to
> launch containers on.
> * Can be generated using
> [[org.apache.spark.scheduler.InputFormatInfo.computePreferredLocations]]
> * from a list of input files or InputFormats for the application.
> */
> @DeveloperApi
> def this(config: SparkConf, preferredNodeLocationData: Map[String,
> Set[SplitInfo]]) = {
> this(config)
> this.preferredNodeLocationData = preferredNodeLocationData
> }
>
> ------------------------------
> bit1...@163.com
>
>
> *From:* Shushant Arora <shushantaror...@gmail.com>
> *Date:* 2015-05-31 22:54
> *To:* user <user@spark.apache.org>
> *Subject:* data localisation in spark
>
> I want to understand how  spark takes care of data localisation in cluster
> mode when run on YARN.
>
> 1.Driver program asks ResourceManager for executors. Does it tell yarn's
> RM to check HDFS blocks of input data and then allocate executors to it.
> And executors remain fixed throughout application or driver program asks
> for new executors when it submits another job in same application , since
> in spark new job is created for each action . If executors are fixed then
> for second job achieving data localisation is impossible?
>
>
>
> 2.When executors are done with their processing, does they are marked as
> free in ResourceManager's resoruce queue and  executors directly tell this
> to Rm  instead of via driver's ?
>
> Thanks
> Shushant
>
>

Reply via email to