spark-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mich Talebzadeh <mich.talebza...@gmail.com>
Subject Re: how to increase threads per executor
Date Thu, 02 Jun 2016 22:28:02 GMT
interesting. a vm with one core!

one simple test

can you try running with

--executor-cores=1

and see it works ok please



Dr Mich Talebzadeh



LinkedIn * https://www.linkedin.com/profile/view?id=AAEAAAAWh2gBxianrbJd6zP6AcPCCdOABUrV8Pw
<https://www.linkedin.com/profile/view?id=AAEAAAAWh2gBxianrbJd6zP6AcPCCdOABUrV8Pw>*



http://talebzadehmich.wordpress.com



On 2 June 2016 at 23:15, Andres M Jimenez T <adrz1@hotmail.com> wrote:

> Mich, thanks for your time,
>
>
> i am launching spark-submit as follows:
>
>
> bin/spark-submit --class com.example.SparkStreamingImpl --master
> spark://dev1.dev:7077 --verbose --driver-memory 1g --executor-memory 1g
> --conf "spark.driver.extraJavaOptions=-Dcom.sun.management.jmxremote
> -Dcom.sun.management.jmxremote.port=8090
> -Dcom.sun.management.jmxremote.rmi.port=8091
> -Dcom.sun.management.jmxremote.authenticate=false
> -Dcom.sun.management.jmxremote.ssl=false" --conf
> "spark.executor.extraJavaOptions=-Dcom.sun.management.jmxremote
> -Dcom.sun.management.jmxremote.port=8092
> -Dcom.sun.management.jmxremote.rmi.port=8093
> -Dcom.sun.management.jmxremote.authenticate=false
> -Dcom.sun.management.jmxremote.ssl=false" --conf
> "spark.scheduler.mode=FAIR" --conf /home/Processing.jar
>
>
> When i use --executor-cores=12 i get "Initial job has not accepted any
> resources; check your cluster UI to ensure that workers are registered and
> have sufficient resources".
>
>
> This, because my nodes are single core, but i want to use more than one
> thread per core, is this possible?
>
>
> root@dev1:/home/spark-1.6.1-bin-hadoop2.6# lscpu
> Architecture:          x86_64
> CPU op-mode(s):        32-bit, 64-bit
> Byte Order:            Little Endian
> CPU(s):                1
> On-line CPU(s) list:   0
> Thread(s) per core:    1
> Core(s) per socket:    1
> Socket(s):             1
> NUMA node(s):          1
> Vendor ID:             GenuineIntel
> CPU family:            6
> Model:                 58
> Model name:            Intel(R) Xeon(R) CPU E5-2690 v2 @ 3.00GHz
> Stepping:              0
> CPU MHz:               2999.999
> BogoMIPS:              5999.99
> Hypervisor vendor:     VMware
> Virtualization type:   full
> L1d cache:             32K
> L1i cache:             32K
> L2 cache:              256K
> L3 cache:              25600K
> NUMA node0 CPU(s):     0
>
>
> Thanks
>
>
>
> ------------------------------
> *From:* Mich Talebzadeh <mich.talebzadeh@gmail.com>
> *Sent:* Thursday, June 2, 2016 5:00 PM
> *To:* Andres M Jimenez T
> *Cc:* user@spark.apache.org
> *Subject:* Re: how to increase threads per executor
>
> What are passing as parameters to Spark-submit?
>
>
> ${SPARK_HOME}/bin/spark-submit \
>                 --executor-cores=12 \
>
> Also check
>
> http://spark.apache.org/docs/latest/configuration.html
> Configuration - Spark 1.6.1 Documentation
> <http://spark.apache.org/docs/latest/configuration.html>
> spark.apache.org
> Spark Configuration. Spark Properties. Dynamically Loading Spark
> Properties; Viewing Spark Properties; Available Properties. Application
> Properties; Runtime Environment
>
>
> Execution Behavior/spark.executor.cores
>
>
> HTH
>
>
> Dr Mich Talebzadeh
>
>
>
> LinkedIn * https://www.linkedin.com/profile/view?id=AAEAAAAWh2gBxianrbJd6zP6AcPCCdOABUrV8Pw
> <https://www.linkedin.com/profile/view?id=AAEAAAAWh2gBxianrbJd6zP6AcPCCdOABUrV8Pw>*
>
>
>
> http://talebzadehmich.wordpress.com
>
>
>
> On 2 June 2016 at 17:29, Andres M Jimenez T <adrz1@hotmail.com> wrote:
>
>> Hi,
>>
>>
>> I am working with Spark 1.6.1, using kafka direct connect for streaming
>> data.
>>
>> Using spark scheduler and 3 slaves.
>>
>> Kafka topic is partitioned with a value of 10.
>>
>>
>> The problem i have is, there is only one thread per executor running my
>> function (logic implementation).
>>
>>
>> Can anybody tell me how can i increase threads per executor to get better
>> use of CPUs?
>>
>>
>> Thanks
>>
>>
>> Here is the code i have implemented:
>>
>>
>> *Driver*:
>>
>>
>> JavaStreamingContext ssc = new JavaStreamingContext(conf, new
>> Duration(10000));
>>
>> //prepare streaming from kafka
>>
>> Set<String> topicsSet = new
>> HashSet<>(Arrays.asList("stage1-in,stage1-retry".split(",")));
>>
>> Map<String, String> kafkaParams = new HashMap<>();
>>
>> kafkaParams.put("metadata.broker.list", kafkaBrokers);
>>
>> kafkaParams.put("group.id", SparkStreamingImpl.class.getName());
>>
>>
>> JavaPairInputDStream<String, String> inputMessages =
>> KafkaUtils.createDirectStream(
>>
>> ssc,
>>
>> String.class,
>>
>> String.class,
>>
>> StringDecoder.class,
>>
>> StringDecoder.class,
>>
>> kafkaParams,
>>
>> topicsSet
>>
>> );
>>
>>
>> inputMessages.foreachRDD(new ForeachRDDFunction());
>>
>>
>> *ForeachFunction*:
>>
>>
>> class ForeachFunction implements VoidFunction<Tuple2<String, String>>
{
>>
>> private static final Counter foreachConcurrent =
>> ProcessingMetrics.metrics.counter( "foreach-concurrency" );
>>
>> public ForeachFunction() {
>>
>> LOG.info("Creating a new ForeachFunction");
>>
>> }
>>
>>
>> public void call(Tuple2<String, String> t) throws Exception {
>>
>> foreachConcurrent.inc();
>>
>> LOG.info("processing message [" + t._1() + "]");
>>
>> try {
>>
>> Thread.sleep(1000);
>>
>> } catch (Exception e) { }
>>
>> foreachConcurrent.dec();
>>
>> }
>>
>> }
>>
>>
>> *ForeachRDDFunction*:
>>
>>
>> class ForeachRDDFunction implements VoidFunction<JavaPairRDD<String,
>> String>> {
>>
>> private static final Counter foreachRDDConcurrent =
>> ProcessingMetrics.metrics.counter( "foreachRDD-concurrency" );
>>
>> private ForeachFunction foreachFunction = new ForeachFunction();
>>
>> public ForeachRDDFunction() {
>>
>> LOG.info("Creating a new ForeachRDDFunction");
>>
>> }
>>
>>
>> public void call(JavaPairRDD<String, String> t) throws Exception {
>>
>> foreachRDDConcurrent.inc();
>>
>> LOG.info("call from inputMessages.foreachRDD with [" +
>> t.partitions().size() + "] partitions");
>>
>> for (Partition p : t.partitions()) {
>>
>> if (p instanceof KafkaRDDPartition){
>>
>> LOG.info("partition [" + p.index() + "] with count [" +
>> ((KafkaRDDPartition) p).count() + "]");
>>
>> }
>>
>> }
>>
>> t.foreachAsync(foreachFunction);
>>
>> foreachRDDConcurrent.dec();
>>
>> }
>>
>> }
>>
>>
>> *The log from driver that tells me my RDD is partitioned to process in
>> parallel*:
>>
>>
>> [Stage 70:>  (3 + 3) / 20][Stage 71:>  (0 + 0) / 20][Stage 72:>  (0 + 0)
>> / 20]16/06/02 08:32:10 INFO SparkStreamingImpl: call from
>> inputMessages.foreachRDD with [20] partitions
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [0] with count [24]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [1] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [2] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [3] with count [19]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [4] with count [19]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [5] with count [20]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [6] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [7] with count [23]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [8] with count [21]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [9] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [10] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [11] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [12] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [13] with count [26]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [14] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [15] with count [27]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [16] with count [0]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [17] with count [16]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [18] with count [15]
>>
>> 16/06/02 08:32:10 INFO SparkStreamingImpl: partition [19] with count [0]
>>
>>
>> *The log from one of executors showing exactly one message per second was
>> processed (only by one thread)*:
>>
>>
>> 16/06/02 08:32:46 INFO SparkStreamingImpl: processing message
>> [f2b22bb9-3bd8-4e5b-b9fb-afa7e8c4deb8]
>>
>> 16/06/02 08:32:47 INFO SparkStreamingImpl: processing message
>> [e267cde2-ffea-4f7a-9934-f32a3b7218cc]
>>
>> 16/06/02 08:32:48 INFO SparkStreamingImpl: processing message
>> [f055fe3c-0f72-4f41-9a31-df544f1e1cd3]
>>
>> 16/06/02 08:32:49 INFO SparkStreamingImpl: processing message
>> [854faaa5-0abe-49a2-b13a-c290a3720b0e]
>>
>> 16/06/02 08:32:50 INFO SparkStreamingImpl: processing message
>> [1bc0a141-b910-45fe-9881-e2066928fbc6]
>>
>> 16/06/02 08:32:51 INFO SparkStreamingImpl: processing message
>> [67fb99c6-1ca1-4dfb-bffe-43b927fdec07]
>>
>> 16/06/02 08:32:52 INFO SparkStreamingImpl: processing message
>> [de7d5934-bab2-4019-917e-c339d864ba18]
>>
>> 16/06/02 08:32:53 INFO SparkStreamingImpl: processing message
>> [e63d7a7e-de32-4527-b8f1-641cfcc8869c]
>>
>> 16/06/02 08:32:54 INFO SparkStreamingImpl: processing message
>> [1ce931ee-b8b1-4645-8a51-2c697bf1513b]
>>
>> 16/06/02 08:32:55 INFO SparkStreamingImpl: processing message
>> [5367f3c1-d66c-4647-bb44-f5eab719031d]
>>
>>
>

Mime
View raw message