Unable to connect to Spark master

问题

I start my DataStax cassandra instance with Spark:

dse cassandra -k

I then run this program (from within Eclipse):

import org.apache.spark.sql.SQLContext
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext

object Start {

  def main(args: Array[String]): Unit = {
    println("***** 1 *****")
    val sparkConf = new SparkConf().setAppName("Start").setMaster("spark://127.0.0.1:7077")
    println("***** 2 *****")
    val sparkContext = new SparkContext(sparkConf)
    println("***** 3 *****")
  }
}

And I get the following output

***** 1 *****
***** 2 *****
Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
15/12/29 15:27:50 INFO SparkContext: Running Spark version 1.5.2
15/12/29 15:27:51 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
15/12/29 15:27:51 INFO SecurityManager: Changing view acls to: nayan
15/12/29 15:27:51 INFO SecurityManager: Changing modify acls to: nayan
15/12/29 15:27:51 INFO SecurityManager: SecurityManager: authentication disabled; ui acls disabled; users with view permissions: Set(nayan); users with modify permissions: Set(nayan)
15/12/29 15:27:52 INFO Slf4jLogger: Slf4jLogger started
15/12/29 15:27:52 INFO Remoting: Starting remoting
15/12/29 15:27:53 INFO Remoting: Remoting started; listening on addresses :[akka.tcp://sparkDriver@10.0.1.88:55126]
15/12/29 15:27:53 INFO Utils: Successfully started service 'sparkDriver' on port 55126.
15/12/29 15:27:53 INFO SparkEnv: Registering MapOutputTracker
15/12/29 15:27:53 INFO SparkEnv: Registering BlockManagerMaster
15/12/29 15:27:53 INFO DiskBlockManager: Created local directory at /private/var/folders/pd/6rxlm2js10gg6xys5wm90qpm0000gn/T/blockmgr-21a96671-c33e-498c-83a4-bb5c57edbbfb
15/12/29 15:27:53 INFO MemoryStore: MemoryStore started with capacity 983.1 MB
15/12/29 15:27:53 INFO HttpFileServer: HTTP File server directory is /private/var/folders/pd/6rxlm2js10gg6xys5wm90qpm0000gn/T/spark-fce0a058-9264-4f2c-8220-c32d90f11bd8/httpd-2a0efcac-2426-49c5-982a-941cfbb48c88
15/12/29 15:27:53 INFO HttpServer: Starting HTTP Server
15/12/29 15:27:53 INFO Utils: Successfully started service 'HTTP file server' on port 55127.
15/12/29 15:27:53 INFO SparkEnv: Registering OutputCommitCoordinator
15/12/29 15:27:53 INFO Utils: Successfully started service 'SparkUI' on port 4040.
15/12/29 15:27:53 INFO SparkUI: Started SparkUI at http://10.0.1.88:4040
15/12/29 15:27:54 WARN MetricsSystem: Using default name DAGScheduler for source because spark.app.id is not set.
15/12/29 15:27:54 INFO AppClient$ClientEndpoint: Connecting to master spark://127.0.0.1:7077...
15/12/29 15:27:54 WARN ReliableDeliverySupervisor: Association with remote system [akka.tcp://sparkMaster@127.0.0.1:7077] has failed, address is now gated for [5000] ms. Reason: [Disassociated] 
15/12/29 15:28:14 ERROR SparkUncaughtExceptionHandler: Uncaught exception in thread Thread[appclient-registration-retry-thread,5,main]
java.util.concurrent.RejectedExecutionException: Task java.util.concurrent.FutureTask@1f22aef0 rejected from java.util.concurrent.ThreadPoolExecutor@176cb4af[Running, pool size = 1, active threads = 1, queued tasks = 0, completed tasks = 0]

So something is happening during the creation of the spark context.

When i look in $DSE_HOME/logs/spark, it is empty. Not sure where else to look.

回答1:

It turns out that the problem was the spark library version AND the Scala version. DataStax was running Spark 1.4.1 and Scala 2.10.5, while my eclipse project was using 1.5.2 & 2.11.7 respectively.

Note that BOTH the Spark library and Scala appear to have to match. I tried other combinations, but it only worked when both matched.

回答2:

I am getting pretty familiar with this part of your posted error:

WARN ReliableDeliverySupervisor: Association with remote system [akka.tcp://...

It can have numerous causes, pretty much all related to misconfigured IPs. First I would do whatever zero323 says, then here's my two cents: I have solved my own problems recently by using IP addresses, not hostnames, and the only config I use in a simple standalone cluster is SPARK_MASTER_IP.

SPARK_MASTER_IP in the $SPARK_HOME/conf/spark-env.sh on your master then should lead the master webui to show the IP address you set:

spark://your.ip.address.numbers:7077

And your SparkConf setup can refer to that.

Having said that, I am not familiar with your specific implementation but I notice in the error two occurrences containing:

/private/var/folders/pd/6rxlm2js10gg6xys5wm90qpm0000gn/T/

Have you looked there to see if there's a logs directory? Is that where $DSE_HOME points? Alternatively connect to the driver where it creates it's webui:

INFO SparkUI: Started SparkUI at http://10.0.1.88:4040

and you should see a link to an error log there somewhere.

More on the IP vs. hostname thing, this very old bug is marked as Resolved but I have not figured out what they mean by Resolved, so I just tend toward IP addresses.

来源：https://stackoverflow.com/questions/34518669/unable-to-connect-to-spark-master

标签

scala

apache-spark

datastax