My Spark's Worker cannot connect Master.Something wrong with Akka?

Question

I want to install Spark Standlone mode to a Cluster with my two virtual machines.
With the version of spark-0.9.1-bin-hadoop1, I execute spark-shell successfully in each vm. I follow the offical document to make one vm(ip:xx.xx.xx.223) as both Master and Worker and to make the other(ip:xx.xx.xx.224) as Worker only.
But the 224-ip vm cannot connect the 223-ip vm. Followed is 223(Master)'s master log:

[@tc-52-223 logs]# tail -100f spark-root-org.apache.spark.deploy.master.Master-1-tc-52-223.out
Spark Command: /usr/local/jdk/bin/java -cp :/data/test/spark-0.9.1-bin-hadoop1/conf:/data/test/spark-0.9.1-bin-hadoop1/assembly/target/scala-2.10/spark-assembly_2.10-0.9.1-hadoop1.0.4.jar -Dspark.akka.logLifecycleEvents=true -Djava.library.path= -Xms512m -Xmx512m org.apache.spark.deploy.master.Master --ip 10.11.52.223 --port 7077 --webui-port 8080

log4j:WARN No appenders could be found for logger (akka.event.slf4j.Slf4jLogger).
log4j:WARN Please initialize the log4j system properly.
log4j:WARN See http://logging.apache.org/log4j/1.2/faq.html#noconfig for more info.
14/04/14 22:17:03 INFO Master: Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
14/04/14 22:17:03 INFO Master: Starting Spark master at spark://10.11.52.223:7077
14/04/14 22:17:03 INFO MasterWebUI: Started Master web UI at http://tc-52-223:8080
14/04/14 22:17:03 INFO Master: I have been elected leader! New state: ALIVE
14/04/14 22:17:06 INFO Master: Registering worker tc-52-223:20599 with 1 cores, 4.0 GB RAM
14/04/14 22:17:06 INFO Master: Registering worker tc_52_224:21371 with 1 cores, 4.0 GB RAM
14/04/14 22:17:06 INFO RemoteActorRefProvider$RemoteDeadLetterActorRef: Message [org.apache.spark.deploy.DeployMessages$RegisteredWorker] from Actor[akka://sparkMaster/user/Master#1972530850] to Actor[akka://sparkMaster/deadLetters] was not delivered. [1] dead letters encountered. This logging can be turned off or adjusted with configuration settings 'akka.log-dead-letters' and 'akka.log-dead-letters-during-shutdown'.
14/04/14 22:17:26 INFO Master: Registering worker tc_52_224:21371 with 1 cores, 4.0 GB RAM
14/04/14 22:17:26 INFO RemoteActorRefProvider$RemoteDeadLetterActorRef: Message [org.apache.spark.deploy.DeployMessages$RegisterWorkerFailed] from Actor[akka://sparkMaster/user/Master#1972530850] to Actor[akka://sparkMaster/deadLetters] was not delivered. [2] dead letters encountered. This logging can be turned off or adjusted with configuration settings 'akka.log-dead-letters' and 'akka.log-dead-letters-during-shutdown'.
14/04/14 22:17:46 INFO Master: Registering worker tc_52_224:21371 with 1 cores, 4.0 GB RAM
14/04/14 22:17:46 INFO RemoteActorRefProvider$RemoteDeadLetterActorRef: Message [org.apache.spark.deploy.DeployMessages$RegisterWorkerFailed] from Actor[akka://sparkMaster/user/Master#1972530850] to Actor[akka://sparkMaster/deadLetters] was not delivered. [3] dead letters encountered. This logging can be turned off or adjusted with configuration settings 'akka.log-dead-letters' and 'akka.log-dead-letters-during-shutdown'.
14/04/14 22:18:06 INFO Master: akka.tcp://sparkWorker@tc_52_224:21371 got disassociated, removing it.
14/04/14 22:18:06 INFO Master: akka.tcp://sparkWorker@tc_52_224:21371 got disassociated, removing it.
14/04/14 22:18:06 INFO LocalActorRef: Message [akka.remote.transport.ActorTransportAdapter$DisassociateUnderlying] from Actor[akka://sparkMaster/deadLetters] to Actor[akka://sparkMaster/system/transports/akkaprotocolmanager.tcp0/akkaProtocol-tcp%3A%2F%2FsparkMaster%4010.11.52.224%3A61550-1#646150938] was not delivered. [4] dead letters encountered. This logging can be turned off or adjusted with configuration settings 'akka.log-dead-letters' and 'akka.log-dead-letters-during-shutdown'.
14/04/14 22:18:06 INFO Master: akka.tcp://sparkWorker@tc_52_224:21371 got disassociated, removing it.
14/04/14 22:18:06 ERROR EndpointWriter: AssociationError [akka.tcp://[email protected]:7077] -> [akka.tcp://sparkWorker@tc_52_224:21371]: Error [Association failed with [akka.tcp://sparkWorker@tc_52_224:21371]] [
akka.remote.EndpointAssociationException: Association failed with [akka.tcp://sparkWorker@tc_52_224:21371]
Caused by: akka.remote.transport.netty.NettyTransport$$anonfun$associate$1$$anon$2: Connection refused: tc_52_224/10.11.52.224:21371
]
14/04/14 22:18:06 INFO Master: akka.tcp://sparkWorker@tc_52_224:21371 got disassociated, removing it.
14/04/14 22:18:06 ERROR EndpointWriter: AssociationError [akka.tcp://[email protected]:7077] -> [akka.tcp://sparkWorker@tc_52_224:21371]: Error [Association failed with [akka.tcp://sparkWorker@tc_52_224:21371]] [
akka.remote.EndpointAssociationException: Association failed with [akka.tcp://sparkWorker@tc_52_224:21371]
Caused by: akka.remote.transport.netty.NettyTransport$$anonfun$associate$1$$anon$2: Connection refused: tc_52_224/10.11.52.224:21371
]
14/04/14 22:18:06 ERROR EndpointWriter: AssociationError [akka.tcp://[email protected]:7077] -> [akka.tcp://sparkWorker@tc_52_224:21371]: Error [Association failed with [akka.tcp://sparkWorker@tc_52_224:21371]] [
akka.remote.EndpointAssociationException: Association failed with [akka.tcp://sparkWorker@tc_52_224:21371]
Caused by: akka.remote.transport.netty.NettyTransport$$anonfun$associate$1$$anon$2: Connection refused: tc_52_224/10.11.52.224:21371
]
14/04/14 22:18:06 INFO Master: akka.tcp://sparkWorker@tc_52_224:21371 got disassociated, removing it.
14/04/14 22:19:03 WARN Master: Removing worker-20140414221705-tc_52_224-21371 because we got no heartbeat in 60 seconds
14/04/14 22:19:03 INFO Master: Removing worker worker-20140414221705-tc_52_224-21371 on tc_52_224:21371  

Followed is 223(Worker)'s worker log:

14/04/14 22:17:06 INFO Worker: Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
14/04/14 22:17:06 INFO Worker: Starting Spark worker tc-52-223:20599 with 1 cores, 4.0 GB RAM
14/04/14 22:17:06 INFO Worker: Spark home: /data/test/spark-0.9.1-bin-hadoop1
14/04/14 22:17:06 INFO WorkerWebUI: Started Worker web UI at http://tc-52-223:8081
14/04/14 22:17:06 INFO Worker: Connecting to master spark://xx.xx.52.223:7077...
14/04/14 22:17:06 INFO Worker: Successfully registered with master spark://xx.xx.52.223:7077

Followed is 224(Worker)'s work log:

Spark Command: /usr/local/jdk/bin/java -cp :/data/test/spark-0.9.1-bin-hadoop1/conf:/data/test/spark-0.9.1-bin-hadoop1/assembly/target/scala-2.10/spark-assembly_2.10-0.9.1-hadoop1.0.4.jar -Dspark.akka.logLifecycleEvents=true -Djava.library.path= -Xms512m -Xmx512m org.apache.spark.deploy.worker.Worker spark://10.11.52.223:7077 --webui-port 8081
========================================

log4j:WARN No appenders could be found for logger (akka.event.slf4j.Slf4jLogger).
log4j:WARN Please initialize the log4j system properly.
log4j:WARN See http://logging.apache.org/log4j/1.2/faq.html#noconfig for more info.
14/04/14 22:17:06 INFO Worker: Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
14/04/14 22:17:06 INFO Worker: Starting Spark worker tc_52_224:21371 with 1 cores, 4.0 GB RAM
14/04/14 22:17:06 INFO Worker: Spark home: /data/test/spark-0.9.1-bin-hadoop1
14/04/14 22:17:06 INFO WorkerWebUI: Started Worker web UI at http://tc_52_224:8081
14/04/14 22:17:06 INFO Worker: Connecting to master spark://xx.xx.52.223:7077...
14/04/14 22:17:26 INFO Worker: Connecting to master spark://xx.xx.52.223:7077...
14/04/14 22:17:46 INFO Worker: Connecting to master spark://xx.xx.52.223:7077...
14/04/14 22:18:06 ERROR Worker: All masters are unresponsive! Giving up.

Followed is my spark-env.sh:

JAVA_HOME=/usr/local/jdk
export SPARK_MASTER_IP=tc-52-223
export SPARK_WORKER_CORES=1
export SPARK_WORKER_INSTANCES=1
export SPARK_MASTER_PORT=7077
export SPARK_WORKER_MEMORY=4g
export MASTER=spark://${SPARK_MASTER_IP}:${SPARK_MASTER_PORT}
export SPARK_LOCAL_IP=tc-52-223

I have googled many solutions, but they cant work. Please help me.

Answer 1

I'm not sure if this is the same issue I encountered but you may want to try setting SPARK_MASTER_IP the same as what spark binds to. In your example is looks like it would be 10.11.52.223 and not tc-52-223.

It should be the same as what you see when you visit the master node web UI on 8080. Something like: Spark Master at spark://ec2-XX-XX-XXX-XXX.compute-1.amazonaws.com:7077

Answer 2

If you are getting a "Connection refused" exception, You can resolve it by checking

=> Master is running on the specific host

netstat -at | grep 7077

You will get something similar to:

tcp        0      0 akhldz.master.io:7077 *:*             LISTEN  

If that is the case, then from your worker machine do a host akhldz.master.io ( replace akhldz.master.io with your master host.If something goes wrong, then add a host entry in your /etc/hosts file)
telnet akhldz.master.io 7077 ( If this is not connecting, then your worker wont connect either. )

=> Adding Host entry in /etc/hosts

Open /etc/hosts from your worker machine and add the following entry (example)

192.168.100.20   akhldz.master.io

PS :In the above case Pillis was having two ip addresses having same host name eg:

192.168.100.40  s1.machine.org
192.168.100.41  s1.machine.org

Hope that help.

Answer 3

There's a lot of answers and possible solutions, and this question is a bit old, but in the interest of completeness, there is a known Spark bug about hostnames resolving to IP addresses. I am not presenting this as the complete answer in all cases, but I suggest trying with a baseline of just using all IPs, and only use the single config SPARK_MASTER_IP. With just those two practices I get my clusters to work and all the other configs, or using hostnames, just seems to muck things up.

So in your spark-env.sh get rid of SPARK_WORKER_IP and change SPARK_MASTER_IP to an IP address, not a hostname.

I have treated this more at length in this answer.

For more completeness here's part of that answer:

Can you ping the box where the Spark master is running? Can you ping the worker from the master? More importantly, can you password-less ssh to the worker from the master box? Per 1.5.2 docs you need to be able to do that with a private key AND have the worker entered in the conf/slaves file. I copied the relevant paragraph at the end.

You can get a situation where the worker can contact the master but the master can't get back to the worker so it looks like no connection is being made. Check both directions.

I think the slaves file on the master node, and the password-less ssh can lead to similar errors to what you are seeing.

Per the answer I crosslinked, there's also an old bug but it's not clear how that bug was resolved.