Storm-yarn enables Storm clusters to be deployed into machines managed by Hadoop YARN.
Storm-yarn enables Storm clusters to be deployed into machines managed by Hadoop YARN. It is still a work in progress.
Feel free to ask questions on storm-yarn's mailing list: http://groups.google.com/group/storm-yarn
Based on the project developed by yahoo, we have added following new features.
We have updated the version of Apache Storm from 0.9.0 to 1.0.1.
We have added StormClusterChecker class, in order to monitor the storm cluster. It can adjust the number of supervisors based on the usage of system resources.
We have added the function, namely removeSupervisors() in order to monitor resources. Its functionality is just opposite to that of addSupervisors().
We have updated the logging framework from logback to log4j2.
Install Java 8 and Maven 3 first. These two software are necessary to compiling and packaging the source code of storm-on-yarn.
Make sure Hadoop YARN have been properly launched.
The storm-on-yarn implementation does not include running Zookeeper on YARN. Make sure the Zookeeper service is independently launched beforehands.
Download the source code of storm-on-yarn, e.g., execute the command
git cloneto get the source code.
Edit pom.xml in storm-on-yarn root directory to set the Hadoop version.
To package items, please execute the following command under storm-on-yarn root directory.
You will see the following execution messages.
17:57:27.810 [main] INFO com.yahoo.storm.yarn.TestIntegration - bin/storm-yarn launch ./conf/storm.yaml --stormZip lib/storm.zip --appname storm-on-yarn-test --output target/appId.txt 17:57:59.681 [main] INFO com.yahoo.storm.yarn.TestIntegration - bin/storm-yarn getStormConfig ./conf/storm.yaml --appId application13721218423690001 --output ./lib/storm/storm.yaml 17:58:04.382 [main] INFO com.yahoo.storm.yarn.TestIntegration - ./lib/storm/bin/storm jar lib/storm-starter-0.0.1-SNAPSHOT.jar storm.starter.ExclamationTopology exclamation-topology 17:58:04.382 [main] INFO com.yahoo.storm.yarn.TestIntegration - ./lib/storm/bin/storm kill exclamation-topology 17:58:07.798 [main] INFO com.yahoo.storm.yarn.TestIntegration - bin/storm-yarn stopNimbus ./conf/storm.yaml --appId application13721218423690001 17:58:10.131 [main] INFO com.yahoo.storm.yarn.TestIntegration - bin/storm-yarn startNimbus ./conf/storm.yaml --appId application13721218423690001 17:58:12.460 [main] INFO com.yahoo.storm.yarn.TestIntegration - bin/storm-yarn stopUI ./conf/storm.yaml --appId application13721218423690001 17:58:15.045 [main] INFO com.yahoo.storm.yarn.TestIntegration - bin/storm-yarn startUI ./conf/storm.yaml --appId application13721218423690001 17:58:17.390 [main] INFO com.yahoo.storm.yarn.TestIntegration - bin/storm-yarn shutdown ./conf/storm.yaml --appId application13721218423690001
If you want to skip the tests, please add
mvn package -DskipTests
After compiling and building the whole project of storm-on-yarn, next you need to install storm-on-yarn and Storm on the Storm Client machine, which is used for submitting the YARN applications to YARN ResourceManager (RM) later.
Please refer to the following guide, step by step to deploy on the Storm Client machine.
So far, you have put storm-on-yarn and storm in the right place on Storm Client machine. You do not need to start running the Storm cluster, as this will be done by running storm-on-yarn later on.
Hence you are suggested to add storm-1.0.1/bin and $(storm-on-yarn root directory)/bin to your PATH environment variable, like this:
Following commands illustrate how to upload the storm.zip from the local directory to "/lib/storm/1.0.1" in HDFS.
hadoop fs -mkdir /lib hadoop fs -mkdir /lib/storm hadoop fs -mkdir /lib/storm/1.0.1 zip -r storm.zip storm-1.0.1 hadoop fs -put storm.zip /lib/storm/1.0.1/
At this point, the deployment step is complete.
Everything should be ready. Now you can start your storm-on-yarn project.
The storm-on-yarn project have a set of specify commands and it use storm-yarn [command] -[arg] xxx as the comman format.
To launch the cluster you can run
storm-yarn launch [storm-yarn-configuration]
[storm-yarn-configuration], which is usually a .yaml file, including all the required configurations during the launch of the Storm cluster.
In this project, we provide two quick ways to create the storm-yarn-configuration file:
a) Edit the storm.yaml file under storm-1.0.1/conf folder
b) Copy the $(storm-on-yarn root directory)/src/main/master_defaults.yaml to storm-1.0.1/conf and rename it to master.yaml, and then edit it where necessary.
In the simplest case, the only configuration you need to add is the Zookeeper cluster information and set the port of supervisor:
When you write a right configure file and execute the command above, the Hadoop YARN will return a application ID, if you deploy completely right. like this:
And, you can see the graphic below on YARN UI and Storm UI respectively.
So far, the storm-on-yarn has run and you can submit topology.
You can run the following command to submit a topology.
storm jar ***.jar topologyName -c nimbus=
For example :
storm jar ~/WordCount2/testWord-1.0-SNAPSHOT.jar storm.topology.WordCountTopology wordcount -c nimbus=192.168.1.25
storm-yarn has a number of new options for configuring the storm ApplicationManager (AM), e.g.,
The procedure of "storm-yarn launch" returns an Application ID, which uniquely identifies the newly launched Storm master. This Application ID will be used for accessing the Storm master.
To obtain a storm.yaml from the newly launch Storm master, you can run
storm-yarn getStormConfig --appId --output
storm.yaml will be retrieved from Storm master.
For a full list of storm-yarn commands and options you can run
Storm-on-yarn is now configured to use Netty for communication between spouts and bolts.
It's pure JVM based, and thus OS independent.
If you are running storm using zeromq (instead of Netty), you need to augment the standard storm.zip file the needed .so files. This can be done with the not ideally named create-tarball.sh script
| Command | Usage | | -------- | ------ | |storm-yarn launch | launch storm on yarn| |storm-yarn help | get help| |storm-yarn version | view storm version| |storm-yarn addSupervisors/removeSupervisor | add/remove supervisor| |storm-yarn startNimbus/stopNimbus | start/stop Nimbus| |storm-yarn startUI/stopUI | start/stop Web UI| |storm-yarn startSupervisors/stopSupervisor | start or stop all supervisors| |storm-yarn shutdown | shutdown storm cluster|
| Argument | Usage| | -------- | ------ | | -appname | (Only for storm-yarn launch) Application Name. Default value – "Storm-on-Yarn" | | -appId | (Required) The storm clusters app ID| | -output | Output file| | -container | (Required for removeSupervisors) the supervisor to be removed| | -supervisors | (Required for addSupervisors) The # of supervisors to be added|
There is no failover when nimbus goes down. Still working on it.
There is no simple way to get to the logs for the different processes.
The use and distribution terms for this software are covered by the Apache License, Version 2.0 (http://www.apache.org/licenses/LICENSE-2.0.html).
This work was partially supported by SingAREN/AWS Cloud Credit for Research Program 2016.