Magpie is a command line tool for deploying and managing the Yarn on Docker cluster.

Build and run an yarn cluster on docker, pass the config item to hadoop configuration files through docker ENV.

Design docs - Yarn on Docker

You should follow the docker project standard to build your own docker image.

• Step1

  Prepare the direcotries for yarn and chown owner to hadoop.

  Download hadoop-2.6.0-cdh5.5.2.tar.gz and install it in docker image. Unzip it and remove the default hadoop configuration files.

• Step2

  Put the codec lib.so files into the hadoop native directory.

  Put custom hadoop configuration files to hadoop conf directory.

• Step3

  Set the ENV and Entrypoint.

• Stop4

  Run a container with sepecific ENV.

Edit Dockerfile and change the base image to your own JDK7 image.

./build.sh
docker build -t hadoop-yarn:v0.1 .

For example

With hadoop ha

docker run -d -e NAMESERVICE=addmp -e ACTIVE_NAMENODE_ID=namenode29 -e STANDBY_NAMENODE_ID=namenode63 -e HA_ZOOKEEPER_QUORUM=192.168.0.1:2181,192.168.0.2:2181,192.168.0.3:2181 -e YARN_ZK_DIR=rmstore -e YARN_CLUSTER_ID= yarnRM -e YARN_RM1_IP=192.168.0.3 -e YARN_RM2_IP=192.168.0.2 -e YARN_JOBHISTORY_IP=192.168.0.3 -e ACTIVE_NAMENODE_IP=192.168.0.1 -e STANDBY_NAMENODE_IP=192.168.0.2  -e HA=yes hadoop-yarn:v0.1 resourcemanager

docker run -d -e NAMESERVICE=addmp -e ACTIVE_NAMENODE_ID=namenode29 -e STANDBY_NAMENODE_ID=namenode63 -e HA_ZOOKEEPER_QUORUM=192.168.0.1:2181,192.168.0.2:2181,192.168.0.3:2181 -e YARN_ZK_DIR=rmstore -e YARN_CLUSTER_ID= yarnRM -e YARN_RM1_IP=192.168.0.3 -e YARN_RM2_IP=192.168.0.2 -e YARN_JOBHISTORY_IP=192.168.0.3 -e ACTIVE_NAMENODE_IP=192.168.0.1 -e STANDBY_NAMENODE_IP=192.168.0.2  -e HA=yes hadoop-yarn:v0.1 nodemanager

Without hadoop ha

docker run -d -e NANENODE_IP=192.168.0.1 -e RESOURCEMANAGER_IP=192.168.0.1 -e YARN_JOBHISTORY_IP=192.168.0.1 -e HA=no hadoop-yarn:v0.1 resourcemanager

docker run -d -e NANENODE_IP=192.168.0.1 -e RESOURCEMANAGER_IP=192.168.0.1 -e YARN_JOBHISTORY_IP=192.168.0.1 -e HA=no hadoop-yarn:v0.1 nodemanager

• HA (default yes)

• NAMESERVICE

• ACTIVE_NAMENODE_IP

• STANDBY_NAMENODE_IP

• ACTIVE_NAMENODE_ID

• STANDBY_NAMENODE_ID

• HA_ZOOKEEPER_QUORUM

• YARN_ZK_DIR

• YARN_CLUSTER_ID

• YARN_RM1_IP

• YARN_RM2_IP

• YARN_JOBHISTORY_IP

• NAMENDOE_IP

• RESOURCEMANAGER_IP

• HISTORYSERVER_IP

• CPU_CORE_NUM

• NODEMANAGER_MEMORY_MB

  ​

• No-password login to all the active resource managers.
• Docker container's name must contain the cluster name.

Use magpie -h for help usage.

Magpie is  a  CLI tool to manage the Yarn on Docker cluster.
 
 Magpie can be used to inspect the docker,swarm and yarn cluster status, scale the yarn cluster and decommising
 nodemanagers or delete the existed containers.
 
 Usage:
   magpie [command]
 
 Available Commands:
   docker      Docker cluster management tool.
   tool        Other management tool.
   yarn        Yarn cluster management tool.
 
 Flags:
       --config string   config file (default is conf/magpie.toml)
   -t, --toggle          Help message for toggle
 
 Use "magpie [command] --help" for more information about a command.

Magpie use viper to resolve the toml configuration file.

Config file default located at ./conf/magpie.toml

You can use --config to sepcify your custom configuration file.

Most configuration items are inherited from docker image, you don't need to specify in the configuration file, unless you want to modify the docker environment variables in the image.

[clusters]
#Yarn clsuter name
cluster_name = ["yarn1","yarn2","yarn3"]
#Swarm master ip address
swarm_master_ip = "192.168.0.1"
swarm_master_port = "4000"
nodemanager_exclude_file = "/usr/local/hadoop/etc/hadoop/exclude.txt"

[resource_managers]
#Yarn active resourcemanager ip address
yarn1 = "172.18.0.2"
yarn2 = "172.18.0.3"
yarn3 = "172.18.0.4"

[base_container]
#Base container used for scale the yarn cluster 
yarn1 = "yarn1-nm1"
yarn2 = "yarn2-nm1"
yarn3 = "yarn3-nm1"

[nodemanager]
#Nodemanager default config
HA = "yes"
NAMESERVICE = "dcnameservice"
ACTIVE_NAMENODE_IP= "namenode-001.test.com"
STANDBY_NAMENODE_IP = "namenode-002.test.com"
ACTIVE_NAMENODE_ID = "namenode1"
STANDBY_NAMENODE_ID = "namenode2"
HA_ZOOKEEPER_QUORUM = "zk-001.test.com:2181,zk-002.test.com:2181,zk-003.test.com:2181"
YARN_ZK_DIR = ""
YARN_CLUSTER_ID = ""
YARN_RM1_IP = ""
YARN_RM2_IP = ""
YARN_JOBHISTORY_IP = ""
NAMENODE_IP = ""
RESOURCEMANAGER_IP = ""
CPU_CORE_NUM = 4
NODEMANAGER_MEMORY_MB = 8192
network_mode = "mynet"
limit_cpus = 5
limit_memory_mb = 12288
image = "docker-registry:5000/library/hadoop-yarn:v0.1"
cmd = ["nodemanager"]

• Inspect the swarm cluster status.
• Inspect the yarn clsuter status.
• Inspect the docker contianer status include contianer's host config and config items.
• Delete docker containers.
• Offline or decomissing yarn nodemanagers.
• Scaling yarn cluster.
• Add a yarn create command to create a new yarn nodemanager for other yarn cluster.

Build environment

go 1.7.4 amd64

Build magpie on your own platform

Edit Makefile to set GOOS and GOARCH to your target environment .

make

• Scale and flux the yarn cluster automatically
• Nodemanager registered on resourcemanager repeatability

Docker remote API version 1.23

YARN RESTful API version 2.6.0

Swarm API version 1.2.2

Author: Jimmy Song

Email: [email protected]

Jimmy Song's Blog

FYI: If you want to create a yarn cluster with multiple nodemanagers, you need a docker plugins to make the docker container on different hosts can be accessed with each others. You need a docker ipam plugin to make the continers located on different hosts can be accessed by each others. Try this:https://github.com/rootsongjc/docker-ipam-plugin You also need a plugin to listen on docker nodes and register container's IP-hostname into a DNS server so that docker containers can recognise each other by the hostname which is the same with the container ID.