ale93p / g5k-flink-cluster

Automated deployment of an Apache Flink cluster in your Grid'5000 reserved nodes.

Grid’5000 Apache Flink Cluster

This script will deploy a basic Apache Flink Cluster in our reserved nodes in Grid’5000. Hope to improve it in the future, any help is welcomed.

Dependencies

• Python 3.x
• Ansible (tested with version 2.5)

To install ansible from the frontend:

export PATH=$HOME/.local/bin:$PATH
easy_install --user ansible netaddr

How to run it

First of all we have to clone this repository in the frontend node in Grid’5000.

Download the disk image and the env file from here, then move them inside the repository folder.

Reserve resources

To reserve nodes in Grid’5000 you just have to run the following command (adapted to your situation):

frontend > oarsub -t deploy -p "cluster='suno'" -I -l nodes=4,walltime=2 -k

In this example we are in the Sophia region, we’re requesting 4 nodes for 2 hours in the cluster named ”suno”.

For further and more specific information follow the Grid’5000’s Getting Started tutorial.

Pre-tasks

Prepare the config file

Open the file cluster.conf and modify the parameters to comply your system configuration (g5k, storm and folders). Be sure to change the username with your grid5000 username, and to specify the correct grid5000 image name. (check also other possible configurations that I may have changed during testing)

Install Storm in your frontend

Download ad extract a binary of flink (download 1.7.2) in your frontend home. It will be needed to copy the configuration for your cluster, so you will be able to submit topologies directly from the frontend.

Run it

frontend > python3 deploy.py

To access your nodes use:

ssh root@node-name

That’s all. Simple, no?

Post-Run

Connect to Flink Dashboard

To connect to the Web UI, we need to open an ssh tunnel to the web service port:

localhost > ssh {{ g5k.username }}@access.grid5000.fr -N -L8081:{{ nimbus_node_address }}:8081

Now the Web Server should be reached through localhost:8081

Multi-Cluster Run

The script is able to deploy storm also in a multi-cluster environment. To make the reservation use:

frontend > oargridsub -t deploy -w '0:59:00' suno:rdef="/nodes=6",parapide:rdef="/nodes=6"

In this case, we don't enter in the job shell, so we don't have the OAR_NODE_FILE systemvariable. We can retrieve the list of the reserved machines using:

frontend > oargridstat -w -l {{ GRID_RESERVATION_ID  }} | sed '/^$/d' > ~/machines

Finally, change the configuration file specifying the location of the file just created (oar.file.location=~/machines) and write "yes" in the multi cluster option (multi.cluster=yes).

For more informations visit the Grid'5000's Multi-site jobs tutorial.

More

Check out also the other script for:

• Apache Storm