Usually on Grid’5000, in order to deploy our own system we use Kadeploy.
The lifetime of the system is very short, the duration of the OAR job.
To keep system modifications, you need to update the deployed Kadeploy environment before the end of OAR job.
This is sufficient in most simple cases, but for systems hosting a lot of datas usable through multiple experiments, it’s not practical.
In this tutorial we will create a persistent system hosting a database, hosted on a node as a virtual machine with KVM, without deployment (only 20 secondes needed to start the service).
This virtual machine will be stored on the Ceph distributed object store available at Rennes and Nantes.
Manage your Ceph object store
Generate your Ceph Key and create an RBD pool
On Grid’5000 Ceph UI (Rennes,Nantes), click on Generate Ceph key (if no key appears) and create a Ceph pool name username_rbd.
Note : A replication size of 3 is required for a safe storage.
Create the virtual machine stored on a Rados Block Device
Configure Ceph client and authentication
From host frontend.rennes.grid5000.fr :
Notes : Replace user_login by your username.
Create a Rados Block Device based on Debian 7 (Wheezy)
Submit an interactive OAR job :
From the node :
Start the virtual machine
From host frontend.rennes.grid5000.fr :
Configure Virtual Machine and MySQL
Set hostname
Here apt-get update command will fail because the system cannot resolve host proxy. SLIRP
does not send by default DNS search field. Add FQDN in /etc/hostname file will quickly solve the problem.
Log in as root/grid5000, and :
Install and configure MySQL
Edit /etc/mysql/my.cnf, comment line bind-address = 127.0.0.1 and restart MySQL service :
Restart VM with MySQL port forwarding
from the node :
Test
From host frontend.rennes.grid5000.fr :
Conclusion
Now you can use a persistent database during multiple experiments.
Network performances are not optimal in this KVM configuration, but it’s very simple to use, without privileges.