This page explains how to setup Gluster network filesystems that can be mounted by multiple virtual machines, in particular, MPI(or GNU/parallel) cluster nodes.
If you have questions and suggestions about this tutorial page and related problems, please, contact Min-Su Shin.

Case A: Filesystem with a single Gluster server

Gluster filesystem can be prepared with a single server and its disk space to serve multiple clients. In this case A, we assume that a single Gluster server is made to be used by multiple VMs (virtual machines) included in a cluster.

Step 1. Creating a single VM

Following KASI Science Cloud : VM instances#Step3.CreateaVMinstance, create a single VM with a VM flavor which fits to your requirement such as the required size of the network filesystem. The created single VM will work as a server serving the Gluster filesystem. Since the required memory for the Gluster server is not huge, the recommended flavor is large-CPU + small-memory for a given size of disk.

Step 2. Configuring the Gluster server

On the created VMs, we install the Gluster server program and configure a Gluster filesystem. For your convenience, you can find a shell script that should be executed by root account on the master node. See https://github.com/astromsshin/cloud_ex and the script https://github.com/astromsshin/cloud_ex/blob/main/tool_setup_glusterfs_single_server.sh which is presented below:

#!/bin/bash

# Directory serving the Gluster filesystem
TARGETDIR="/glusterfs/vol"
# Gluster filesystem volume available for clients
TARGETVOL="gvol"
# Name of the Gluster server
VMNAME="test-gl-vm"

# Install the server
apt install glusterfs-server -y
systemctl enable --now glusterd

# Information
# (optional)
#gluster --version
#systemctl status glusterd

# Prepare the directory
mkdir -p ${TARGETDIR}
# Produce the Gluster filesystem volume with the above directory
gluster volume create ${TARGETVOL} ${VMNAME}:${TARGETDIR} force
# Make the volume available
gluster volume start ${TARGETVOL}

# Information
# (optional)
#gluster volume info
#gluster volume status

Step 3. Mounting the created Gluster filesystemt on client VMs

The created network filesystem can be mounted on client VMs by using mount command with required options. A simple script is available as https://github.com/astromsshin/cloud_ex/blob/main/tool_setup_glusterfs_client_all_nodes-single_server.sh in the github repository https://github.com/astromsshin/cloud_ex. The script, which needs to be executed on the client cluster master node, is given below:

#!/bin/bash

# Name of the cluster which defines the clients' names
CLUSTERNAME="ml-image"
# Last integer index of the minions in the cluster
MINIONLASTIND="8"
# IP of the Gluster server
GLUSTERSERVERIP="10.0.100.150"
# Name of the Gluster server
GLUSTERSERVERNAME="test-gl-vm"
# Directory name which is a mount point on clients
TARGETDIR="/mnt/gluster"


RUNCMD="apt -y install glusterfs-client; echo ${GLUSTERSERVERIP} ${GLUSTERSERVERNAME} >> /etc/hosts;mkdir ${TARGETDIR}; mount -t glusterfs ${GLUSTERSERVERNAME}:/gvol ${TARGETDIR}; chmod -R 777 ${TARGETDIR}"

echo "... setuping on ${CLUSTERNAME}-master"
echo $RUNCMD | bash

for ind in $(seq 0 ${MINIONLASTIND})
do
echo "... setuping on ${CLUSTERNAME}-minion-${ind}"
ssh ${CLUSTERNAME}-minion-${ind} "${RUNCMD}"
done

In the above script, we assume that the filesystem is mounted on multiple VMs created as explained in Deploy Message Passing Interface (MPI) (and/or GNU parallel) cluster. Edit the variables CLUSTERNAME, MINIONLASTIND, GLUSTERSERVERIP, GLUSTERSERVERNAME, and TARGETDIR for your cluster configuration.

Case B: Filesystem with multiple Gluster servers (in the form of distributed volumes)

We explain how to setup the Gluster filesystem in the form of distributed volumes by using multiple Gluster servers. As depicted in the following figure and explained in https://docs.gluster.org/en/latest/Administrator-Guide/Setting-Up-Volumes/,

there is a type of distrbiuted volume where files are distributed in multiple gluster servers.

Step 1. Creating a basic cluster of VMs

You need to deploy multiple VMs as Gluster servers. The basic procedure follows the tutorial given in Deploy Message Passing Interface (MPI) (and/or GNU parallel) cluster. However, you do not need to setup MPI and NFS share directories explained in the tutorial. Use a cluster template KASI-Cluster-Basic-Block in the KASI Cluster Templates to produce a VM cluster for the Gluster filesystem with an extra block device (i.e., /dev/vdb). The recommended configuration of the flavor is C2M8D40, and you need to estimate the required number of minions for the required Gluster volume size and the chosen extra block size. For example, if you need 30TB volume, you need 9 minions as well as 1 master (i.e., 10 virtual machines in total) with the block size 3TB (i.e., 3072GB).

Step 2. Configuring the Gluster server with multiple bricks on the cluster VMs

The configuration of the Gluster VMs can be done with the provided script https://github.com/astromsshin/cloud_ex/blob/main/tool_setup_glusterfs_multiple_servers.sh on the github repository https://github.com/astromsshin/cloud_ex. The script is given below. You should edit the CLUSTERNAME and MINIONLASTIND in the below script for your own configuration of the Gluster cluster.

#!/bin/bash

# Run this script with root permission
# on the master node of the GlusterFS cluster

# Block device
TARGETDEV="/dev/vdb"
# Directory serving the Gluster filesystem
TARGETDIR="/glusterfs/vol"
# Gluster filesystem volume available for clients
TARGETVOL="gvol"
# Name of the Gluster cluster
CLUSTERNAME="gluster-test"
MINIONLASTIND="8"
# Saving Gluster server information
HOSTNAMEFN="gluster_server_hostnames.txt"
HOSTIPFN="gluster_server_ips.txt"

# Install the server
echo "... install required software on ${CLUSTERNAME}-master"
apt install glusterfs-server -y
systemctl enable --now glusterd
for ind in $(seq 0 ${MINIONLASTIND})
do
echo "... install required software on ${CLUSTERNAME}-minion-${ind}"
ssh ${CLUSTERNAME}-minion-${ind} "apt install glusterfs-server -y; systemctl enable --now glusterd"
done

echo
read -p "[CHECK] Are softwares installed successfully on all nodes? [YyNn]" -n 1 -r
echo
if [[ ! $REPLY =~ ^[Yy]$ ]]
then
echo "[STOP] please, re-run the script to make sure that all nodes have the softwares."
exit 1
fi

# Probing
for ind in $(seq 0 ${MINIONLASTIND})
do
echo "... probing ${CLUSTERNAME}-minion-${ind}"
gluster peer probe ${CLUSTERNAME}-minion-${ind}
done

# Prepare the block device with the XFS filesystem
echo "... preparing the block dev ${TARGETDEV} on ${CLUSTERNAME}-master"
umount -f ${TARGETDEV}
mkfs.xfs -f -i size=512 ${TARGETDEV}
for ind in $(seq 0 ${MINIONLASTIND})
do
echo "... preparing the block dev ${TARGETDEV} on ${CLUSTERNAME}-minion-${ind}"
ssh ${CLUSTERNAME}-minion-${ind} "umount -f ${TARGETDEV}; mkfs.xfs -f -i size=512 ${TARGETDEV}"
done

# Prepare the directory
echo "... creating directory on ${CLUSTERNAME}-master"
mkdir -p ${TARGETDIR}
for ind in $(seq 0 ${MINIONLASTIND})
do
echo "... creating directory on ${CLUSTERNAME}-minion-${ind}"
ssh ${CLUSTERNAME}-minion-${ind} "mkdir -p ${TARGETDIR}"
done

# Mount the block device
echo "... mounting ${TARGETDEV} on ${CLUSTERNAME}-master"
mount -t xfs ${TARGETDEV} ${TARGETDIR}
for ind in $(seq 0 ${MINIONLASTIND})
do
echo "... mounting ${TARGETDEV} on ${CLUSTERNAME}-minion-${ind}"
ssh ${CLUSTERNAME}-minion-${ind} "mount -t xfs ${TARGETDEV} ${TARGETDIR}"
done

# Information
gluster --version
gluster peer status

# Produce the Gluster filesystem volume with the above directory
VOLSTR="${CLUSTERNAME}-master:${TARGETDIR}"
for ind in $(seq 0 ${MINIONLASTIND})
do
VOLSTR="${VOLSTR} ${CLUSTERNAME}-minion-${ind}:${TARGETDIR}"
done
gluster volume create ${TARGETVOL} ${VOLSTR} force
# Make the volume available
gluster volume start ${TARGETVOL}

# Information
# (optional)
gluster volume info
gluster volume status

echo "Use ${HOSTNAMEFN} for clients"
gluster volume info | grep "^Brick[0-9]*:" | grep -v 'Bricks:' | awk -F':' '{print $2}' | xargs | tee ${HOSTNAMEFN}
echo "Use ${HOSTIPFN} for clients"
gluster volume info | grep "^Brick[0-9]*:" | grep -v 'Bricks:' | awk -F':' '{print $2}' | xargs -I {} grep -m 1 "{}" /etc/hosts | cut -d' ' -f1 | xargs | tee ${HOSTIPFN}

Step 3. Mounting the created Gluster filesystem on client VMs

The configuration of the Gluster firesystem affects how the client VMs mount and access the Gluster filesystem. The following script, which is available on the github repository https://github.com/astromsshin/cloud_ex as https://raw.githubusercontent.com/astromsshin/cloud_ex/main/tool_setup_glusterfs_client_all_nodes-multiple_servers.sh, should be executed on the master node of your cluster, i.e., client VMs. Edit the variables CLUSTERNAME, MINIONLASTIND, and TARGETDIR depending on your cluster configuration and usage directory name.

#!/bin/bash

# Name of the cluster which defines the clients' names
CLUSTERNAME="lice"
# Last integer index of the minions in the cluster
MINIONLASTIND="14"
# Directory name which is a mount point on clients
TARGETDIR="/mnt/gluster-input"

# IP of the Gluster server
GLUSTERSERVERIP="10.0.110.167 10.0.110.92 10.0.110.124 10.0.110.68 10.0.110.104 10.0.110.172 10.0.110.44 10.0.110.147 10.0.110.67 10.0.110.185 10.0.110.31 10.0.110.225"
# Name of the Gluster server
GLUSTERSERVERNAME="gluster-input-master gluster-input-minion-0 gluster-input-minion-1 gluster-input-minion-2 gluster-input-minion-3 gluster-input-minion-4 gluster-input-minion-5 gluster-input-minion-6 gluster-input-minion-7 gluster-input-minion-8 gluster-input-minion-9 gluster-input-minion-10"

IPARRAY=($GLUSTERSERVERIP)
NAMEARRAY=($GLUSTERSERVERNAME)

# Deleting exisitng entries in /etc/hosts
echo "Deleting existing entries in /etc/hosts"
# ... Master
for (( i=0; i<=${#IPARRAY[@]}; i++ ))
do
sed -i "/ ${NAMEARRAY[$i]}$/d" /etc/hosts
sed -i "/ ${IPARRAY[$i]}$/d" /etc/hosts
done
# ... Minion
for ind in $(seq 0 ${MINIONLASTIND})
do
for (( i=0; i<=${#IPARRAY[@]}; i++ ))
do
ssh ${CLUSTERNAME}-minion-${ind} "sed -i \"/${NAMEARRAY[$i]}/d\" /etc/hosts; sed -i \"/${IPARRAY[$i]}/d\" /etc/hosts"
done
done


# Updating /etc/hosts
echo "Adding entries in /etc/hosts"
# ... Master
for (( i=0; i<=${#IPARRAY[@]}; i++ ))
do
echo ${IPARRAY[$i]} ${NAMEARRAY[$i]} >> /etc/hosts
done
# ... Minion
for ind in $(seq 0 ${MINIONLASTIND})
do
for (( i=0; i<=${#IPARRAY[@]}; i++ ))
do
ssh ${CLUSTERNAME}-minion-${ind} "echo ${IPARRAY[$i]} ${NAMEARRAY[$i]} >> /etc/hosts"
done
done


# Rest of tasks 
echo "Running the relevant commands"
RUNCMD="apt -y install glusterfs-client; mkdir ${TARGETDIR}; mount -t glusterfs ${NAMEARRAY[0]}:/gvol ${TARGETDIR}; chmod -R 777 ${TARGETDIR}"
# ... Master
echo "... setuping on ${CLUSTERNAME}-master"
echo $RUNCMD | bash
# ... Minion
for ind in $(seq 0 ${MINIONLASTIND})
do
echo "... setuping on ${CLUSTERNAME}-minion-${ind}"
ssh ${CLUSTERNAME}-minion-${ind} "${RUNCMD}"
done

Since there are multiple Gluster servers, you need to type their ip addresses and names in the above script. Edit the variables GLUSTERSERVERIP and GLUSTERSERVERNAME in the above script. You can easily find the ip addresses and names in the cloud dashboard or the files generated in the step of configuring the Gluster cluster.

Useful Tips

When you need to unmount the Gluster filesystems on the client machines, you can use the https://github.com/astromsshin/cloud_ex/blob/main/tool_umount_fs_all_nodes.sh script.
There are several Gluster-related commands which help you figure out the current status. See https://docs.gluster.org/en/latest/CLI-Reference/cli-main/ . You have to login the Gluster cluster machines in order to execute the Gluster commands. Check https://data.kasi.re.kr/confluence/x/PoDVAg about how to access virtual machines.