Glusterfs replica + arbiter

TAGS: gluster linux

Glusterfs replica 2+1

Intro

Today we will discuss about glusterfs on two storage nodes with one management node aka arbiter scheme.
If you have only two cool hardware nodes and you want some nice HA in your storage, then arbiter scheme is what you need.
In case of failure this scheme is working like this: 1) When one of your storage node goes down then no one will suspect something.
2) Same behaviour if arbiter goes down.
3) If arbiter node and storage node are down, then your gluster cluster becomes readonly.
4) With two storage nodes in down state your transport point will be unavailable.
Also with this arbiter scheme you will always have quorum in write mode so you can be sure that your data is consistent.

How to deploy

What we have:

  • 3 ubuntu 16.04 nodes cat /etc/hosts storage-1 10.1.39.100 storage-2 10.1.39.101 meta-1 10.1.39.102

Run on all nodes
sudo passwd root add-apt-repository ppa:gluster/glusterfs-3.12 apt update apt install glusterfs-server gluster peer probe storage-{1..2}; gluster peer probe meta-1 mkdir /mnt/gluster-brick mkdir /opt/gluster-data gluster volume create data replica 3 arbiter 1 transport tcp storage-1:/mnt/gluster-brick storage-2:/mnt/gluster-brick meta-1:/mnt/gluster-brick gluster volume start data gluster volume info

client
mount -t glusterfs storage-1/data MOUNTDIR

Written on October 16, 2017

Archive

linux openstack kubernetes gluster ceph ansible docker systemctl python openshift log centos sed registry kolla kibana keepalived elasticsearch deploy cloud-init auth HA zabbix vsphere vmware venv tools swift ssl ssh scm ruby rsyslog rhel rbac rabbitmq prometheus postgres policy pgpool2 patrole pacemaker ntp nfs net mq monitoring mongo mbr lvm logstash kubespray kubeadm kolla-ansible jenkins jekyII java ironic helm heketi harbor grub github ganesha fluentd fluent-bit fio drbd diskimage-builder devstack debug deb db cisco cicd ceph-ansible calico build bitbucket benchmarking apparmor api NUMA Linux Golang Go ELK 3par