GlusterFS: How-to Install and Overview

UPDATE: Note that GluserFS Cinder support was deprecated in OpenStack Newton & will not be supported on Platform9 as of 3.6 release.

Traditional hardware-based storage has two issues: scalability and centralized storage. The storage arrays scale vertically so storage capacity cannot be increased on-demand, and once the array is filled, another array must be bought. Centralized storage requires specialized hardware, including arrays and in some cases, specialized storage networks using Fiber Channel. Information becomes expensive and is difficult to locate and manage. This is why cloud storage is so attractive to organizations, but cloud storage requires the ability to grow or shrink storage capacity on demand. Gluster File System allows you to scale out large, unstructured data on premise, in Linux containers, or in public, private, or hybrid clouds. This blog shows how-to install GlusterFS and gives an overview.

GlusterFS connects storage servers over TCP/IP or Infiniband Remote Direct Memory Access(RDMA), clustering disk and memory together. Data management is centralized under a global namespace so that data is accessible throughout multiple data storage locations. It supports multi-tenancy by dividing users and groups into logical volumes on shared storage, which eliminates user dependency on traditional storage arrays. Capacity may be added throughout a variety of workloads without affecting performance, thus increasing storage efficiency and capacity.

Gluster Deployment Considerations

One thing to watch out for before you start building your Gluster volume is split brain, or in other words, when a cluster of nodes gets partitioned so that each divided set of nodes acts independently. Split brain is specific to GlusterFS with replication. The most effective way of preventing split brain is to ensure you have the ability to tolerate multiple simultaneous failures in each volume. In GlusterFS with replication, this means that you’ll lose 67% of your data (or more) to redundancy. This is where Arbiter volumes come in handy. An Arbiter volume is a mechanism to define a smaller “Voting” disk that doesn’t store any real data but does “Vote” on which “side” of the replica pair has the most consistent state of data. Note that Arbiter volumes do nothing to increase redundancy.

Which leads into our next consideration, failures to tolerate. When it comes to software defined storage, remember that the data is being hosted by servers. Servers need maintenance windows from time to time. If you only tolerate one failure, you will be without any redundancy to speak of during a standard maintenance window. This is why most Software Defined Storage recommend (or require) the ability to tolerate two simultaneous failures.

GlusterFS allows for multiple methods to protect your data. These methods are Replication and Erasure Coding. Each of these methods have their pros and cons, and we’ll be diving deeper into each of these.

Getting started with Gluster

Assumptions

CentOS 7 Installed
6x Disks(/dev/sdb-/dev/sdg)
Server names: Gluster01-Gluster06
Logged in as root

Below are the steps for a successful GlusterFS deployment, followed by details for each step:

Install GlusterFS
Format and mount the disks that will be used for GlusterFS bricks
Start the GlusterFS service
Probe your hosts to form a cluster
Choose your data protection method and Create your volume
Expand your volume
Start your volume
Mount your volume
Monitor your filesystem as well as hosts

1) Install GlusterFS

Add the upstream and gluster repos

$ yum -q -y install centos-release-gluster epel-release

Install GlusterFS and dependencies

$ yum -q -y install glusterfs-server glusterfs-fuse attr xfsprogs xfsdump

2) Prepare bricks

The first thing we’ll do is prepare our raw disks to be added into a GlusterFS volume.

Create folders to mount bricks

$ mkdir -p /bricks/brick{1..6}

Format the volumes

$ mkfs.xfs -f /dev/sdb

mkfs.xfs -f /dev/sdc

mkfs.xfs -f /dev/sdd

mkfs.xfs -f /dev/sde

mkfs.xfs -f /dev/sdf

mkfs.xfs -f /dev/sdg

Add your mount points to fstab for persistence

$ cat <<-EOF >> /etc/fstab

/dev/sdb  /bricks/brick1    xfs    defaults    0 0

/dev/sdv  /bricks/brick2    xfs    defaults    0 0

/dev/sdd  /bricks/brick3    xfs    defaults    0 0

/dev/sdr  /bricks/brick4    xfs    defaults    0 0

/dev/sdf  /bricks/brick5    xfs    defaults    0 0

/dev/sdg  /bricks/brick6    xfs    defaults    0 0

EOF

Mount all of the bricks

$ mount -a

Create an empty folder on each brick for the gluster data to live in

$ mkdir -p /bricks/brick{1..6}/brick

3) Start and enable the GlusterFS service

Enable Glusterd

$ systemctl enable glusterd.service

Start Glusterd

$ systemctl start glusterd.service

4) Probe your hosts to form a cluster

This will loop through and probe hosts 2-6 to add them to a cluster

Only run this from the gluster01 host:

$ for i in {2..6}; do gluster peer probe gluster0$i; done<

5) Choose your data protection method

Replication

GlusterFS with replication provides simple redundancy over multiple storage nodes by creating multiple copies of data throughout other storage locations. So, in the case of disk failure, data availability is unaffected.

Pros
- Less CPU intensive = faster write performance
- Simple restore = faster rebuild performance
Cons
- Requires 2x or more of the original storage space
- Susceptible to Split-Brain

Here are the steps to create a Gluster volume using replication:

Let’s say you are starting off with 36 disks total. If we wanted tolerate two simultaneous failures without data loss, we would use 12 disks as data disks, and 24 as redundant disks.

Create the volume using the following command. *The order of bricks matters because you are creating replica pairs. For example, Gluster1:Brick1 pairs with Gluster2:Brick1 and Gluster3:Brick1.

$ gluster volume create pf9 replica 3 transport tcp \

gluster01:/bricks/brick1/brick gluster02:/bricks/brick1/brick gluster03:/bricks/brick1/brick \

gluster04:/bricks/brick1/brick gluster05:/bricks/brick1/brick gluster06:/bricks/brick1/brick \

gluster01:/bricks/brick2/brick gluster02:/bricks/brick2/brick gluster03:/bricks/brick2/brick \

gluster04:/bricks/brick2/brick gluster05:/bricks/brick2/brick gluster06:/bricks/brick2/brick \

gluster01:/bricks/brick3/brick gluster02:/bricks/brick3/brick gluster03:/bricks/brick3/brick \

gluster04:/bricks/brick3/brick gluster05:/bricks/brick3/brick gluster06:/bricks/brick3/brick \

gluster01:/bricks/brick4/brick gluster02:/bricks/brick4/brick gluster03:/bricks/brick4/brick \

gluster04:/bricks/brick4/brick gluster05:/bricks/brick4/brick gluster06:/bricks/brick4/brick \

gluster01:/bricks/brick5/brick gluster02:/bricks/brick5/brick gluster03:/bricks/brick5/brick \

gluster04:/bricks/brick5/brick gluster05:/bricks/brick5/brick gluster06:/bricks/brick5/brick \

gluster01:/bricks/brick6/brick gluster02:/bricks/brick6/brick gluster03:/bricks/brick6/brick \

gluster04:/bricks/brick6/brick gluster05:/bricks/brick6/brick gluster06:/bricks/brick6/brick

Erasure Coding

Erasure coding is a parity based protection method where data is broken up into fragments, encoded, and stored across multiple storage locations with some fragments being redundant. Erasure coding enables corrupted data in the disk storage process to be reconstructed by using data stored elsewhere in the array. In the case that node partitioning occurs, data can be evaluated as good or bad by having a quorum and at least 2 redundancy bricks. In comparison to traditional RAID methods, erasure coding is faster for data reconstruction, but is CPU-intensive, as well as having a penalty on writes in order to calculate the parity checksum.

Pros
- Consumes less storage than replication – good for cheap/deep
- Allows for the failure of one or more elements of the storage system
Cons
- Parity calculation is CPU-intensive
- Rebuild times are slower and are CPU-intensive
- Increased latency can slow production writes and rebuilds
  - 10Gbps Nics and SSDs can alleviate most of this.

Here are the following steps to erasure coding:

Let’s say you are starting off with 36 disks total. If we wanted tolerate two simultaneous failures without data loss, we would have 24 disks used as data disks, and 12 as redundant disks.

In order to create a volume using erasure coding, do the following: *The order of bricks matters because you are creating dispersed pairs. For example, Gluster1:Brick1 pairs with Gluster2:Brick1, Gluster3:Brick1, Gluster4:Brick1, Gluster5:Brick1, and Gluster6:Brick1.

$ gluster volume create pf9 disperse-data 4 redundancy 2 transport tcp \

gluster01:/bricks/brick1/brick gluster02:/bricks/brick1/brick gluster03:/bricks/brick1/brick \

gluster04:/bricks/brick1/brick gluster05:/bricks/brick1/brick gluster06:/bricks/brick1/brick \

gluster02:/bricks/brick2/brick gluster03:/bricks/brick2/brick gluster04:/bricks/brick2/brick \

gluster05:/bricks/brick2/brick gluster06:/bricks/brick2/brick gluster01:/bricks/brick2/brick \

gluster03:/bricks/brick3/brick gluster04:/bricks/brick3/brick gluster05:/bricks/brick3/brick \

gluster06:/bricks/brick3/brick gluster01:/bricks/brick3/brick gluster02:/bricks/brick3/brick \

gluster04:/bricks/brick4/brick gluster05:/bricks/brick4/brick gluster06:/bricks/brick4/brick \

gluster01:/bricks/brick4/brick gluster02:/bricks/brick4/brick gluster03:/bricks/brick4/brick \

gluster05:/bricks/brick5/brick gluster06:/bricks/brick5/brick gluster01:/bricks/brick5/brick \

gluster02:/bricks/brick5/brick gluster03:/bricks/brick5/brick gluster04:/bricks/brick5/brick \

gluster06:/bricks/brick6/brick gluster01:/bricks/brick6/brick gluster02:/bricks/brick6/brick \

gluster03:/bricks/brick6/brick gluster04:/bricks/brick6/brick gluster05:/bricks/brick6/brick

6) Expand your volume

Replication:

To add a new node to your gluster cluster, use the following command:

$ gluster peer probe gluster07

Relocate the replica bricks by allocating them across the new node to make sure data is spread evenly:

$ gluster volume replace-brick pf9 gluster03:/bricks/brick1/brick gluster07:/bricks/brick1/brick commit force

$ gluster volume replace-brick pf9 gluster04:/bricks/brick2/brick gluster07:/bricks/brick2/brick commit force

$ gluster volume replace-brick pf9 gluster05:/bricks/brick3/brick gluster07:/bricks/brick3/brick commit force

$ gluster volume replace-brick pf9 gluster06:/bricks/brick4/brick gluster07:/bricks/brick4/brick commit force

You must clean up the old bricks to be able to use these bricks again to expand the volume.

$ gluster03:$ rm -rf /bricks/brick1/brick

gluster03:$ mkdir –p /bricks/brick1/brick

gluster04:$ rm -rf /bricks/brick2/brick

gluster04:$ mkdir –p /bricks/brick2/brick

gluster05:$ rm -rf /bricks/brick3/brick

gluster05:$ mkdir –p /bricks/brick3/brick

gluster06:$ rm -rf /bricks/brick4/brick

gluster06:$ mkdir –p /bricks/brick4/brick

To expand volume, add new bricks using the old bricks that you have relocated and the new bricks from the new node you have created:

$ gluster volume add-brick pf9 replica 3 gluster03:/bricks/brick1/brick gluster04:/bricks/brick2/brick gluster07:/bricks/brick5/brick

$ gluster volume add-brick pf9 replica 3 gluster05:/bricks/brick3/brick gluster06:/bricks/brick4/brick gluster07:/bricks/brick6/brick

Rebalance volume:

$ gluster volume rebalance pf9 start

You have now replicated some of your disks by expanding volume to 42 total disks, with 28 disks being redundant and 14 being data disks

Erasure Coding

If you want to expand this volume, you first need to add a new node with available disks.

$ gluster peer probe gluster07

You will now need to relocate existing bricks from the old nodes to the new nodes in order to make room to create another dispersed data set.

$ gluster volume replace-brick pf9 gluster01:/bricks/brick1/brick gluster07:/bricks/brick1/brick commit force

$ gluster volume replace-brick pf9 gluster02:/bricks/brick2/brick gluster07:/bricks/brick2/brick commit force

$ gluster volume replace-brick pf9 gluster03:/bricks/brick3/brick gluster07:/bricks/brick3/brick commit force

$ gluster volume replace-brick pf9 gluster04:/bricks/brick5/brick gluster07:/bricks/brick5/brick commit force

$ gluster volume replace-brick pf9 gluster06:/bricks/brick6/brick gluster07:/bricks/brick6/brick commit force

Clean up old bricks to be able to use these bricks again to expand the volume:

gluster01:$ rm -rf /bricks/brick1/brick

gluster01:$ mkdir –p /bricks/brick1/brick

gluster02:$ rm -rf /bricks/brick2/brick

gluster02:$ mkdir –p /bricks/brick2/brick

gluster03:$ rm -rf /bricks/brick3/brick

gluster03:$ mkdir –p /bricks/brick3/brick

gluster04:$ rm -rf /bricks/brick5/brick

gluster04:$ mkdir –p /bricks/brick5/brick

gluster06:$ rm -rf /bricks/brick6/brick

gluster06:$ mkdir –p /bricks/brick6/brick

Expand the volume by adding new bricks by using the old bricks that you relocated and new bricks from the new node you created.

$ gluster volume add-brick pf9 disperse-data 4 redundancy 2 \

gluster01:/bricks/brick1/brick gluster02:/bricks/brick2/brick gluster03:/bricks/brick3/brick \

gluster04:/bricks/brick5/brick gluster06:/bricks/brick6/brick gluster07:/bricks/brick4/brick

Rebalance volume:

$ gluster volume rebalance pf9 start

You are left with 42 disks, with 14 redundant disks and 28 data disks.

7) Start your volume:

You’ll need to start the GlusterFS volume

$ gluster volume start pf9

8) Mount your volume:

Create a directory to mount your gluster volume

$ mkdir -p /volumes/glusterfs/pf9

Add your gluster volume to fstab

$ echo 'gluster01:/pf9 /volumes/glusterfs/pf9 glusterfs defaults,_netdev 0 0' >> /etc/fstab

Mount the volume

$ mount -a

Bitrot Detection

Bitrot Daemon is used to locate silent data corruption on a disk where there is no indication from the disk to the storage software layer that a disk error occurred. When BitRot Daemon is enabled on a volume, it signs all of the files and objects in the volume with a hash, which is a checksum calculated on the data of the file and is stored in a persistent data store. It checks data periodically for signature verification and makes note of all anomalies and records them on log files. Files or objects are only signed when they are considered stable, meaning there are no modifications to the file or object for “X” amount of time and all file descriptors are closed (release()). Modification is indicated by O_WRONLY or O_RDWR.

File Integrity

Scrubber validates the integrity of a file by comparing the checksum on the file data with the stored checksum. When there is silent data corruption, scrubbing will mark the files as bad and try to recover the files. Since scrubbing must be done on the entire filesystem tree for the checksum computations, it is CPU-intensive and can be done bi-weekly or monthly.

Other Filesystems

Other open source file systems like GlusterFS include Ceph, OpenStack Swift, Lustre, OpenAFS, HDFS etc. Ceph is one of GlusterFS’s main competitors, each offering different approach to file systems solutions. Ceph, along with OpenStack Swift and Amazon S3, are object-store systems where data is stored as binary objects. However, GlusterFS’s specialty is that it does not use any metadata servers; instead it uses the extended attributes of the back-end filesystem for storing metadata.

Summary

GlusterFS allows you to build a multi-tenant storage system that is centrally managed, so your data is accessible by your IT staff from any location. GlusterFS improves back-end storage by enhancing the performance of frequently accessed files through automatic storage tiering and allowing for quick deployment of additional storage capacity. The risk of data corruption is lowered while the integrity of your data is preserved.

Author
Recent Posts

Platform9

Platform9 is the only container management and orchestration solution that supports the complete lifecycle of the cloud-native journey on infrastructure anywhere. Our unique management plane simplifies operations, including observability & monitoring, increases uptime, minimizes costs, and scales as you scale. Built on open foundations, our platform enables enterprises to keep up with the ever-changing cloud-native ecosystem. Our cloud-native wizards provide Always-on Assurance™ every step of the way to ensure a smooth, simplified journey. Innovative enterprises like Juniper, Kingfisher Plc, Mavenir, Redfin, and Cloudera achieve 4x faster time-to-market, up to 90% reduction in operational costs, and 99.99% uptime.Platform9 is a better way to go cloud native, paving the way for an open distributed cloud

How Three Enterprises Implemented OpenStack and Kubernetes

By Platform9

Migrating Off AWS: How to Move Production from AWS to Enterprise Data Centers Running OpenStack and Kubernetes

By Platform9

Categorized within: OpenStack Tags: Distributed Storage, Erasure Coding, Gluster, GlusterFS, Replication