7.8 KiB

Raw Blame History

Gluster charm

GlusterFS is an open source, distributed file system capable of scaling to several petabytes (actually, 72 brontobytes!) and handling thousands of clients. GlusterFS clusters together storage building blocks over Infiniband RDMA or TCP/IP interconnect, aggregating disk and memory resources and managing data in a single global namespace. GlusterFS is based on a stackable user space design and can deliver exceptional performance for diverse workloads.

Usage

The gluster charm has defaults in the config.yaml that you will want to change for production. Please note that volume_name, cluster_type, and replication_level are immutable options. Changing them post deployment will have no effect. This charm makes use of juju storage. Please read the docs to learn about adding block storage to your units.

volume_name:
    Whatever name you would like to call your gluster volume.
cluster_type:
    The default here is Replicate but you can also set it to
     * Arbiter
     * Distribute
     * Stripe
     * Replicate
     * Striped-Replicate
     * Disperse
     * Distributed-Stripe
     * Distributed-Replicate
     * Distributed-Striped-Replicate
     * Distributed-Disperse
replication_level:
    The default here is 3
    If you don't know what any of these mean don't worry about it. 
    The defaults are sane.

Actions

This charm several actions to help manage your Gluster cluster.

Creating volume quotes. Example: juju action do --unit gluster/0 create-volume-quota volume=test usage-limit=1000MB
Deleting volume quotas. Example: juju action do --unit gluster/0 delete-volume-quota volume=test
Listing the current volume quotas. Example: juju action do --unit gluster/0 list-volume-quotas volume=test
Setting volume options. This can be used to set several volume options at once. Example: juju action do --unit gluster/0 set-volume-options volume=test performance-cache-size=1GB performance-write-behind-window-size=1MB

Building from Source

Configure

Create a config.yaml file to set any options you would like to change from the defaults.

Deploy

This charm requires juju storage. It requires at least 1 block device. For more information please check out the docs

Example EC2 deployment on Juju 2.1:
juju deploy cs:~xfactor973/xenial/gluster-3 -n 3 --config=~/gluster.yaml --storage brick=ebs,10G,2

To scale out the service use this command:
juju add-unit gluster

(keep adding units to keep adding more bricks and storage)

Scale Out

Note that during scale out operation if your cluster has existing files on there they will not be migrated to the new bricks until a gluster volume rebalance start operation is performed. This operation can slow client traffic so it is left up to the administrator to perform at the appropriate time.

Rolling Upgrades

The config.yaml source option is used to kick off a rolling upgrade of your cluster. The current behavior is to install the new packages on the server and upgrade it one by one. A UUID sorted order is used to define the upgrade order. Please note that replica 3 is required to use rolling upgrades. With replica 2 it's possible to have split brain issues.

Testing

For a simple test deploy 4 gluster units like so:

juju deploy gluster -n 4 --config=~/gluster.yaml --storage brick=local,10G

Once the status is started the charm will bring both units together into a cluster and create a volume. You will know the cluster is ready when you see a status of active.

Now you can mount the exported GlusterFS filesystem with either fuse or NFS. Fuse has the advantage of knowing how to talk to all replicas in your Gluster cluster so it will not need other high availablity software. NFSv3 is a point to point so it will need something like virtual IP's, DNS round robin or something else to ensure availability if a unit should die or go away suddenly. Install the glusterfs-client package on your host. You can reference the ./hooks/install file to show you how to install the glusterfs packages.

On your juju host you can mount Gluster with fuse like so:

mount -t glusterfs <ip or hostname of unit>:/<volume_name> mount_point/

High Availability

There's 3 ways you can achieve high availability with Gluster.

The first an easiest method is to simply use the glusterfs fuse mount on all clients. This has the advantage of knowing where all servers in the cluster are at and will reconnect as needed and failover gracefully.
Using virtual ip addresses with a DNS round robin A record. This solution applies to NFSv3. This method is more complicated but has the advantage of being usable on clients that only support NFSv3. NFSv3 is stateless and this can be used to your advantage by floating virtual ip addresses that failover quickly. To use this setting please set the virtual_ip_addresses config.yaml setting after reading the usage.
Using the Gluster coreutils. If you do not need a mount point then this is a viable option. glusterfs-coreutils provides a set of basic utilities such as cat, cp, flock, ls, mkdir, rm, stat and tail that are implemented specifically using the GlusterFS API commonly known as libgfapi. These utilities can be used either inside a gluster remote shell or as standalone commands with 'gf' prepended to their respective base names. Example usage is shown here: Docs

MultiTenancy

Gluster provides a few easy ways to have multiple clients in the same volume without them knowing about one another.

Deep Mounting. Gluster NFS supports deep mounting which allows the sysadmin to create a top level directory for each client. Then instead of mounting the volume you mount the volume + the directory name. Now the client only sees their files. This doesn't stop a malacious client from remounting the top level directory.

This can be combined with posix acl's if your tenants are not trustworthy.
Another option is combining with Netgroups. This feature allows users to restrict access specific IPs (exports authentication) or a netgroup (netgroups authentication), or a combination of both for both Gluster volumes and subdirectories within Gluster volumes.

Filesystem Support:

The charm supports several filesystems currently. Btrfs, Ext4, Xfs and ZFS. The default filesystem can be set in the config.yaml. The charm currently defaults to XFS but ZFS would likely be a safe choice and enable advanced functionality such as bcache backed gluster bricks. Note: The ZFS filesystem requires Ubuntu 16.04 or greater

Notes:

If you're using containers to test Gluster you might need to edit /etc/default/lxc-net and read the last section about if you want lxcbr0's dnsmasq to resolve the .lxc domain

Now to show that your cluster can handle failure you can:

juju destroy-machine n;

This will remove one of the units from your cluster and simulate a hard failure. List your files on the mount point to show that they are still available.

Reference

For more information about Gluster and operation of a cluster please see: https://gluster.readthedocs.org/en/latest/ For more immediate and interactive help please join IRC channel #gluster on Freenode. Gluster also has a users mailing list: https://www.gluster.org/mailman/listinfo/gluster-users For bugs concerning the Juju charm please file them on Github

7.8 KiB Raw Blame History