Storage Groups

Alteeve Wiki :: How To :: Storage Groups

In an Anvil! cluster, a "storage group" is a logical group of LVM Volume Groups. When a server is created, the selected storage group tells the cluster where to pull the physical storage from to create the server's virtual hard drive. Most Anvil! nodes have only one VG per subnode and they are generally the same size. As such, the Anvil! is able to automatically link them into a storage group automatically.

Multi-VG Example

Consider a more complex storage configuration, however.

Lets use the example of subnodes with two RAID arrays; The first is made of high speed flash drives, and the second made up of lower speed bulk storage drives. In this case, there will be two storage devices on the nodes;

Block Device	Purpose
/dev/sda	Bulk storage, slow drives. The subnode's OS is installed on this array.
/dev/sdb	Smaller but faster drives, reserved for specific guest servers with high speed storage requirements.

During the OS installation stage, /dev/sda3 is used as the LVM physical volume, or "PV".

Node: an-anvil-01

Subnode	Physical Volume	Size	Volume Group	Type
an-a01n01	/dev/sda3	500 GiB	an-a01n01_vg0	Bulk storage, slow drives. The subnode's OS is installed on this array.
an-a01n02	/dev/sda3	500 GiB	an-a01n02_vg0

After the OS is installed, the /dev/sdb1 partition is used as the PV and used to back a second VG;

Node: an-anvil-01

Subnode	Physical Volume	Size	Volume Group	Type
an-a01n01	/dev/sdb1	10 TiB	an-a01n01_vg1	Smaller but faster drives, reserved for specific guest servers with high speed storage requirements.
an-a01n02	/dev/sdb1	10 TiB	an-a01n02_vg1

Virtual Hard Drives

When a server is created, a virtual hard drive is created to back the storage of that server. Behind the scenes though, an LVM logical volume is created on each subnode, and those are used to back the replicates storage device.

Consider an example server we'll call 'srv01-database'. This looks like this

Server: srv01-database on an-a01n01.

Subnode	Volume Group	Logical Volume	Replicated Volume
an-a01n01	an-a01n01_vg0	/dev/an-a01n01_vg0/srv01-database_0	/dev/drbd/by-res/srv01-database/0
an-a01n02	an-a01n02_vg0	/dev/an-a01n02_vg0/srv01-database_0

At this point, you might be asking yourself; "How did the new server decide to use the bulk storage array? What if I wanted to use the high speed storage?".

This is where Storage Groups come in to play!

Storage Groups; Linking VGs

Note: The Anvil! tries to auto-form storage groups by matching VG sizes. There is no requirement to name volume groups in any specific way, so we can't use the VG names as a reliable way to form storage groups, so size is the only way we can guess which belong together.

In the example above, the high-speed storage VGs were both 500 GiB, and both bulk storage VGs were 10 TiB. As the sizes matched, the Anvil! would auto-create two Storage Groups;

Node: an-a01n01.

Subnode	Volume Group	Storage Group	Type
an-a01n01	an-a01n01_vg0	Storage Group 1	Bulk storage, slow drives. The subnode's OS is installed on this array.
an-a01n02	an-a01n02_vg0
an-a01n01	an-a01n01_vg1	Storage Group 2	Smaller but faster drives, reserved for specific guest servers with high speed storage requirements.
an-a01n02	an-a01n02_vg1

Note: The auto-created storage groups have generic names. You can rename the storage groups. Example; 'anvil-manage-storage-groups --anvil an-anvil-01 --group "Storage Group 1" --new-name "Bulk Storage 1"'.

If you have mis-matched hardware as subnodes in your Anvil! node, then your storage groups may not auto-form. If this happens, you can use anvil-manage-storage-groups' to manually groups VGs into SGs. Please see the man page for details.

Use of Storage Groups

With your storage groups created, you can now use them when you provision a server to tell the Anvil! which type of storage to use when you create a new server!

You can also use anvil-manage-server-storage' to add a second (or third) virtual hard drive to an existing server.

Consider the example 'srv01-database' server from before. In that case, the first disk, and so the disk used to install the guest OS on, was using the bulk 'Storage Group 1'. Now lets say you want to add a second hard drive using 100 GiB from the high speed storage to back the database engine and data. You can do so with a command like this;

• anvil-manage-server-storage --server srv01-database --disk vdb --add 100GiB --storage-group "Storage group 2"

Your server will now have two hard drives, one for either SG!

Server: srv01-database on an-a01n01.

Subnode	Volume Group	Logical Volume	Replicated Volume
an-a01n01	an-a01n01_vg0	/dev/an-a01n01_vg0/srv01-database_0	/dev/drbd/by-res/srv01-database/0
an-a01n02	an-a01n02_vg0	/dev/an-a01n02_vg0/srv01-database_0
an-a01n01	an-a01n01_vg1	/dev/an-a01n01_vg1/srv01-database_1	/dev/drbd/by-res/srv01-database/1
an-a01n02	an-a01n02_vg1	/dev/an-a01n02_vg1/srv01-database_1

An Important Note on Write Order

Note: One of the important features of the Anvil! is write ordering.

In this example, the two virtual hard drives exist on entirely separate RAID arrays. Even so, the order that writes are sent to the two volumes (the two virtual hard drives) are maintained. This is critical to protecting data integrity.

Consider the scenario where you have a write-ahead log on volume 0, and the database data is on volume 1. The write sequence to the database would be:

1. . Write the transaction to be done to volume 0
2. . Write the actual transaction to the database's data set on volume 1
3. . Clear the completed transaction from the log on volume 0

In some systems, the storage would see two writes to volume 0 and one write to volume 1, and decide that it's more efficient to group the writes. If this happened, the actual sequence of writes would be;

1. . Write the transaction to be done to volume 0
2. . Clear the completed transaction from the log on volume 0
3. . Write the actual transaction to the database's data set on volume 1

That's ... not good. If there was an unpredictable failure in between writing to the disk, you could end up in a situation where the write-ahead log is written and cleared as complete, and then the system shuts down mid transaction when working on the real data. This could leave your database in an inconsistent or corrupt state!

The Anvil! ensures that write ordering is maintained across disks to make sure this never happens.

 

Any questions, feedback, advice, complaints or meanderings are welcome.
Alteeve's Niche!	Enterprise Support: Alteeve Support	Community Support
© Alteeve's Niche! Inc. 1997-2024		Anvil! "Intelligent Availability®" Platform
legal stuff: All info is provided "As-Is". Do not use anything here unless you are willing and able to take responsibility for your own actions.