大网数据广告活动
服务器运维问答

GlusterFS Volume Creation Suggestion

· 52CE 我爱测

I have to deploy multiple Openshift Clusters from 3 nodes to 10 nodes. For 3 nodes i am creating volumes as replicated. But for 4 and above it doesn't look nice to create replicated volume, so each node has 300GB disk and replica…

问题原文

原帖链接 · ImranRazaKhan · CC BY-SA 4.0

I have to deploy multiple Openshift Clusters from 3 nodes to 10 nodes. For 3 nodes i am creating volumes as replicated.

But for 4 and above it doesn't look nice to create replicated volume, so each node has 300GB disk and replicating it to 10 nodes is not optimal. I am looking for formula to use like

For 4 nodes create volume as disperse:2:1
For 5 nodes create volume as disperse:?:?
For 6 nodes create volume as disperse:?:?
For 7 nodes create volume as disperse:?:?
For 8 nodes create volume as disperse:?:?
For 9 nodes create volume as disperse:?:?
For 10 nodes create volume as disperse:?:?

Environment: I will use these volumes for MYSQL 5.7.28 and Each Server has 300GB disk, Out of 300GB i will create volume with size of 250GB for MYSQL.

OpenShift 3.11 version

# gluster --version
glusterfs 6.1

PS: i have no background of storage, so excuse if i am missing some obvious point i tried to search on google but couldn't extract required info.

已采纳解答

原帖链接 · Aravinda VK · CC BY-SA 4.0

Are you planning to use all nodes as storage nodes or only a subset of the nodes as storage nodes? Based on your question, MySQL uses 250GiB, what other applications need storage?

Replicate Volume: Effective storage space available will be

volume_size = sum of storage available from three nodes / 3

In your case, Volume size will be 300GiB using three storage nodes.

Disperse Volume: Effective storage space available will be

volume_size = storage in single node * (number of bricks - redundancy count)

In your case, Volume size will be 300 * (3-1) = 600GiB. More detail is available here https://docs.gluster.org/en/v3/Administrator%20Guide/Setting%20Up%20Volumes/#creating-dispersed-volumes Disperse Volumes are good for archival purposes since it can save space compared to Replica volumes. But it may be slow compared to Replica because of the computation involved during every IO.

Kadalu(https://kadalu.io) project provides a different approach to provision Volumes in Kubernetes. It creates a single Gluster Volume from the storage and provides subvolumes out of that volume when PV is requested(In your case Storage for Mysql).

Kadalu currently supports Replica 1 and Replica 3 Volumes. Replica 1 is useful when the storage device is claimed from other storage providers, for example, AWS/Azure. Replica 3 provides a high availability of storage for applications even though one out of three nodes goes down. The recent blog post(https://kadalu.io/blog/kadalu-kubernetes-storage) explains multiple configurations available with Kadalu and using it with the existing storage.

Kadalu uses GlusterFS and integrated with Kubernetes natively, without using Gluster management daemon - glusterd.

Update: Added calculations for Disperse Volume

number of disperse bricks = data bricks + redundancy count

If 3 storage devices are available,

2 data bricks + 1 redundancy bricks

In case of 6 storage devices,

4 data bricks + 2 redundancy bricks

If number of redundancy bricks increases then the usable Volume size will get reduced. The volume will be available to applications even though number of bricks equivalent to redundancy bricks goes down. For example, in 4+2 configuration, Volume will be available even though 2 bricks out of 6 go down.

摘自公开归档,保留原文语言,版式经整理,外部媒体与邮箱地址已移除。

资料来源与适用范围

资料来源
Server Fault
适用范围
以原帖系统、软件版本与发布日期为准(2020-01-17)
内容语言
中文;命令、路径和错误标识保留技术原文
本地收录日期
2026-11-15
授权说明
CC BY-SA 4.0

正文与所需资源由本站本地存储。查看来源及更新记录 ↗

请按适用版本和实际日志确认原因;本站没有对每一种系统或设备逐项执行文中的操作。