[LUDOC-15] Recommend separate MGT device on backup MDS node

• Status: Open
• Project: Lustre Documentation
• Type: Improvement
• Priority: Minor
• Reporter: Andreas Dilger
• Assignee: Lustre Manual Triage
• Resolution: Unresolved
• Story Points: 3

Description

For Imperative Recovery and future Sequoia (ZFS) configurations, it is desirable to configure the MGS to be running on the backup MDS node. The use of IR will increase the load on the MGS server, and by having the MGS on a separate node it will allow IR to speed up MDS recovery time. Having a separate MGS device also avoids problems in the future if there are multiple MDTs on a single node, since it is otherwise complex to stop an MDT without also stopping the MGS on the node, and impossible to run e2fsck on the shared device.

The manual should be updated to discourage the use of shared MGS devices, and explain the reasoning.

Comments

[Jinshan Xiong (Inactive)]

I tend to think this is NOT only a documentation work. Actually we need some development to make it work because there is no MDS in current metadata stack. So active-active failover may not work.

[Andreas Dilger]

I think Di was working on this already in LU-718 (http://review.whamcloud.com/1418).

[Jodi Levi (Inactive)]

Andreas,
Could you have a look at this and let us know if this is still a current issue?

[Andreas Dilger]

I think the coding work is done, but we still may need an update to the documentation. I don't recall that it was done, and nothing is referenced in the ticket here.