[LU-14955] LNet: change use of fatal error flag for ni selection to be a part of health feature

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Assignee: Serguei Smirnov
• Reporter: Serguei Smirnov

Description

Making use of fatal error flag for NI selection to be a part of health feature allows the user to control it by turning the health feature on/off. Some user may decide to turn of fatal link state detection if LNet is configured with a single NI.

Comments

[Peter Jones]

Landed for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/44746/
Subject: LU-14955 lnet: Use fatal NI if none other available
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ff3322fd0c77a8042558711d9f410326d2aa6375

[Gerrit Updater]

"Serguei Smirnov <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/44746
Subject: LU-14955 lnet: make fatal ni handling part of health feature
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 2f9ae83236e0c4cda5b5a1ae04ab1f71a3cf6036

[Serguei Smirnov]

Andreas, yes, after some debating, the patch that I'm about to push is going to allow the only NI to be picked regardless of its fatal state. As an extrapolation, if there are many NIs and they are all in fatal state, LNet is also going to be able to select one.

[Andreas Dilger]

Serguei, is there any real benefit for the only link/interface on a node to be marked unavailable? I don't think that makes sense. In general, I guess enabling LNet Health doesn't make sense for a system with only a single link/interface, since there isn't any choice but to continue using that one interface. In most such cases, the error will be transient, so retrying will fix the problem, and if the only interface on a client is permanently broken, then there isn't anything that can be done anyway.