[LU-12889] Do not assume peers are MR capable

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.14.0
• Affects Version/s: Lustre 2.13.0
• Severity: 3
• Assignee: Chris Horn
• Reporter: Chris Horn

Description

If a peer has discovery disabled then it will not consolidate peer
NI information. This means we need to use a consistent source NI
when sending to it just like we do for non-MR peers.

A comment in lnet_discovery_event_reply() indicates that this was a
known issue, but the situation is not handled properly.

Do not assume peers are multi-rail capable when peer objects are
allocated and initialized.

Do not mark a peer as multi-rail capable unless all of the following
conditions are satisified:
1. The peer has the MR feature flag set
2. The peer has discovery enabled.
3. We have discovery enabled locally

Marked ticket as critical as it can break setups where one side has discovery enabled and the other side has it disabled.

Issue Links

• is related to: LU-12955 AST replies are dropped when servers are non-MR, clients and routers are MR