After some servers/VMs have experienced an IB issue, some clients have been found in a state where access to a target can not be restored by recovery/HA mechanism.

This unexpected behaviour is always showing the same set of symptoms like, import stuck in LUSTRE_IMP_CONNECTING state, SEC_CTX_INIT requests timing out and being resent in a loop to a wrong service node/VM without a new alternate NID being elected, ...

It is interesting to note that both concerned Client/Server versions, respectively lustre-2.14.0_ddn224/lustre-2.14.0_ddn207, each contain the full set of patches that have been identified to have "fixed" similar issues before

A workaround has been found, that is. to manually migrate the concerned target to the expected service node/VM.

And according to Client/Server logs content, associated Lustre versions source code browsing of concerned parts, and forced crash-dump (taken when problem was still alive) analysis, I should able to propose a fix soon since master branch may also be concerned by this problem.