check_parameter() in lustre/tests/sanity-lnet.sh compares the number of NIs reporting a tunable against ${#INTERFACES[@]}:

check_parameter() {
	local para=$1
	local value=$2

	echo "check parameter ${para} value ${value}"

	return $(( $(do_lnetctl net show -v | \
		     tee /dev/stderr | \
		     grep -c "^ \+${para}: ${value}$") != ${#INTERFACES[@]} ))
}

test_241 configures only INTERFACES[0], so on any node with two or more LNet interfaces the count is 1 while ${#INTERFACES[@]} is 2 or more, and the test fails:

 sanity-lnet test_241: @@@@@@ FAIL: Expected LND timeout 51

The tunable is in fact set correctly. The same run shows:

            lnd tunables:
                  conns_per_peer: 1
                  timeout: 51
                  tos: -1

So this is a defect in the test helper, not in the LND timeout plumbing.

test_310 uses the same helper but configures all interfaces via lnet configure, so it passes today only because the count happens to equal ${#INTERFACES[@]}.

Reproduced on a 3-node rocky9 cluster whose LNet net runs on two interfaces (eth1, eth2).

Suggested fix: derive the expected count from the NIs actually configured on ${NETTYPE} rather than from ${INTERFACES[@]}.