Whamcloud - gitweb
LU-13501 lnet: Skip health and resends for single rail configs 48/38448/7
authorChris Horn <chris.horn@hpe.com>
Tue, 26 May 2020 16:31:26 +0000 (11:31 -0500)
committerOleg Drokin <green@whamcloud.com>
Fri, 19 Jun 2020 16:50:56 +0000 (16:50 +0000)
commitc5381d73b1d83deed561456bc476d63696b9af16
treef60df57eef883fc135f2ac26b244c13a67c8d67c
parent1970a12d7d559dd6d08765047af2b99a3a388524
LU-13501 lnet: Skip health and resends for single rail configs

If the sender of a message only has a single interface it doesn't
make sense to have LNet track the health of that interface, nor
should it attempt to resend a message when it encounters a local
error. There aren't any alternative interfaces to use for a resend.

Similarly, we needn't track health values of a peer's NIs if the peer
only has a single interface. Nor do we need to attempt to resend
a message to a peer with a single interface. There's an exception for
routers. We rely on NI health to determine route aliveness, so even
if a router only has a single interface we still need to track its
health.

We can use the ln_ping_target to get the count of local NIs, and the
lnet_peer struct already contains a count of the number of peer NIs.

HPE-bug-id: LUS-8826
Signed-off-by: Chris Horn <hornc@cray.com>
Change-Id: Id89159a5d07c1668c1cbdfa9050535380f68d1f6
Reviewed-on: https://review.whamcloud.com/38448
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: Serguei Smirnov <ssmirnov@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Alexander Boyko <alexander.boyko@hpe.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
lnet/lnet/lib-msg.c