Whamcloud - gitweb
LU-15275 lnet: Skip router discovery on send path
authorChris Horn <chris.horn@hpe.com>
Tue, 30 Nov 2021 16:57:34 +0000 (10:57 -0600)
committerAndreas Dilger <adilger@whamcloud.com>
Fri, 9 Sep 2022 01:38:53 +0000 (01:38 +0000)
commitbf7b1968ebeb511fb01e65408c85db864f18e83f
tree11968ee4d80cb78ad15a2650ce048e028a894391
parent1a667fec1ee1397be1b66ddcc3d5069d7883e7e7
LU-15275 lnet: Skip router discovery on send path

When the router checker is enabled, routes are regularly marked as out
of date w.r.t. discovery. This can cause upper level messages to be
delayed while the router undergoes discovery. We can avoid delaying
messages by relying on the router checker to initiate discovery of
routers. If we happen to send a message to a router before it has
been discovered then the worst case scenario is that the route is
actually down or we end up utilizing a subset of a multi-rail router's
interfaces. Both situations can be remedied by utilizing the
check_routers_before_use parameter.

Change the logic in lnet_handle_find_routed_path() so that we only
initiate discovery if the alive_router_check_interval is <= 0 (i.e.
router checker pings are disabled).

Lustre-change: https://review.whamcloud.com/45684
Lustre-commit: c8e74c395d5634dbb0d9d8a86605bb36ab2b8233

Test-Parameters: trivial testlist=sanity-lnet
Signed-off-by: Chris Horn <chris.horn@hpe.com>
Change-Id: If0332c21f6157117598b7b908fe17f2d2690fc1d
Reviewed-on: https://review.whamcloud.com/48383
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Frank Sehr <fsehr@whamcloud.com>
Reviewed-by: Alexandre Ioffe <aioffe@ddn.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
lnet/lnet/lib-move.c