Whamcloud - gitweb
LU-16378 lnet: handles unregister/register events
authorCyril Bordage <cbordage@whamcloud.com>
Mon, 12 Dec 2022 10:49:11 +0000 (11:49 +0100)
committerAndreas Dilger <adilger@whamcloud.com>
Sat, 17 Dec 2022 02:23:38 +0000 (02:23 +0000)
commita7ee03d7ca4185e7aadded253f2fea60192301d6
tree92994bccc64ff1bb9ba1fb57b6ae65e58132da9f
parentc522bcd8a1042af099a9892311cc509ce0e1b4ad
LU-16378 lnet: handles unregister/register events

When network is restarted, devices are unregistered and then
registered again. When a device registers using an index that is
different from the previous one (before network was restarted), LNet
ignores it. Consequently, this device stays with link in fatal state.

To fix that, we catch unregistering events to clear the saved index
value, and when a registering event comes, we save the new value.

Lustre-change: https://review.whamcloud.com/49375/
Lustre-commit: TBD (from 7442710a56a8f38453441c62253c0ad891fe9b8c)

Signed-off-by: Cyril Bordage <cbordage@whamcloud.com>
Change-Id: I17e93a1103d588f3e630a9c7446b345f4d472b97
Reviewed-on: https://review.whamcloud.com/c/ex/lustre-release/+/49376
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Serguei Smirnov <ssmirnov@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
lnet/klnds/socklnd/socklnd.c