Whamcloud - gitweb
LU-14566 lnet: Skip discovery in LNetPrimaryNID if DD disabled 41/43141/2
authorChris Horn <chris.horn@hpe.com>
Fri, 26 Mar 2021 16:28:18 +0000 (11:28 -0500)
committerOleg Drokin <green@whamcloud.com>
Wed, 28 Apr 2021 02:11:13 +0000 (02:11 +0000)
If discovery is disabled locally then the discovery thread will not
modify any peer objects as a result of the discovery process. Thus,
the primary NID of any peer we're asked to discover will not change
as a result of discovery. Therefore, we do not need to actually
perform discovery in LNetPrimaryNID() if discovery is disabled
locally. Since this routine can result in long client mount times
when a Lustre server is down we should avoid this unnecessary

Test-Parameters: trivial
HPE-bug-id: LUS-9887
Signed-off-by: Chris Horn <chris.horn@hpe.com>
Change-Id: I6d188e16422ad47a146d52bb24cdd1b77a30aa71
Reviewed-on: https://review.whamcloud.com/43141
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Serguei Smirnov <ssmirnov@whamcloud.com>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

index 20ae558..d70415e 100644 (file)
@@ -1347,7 +1347,13 @@ LNetPrimaryNID(lnet_nid_t nid)
        lp = lpni->lpni_peer_net->lpn_peer;
-       while (!lnet_peer_is_uptodate(lp)) {
+       /* If discovery is disabled locally then we needn't bother running
+        * discovery here because discovery will not modify whatever
+        * primary NID is currently set for this peer. If the specified peer is
+        * down then this discovery can introduce long delays into the mount
+        * process, so skip it if it isn't necessary.
+        */
+       while (!lnet_peer_discovery_disabled && !lnet_peer_is_uptodate(lp)) {
                /* force a full discovery cycle */
                lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
@@ -1368,7 +1374,11 @@ LNetPrimaryNID(lnet_nid_t nid)
                lp = lpni->lpni_peer_net->lpn_peer;
-               /* Only try once if discovery is disabled */
+               /* If we find that the peer has discovery disabled then we will
+                * not modify whatever primary NID is currently set for this
+                * peer. Thus, we can break out of this loop even if the peer
+                * is not fully up to date.
+                */
                if (lnet_is_discovery_disabled(lp))