LU-12292 lnet: keep health even if recovery failed

[fs/lustre-release.git] / lnet / lnet / lib-msg.c

diff --git a/lnet/lnet/lib-msg.c b/lnet/lnet/lib-msg.c
index e92b5c8..3e2c1f2 100644 (file)
--- a/lnet/lnet/lib-msg.c
+++ b/lnet/lnet/lib-msg.c
@@ -490,7 +490,11 @@ lnet_handle_remote_failure_locked(struct lnet_peer_ni *lpni)
        __u32 sensitivity = lnet_health_sensitivity;
        __u32 lp_sensitivity;
 
-       /* lpni could be NULL if we're in the LOLND case */
+       /*
+        * NO-OP if:
+        * 1. lpni could be NULL if we're in the LOLND case
+        * 2. this is a recovery message
+        */
        if (!lpni)
                return;
 
@@ -875,7 +879,12 @@ lnet_health_check(struct lnet_msg *msg)
        case LNET_MSG_STATUS_LOCAL_ABORTED:
        case LNET_MSG_STATUS_LOCAL_NO_ROUTE:
        case LNET_MSG_STATUS_LOCAL_TIMEOUT:
-               lnet_handle_local_failure(ni);
+               /*
+                * don't further decrement the health value if the
+                * recovery message failed.
+                */
+               if (!msg->msg_recovery)
+                       lnet_handle_local_failure(ni);
                if (msg->msg_tx_committed)
                        /* add to the re-send queue */
                        return lnet_attempt_msg_resend(msg);
@@ -886,7 +895,12 @@ lnet_health_check(struct lnet_msg *msg)
         * finalize the message
         */
        case LNET_MSG_STATUS_LOCAL_ERROR:
-               lnet_handle_local_failure(ni);
+               /*
+                * don't further decrement the health value if the
+                * recovery message failed.
+                */
+               if (!msg->msg_recovery)
+                       lnet_handle_local_failure(ni);
                return -1;
 
        /*
@@ -894,7 +908,8 @@ lnet_health_check(struct lnet_msg *msg)
         * attempt a resend safely.
         */
        case LNET_MSG_STATUS_REMOTE_DROPPED:
-               lnet_handle_remote_failure(lpni);
+               if (!msg->msg_recovery)
+                       lnet_handle_remote_failure(lpni);
                if (msg->msg_tx_committed)
                        return lnet_attempt_msg_resend(msg);
                break;
@@ -902,7 +917,8 @@ lnet_health_check(struct lnet_msg *msg)
        case LNET_MSG_STATUS_REMOTE_ERROR:
        case LNET_MSG_STATUS_REMOTE_TIMEOUT:
        case LNET_MSG_STATUS_NETWORK_TIMEOUT:
-               lnet_handle_remote_failure(lpni);
+               if (!msg->msg_recovery)
+                       lnet_handle_remote_failure(lpni);
                return -1;
        default:
                LBUG();