cluster/afr: gluster volume heal $VN info sporadically reports nothing to heal

PROBLEM:
Whenever rsync is done at the mount point of a replicate volume and during the
rsycn glusterd and bricks of one of the machine gets killed then executing
gluster volume heal info gives results sporadically.Sometimes it says Number of
entries are zero and other time it gives different output. When glusterd and
brick of one of the machine gets killed then at other machine,
RPC_CLNT_DISCONNECT will get triggered periodically. So it may happen that
during the execution of gluster volume heal info command, RPC_CLNT_DISCONNECT
also gets triggered and glusterd will assume that it got a new rpc_disconnet
notification and glusterd will call the glusterd_op_ac_send_unlock_drain
function instead of calling the glusterd_op_ac_rcvd_brick_op_acc function which
is responsible for printing the entries. So because of RCP_CLNT_DISCONNECT
notification, glusterd will execute functions responsible for
RPC_CLNT_DISCONNECT(glusterd_op_ac_send_unlock_drain) rather than executing
functions responsible for gluster volume heal info
command(glusterd_op_ac_rcvd_brick_op_acc).

FIX:
Put a condition in for RPC_CLNT_DISCONNECT in the glusterd_peer_rpc_notify
function so that if RPC_DISCONNECT is triggered for the first for a peer then
glusterd mark that peer as disconnected so that if RPC_DISCONNET is triggered
again, because peer is already marked as disconnected, glusterd will not inject
EVENTR_ACK_DRAIN and prints the Number of entries.

Change-Id: I770e3175b050fd7c7beb801b611a2dff9bfcbac8
BUG: 857503
Signed-off-by: Venkatesh Somyajulu <vsomyaju@redhat.com>
Reviewed-on: http://review.gluster.org/4043
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

1 files changed, 29 insertions, 24 deletions

diff --git a/xlators/mgmt/glusterd/src/glusterd-handler.c b/xlators/mgmt/glusterd/src/glusterd-handler.c
index cd395c49e73..ae4138ebe7a 100644
--- a/xlators/mgmt/glusterd/src/glusterd-handler.c
+++ b/xlators/mgmt/glusterd/src/glusterd-handler.c
@@ -2892,41 +2892,46 @@ glusterd_peer_rpc_notify (struct rpc_clnt *rpc, void *mydata,
                 gf_log (this->name, GF_LOG_DEBUG, "got RPC_CLNT_DISCONNECT %d",
                         peerinfo->state.state);
 
-                peerinfo->connected = 0;
-
                 /*
                   local glusterd (thinks that it) is the owner of the cluster
                   lock and 'fails' the operation on the first disconnect from
                   a peer.
                 */
-                glusterd_get_lock_owner (&owner);
-                if (!uuid_compare (conf->uuid, owner)) {
-                        ret = glusterd_op_sm_inject_event (GD_OP_EVENT_START_UNLOCK,
-                                                           NULL);
+
+                if (peerinfo->connected) {
+                        glusterd_get_lock_owner (&owner);
+                        if (!uuid_compare (conf->uuid, owner)) {
+                                ret = glusterd_op_sm_inject_event
+                                      (GD_OP_EVENT_START_UNLOCK, NULL);
+                                if (ret)
+                                        gf_log (this->name, GF_LOG_ERROR,
+                                                "Unable to enqueue cluster "
+                                                "unlock event");
+                                break;
+                        }
+
+                        peer_uuid = GF_CALLOC (1, sizeof (*peer_uuid),
+                                               gf_common_mt_char);
+                        if (!peer_uuid) {
+                                ret = -1;
+                                break;
+                        }
+
+                        uuid_copy (*peer_uuid, peerinfo->uuid);
+                        ret = glusterd_op_sm_inject_event
+                              (GD_OP_EVENT_LOCAL_UNLOCK_NO_RESP, peer_uuid);
                         if (ret)
                                 gf_log (this->name, GF_LOG_ERROR, "Unable"
-                                        " to enqueue cluster unlock event");
-                        break;
-                }
-
-                peer_uuid = GF_CALLOC (1, sizeof (*peer_uuid), gf_common_mt_char);
-                if (!peer_uuid) {
-                        ret = -1;
-                        break;
-                }
+                                        " to enque local lock flush event.");
 
-                uuid_copy (*peer_uuid, peerinfo->uuid);
-                ret = glusterd_op_sm_inject_event (GD_OP_EVENT_LOCAL_UNLOCK_NO_RESP,
-                                                   peer_uuid);
-                if (ret)
-                        gf_log (this->name, GF_LOG_ERROR, "Unable"
-                                " to enque local lock flush event.");
+                        //Inject friend disconnected here
+                        if (peerinfo->state.state == GD_FRIEND_STATE_DEFAULT)  {
+                                glusterd_friend_remove_notify (peerctx);
+                        }
 
-                //Inject friend disconnected here
-                if (peerinfo->state.state == GD_FRIEND_STATE_DEFAULT)  {
-                        glusterd_friend_remove_notify (peerctx);
                 }
 
+                peerinfo->connected = 0;
                 //default_notify (this, GF_EVENT_CHILD_DOWN, NULL);
                 break;
         }