From 9a5c1057ba242e3999c68fe8d09bddb5e9a06c97 Mon Sep 17 00:00:00 2001
From: Dave Parfitt <dparfitt@basho.com>
Date: Mon, 22 Jul 2013 11:17:56 -0400
Subject: [PATCH 1/4] cleanup rt_dirty behavior

---
 src/riak_repl2_fscoordinator.erl | 23 +++++++++++++++++++++--
 src/riak_repl_stats.erl          |  6 ++++--
 2 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/src/riak_repl2_fscoordinator.erl b/src/riak_repl2_fscoordinator.erl
index fba95217..bf2a03e7 100644
--- a/src/riak_repl2_fscoordinator.erl
+++ b/src/riak_repl2_fscoordinator.erl
@@ -350,6 +350,7 @@ handle_cast(start_fullsync,  State) ->
             lager:info("Starting fullsync (source) with max_fssource_node=~p and max_fssource_cluster=~p",
                        [MaxSource, MaxCluster]),
             {ok, Ring} = riak_core_ring_manager:get_my_ring(),
+            check_nodes_for_rt_dirty(Ring),
             N = largest_n(Ring),
             Partitions = sort_partitions(Ring),
             State2 = State#state{
@@ -1007,8 +1008,17 @@ notify_rt_dirty_nodes(State = #state{dirty_nodes = DirtyNodes,
             NodesToNotify = lists:subtract(AllNodesList,
                                            ordsets:to_list(DirtyNodesDuringFS)),
             lager:debug("Notifying nodes ~p", [ NodesToNotify]),
-            _ = rpc:multicall(NodesToNotify, riak_repl_stats, clear_rt_dirty, []),
-            State#state{dirty_nodes=ordsets:new()};
+            {_, BadNodes} = rpc:multicall(NodesToNotify,
+                                          riak_repl_stats,
+                                          clear_rt_dirty, []),
+            case BadNodes of
+                [] ->
+                    %% all nodes nodified, clear rt_dirty state
+                    State#state{dirty_nodes=ordsets:new()};
+                Nodes ->
+                    lager:warning("Failed to clear rt_dirty on ~p", [Nodes]),
+                    State
+            end;
         false ->
             lager:debug("No dirty nodes before fullsync started"),
             State
@@ -1063,3 +1073,12 @@ flush_exit_message(Pid) ->
             ok
     end.
 
+%% check all nodes in the cluster for existing rt_dirty files
+%% and reset their rt_dirty flag if it exists
+check_nodes_for_rt_dirty(Ring) ->
+    Owners = riak_core_ring:all_owners(Ring),
+    [ case rpc:call(Node, riak_repl_stats, is_rt_dirty,[]) of
+            false -> ok;
+            _ -> riak_repl2_fscoordinator:node_dirty(Node)
+        end || Node <- Owners ].
+
diff --git a/src/riak_repl_stats.erl b/src/riak_repl_stats.erl
index f654ce49..093ad799 100644
--- a/src/riak_repl_stats.erl
+++ b/src/riak_repl_stats.erl
@@ -140,8 +140,10 @@ rt_dirty() ->
                         riak_repl2_fscoordinator:node_dirty(node())
                       catch
                         _:_ ->
-                         lager:debug("Failed to notify coordinator of rt_dirty status")
-                      end
+                         %% This could be triggered on startup if the
+                         %% fscoordinator isn't running
+                         lager:warning("Failed to notify coordinator of rt_dirty status.")
+                       end
             end),
             ok;
         false -> ok

From de33a5808cc577e6d927d4a05bb8ed36185ea244 Mon Sep 17 00:00:00 2001
From: Micah Warren <mwarren@basho.com>
Date: Tue, 9 Dec 2014 14:28:43 -0600
Subject: [PATCH 2/4] Fixed node pulling from owners proplist.

---
 src/riak_repl2_fscoordinator.erl | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/riak_repl2_fscoordinator.erl b/src/riak_repl2_fscoordinator.erl
index bf2a03e7..6a3a772d 100644
--- a/src/riak_repl2_fscoordinator.erl
+++ b/src/riak_repl2_fscoordinator.erl
@@ -1080,5 +1080,5 @@ check_nodes_for_rt_dirty(Ring) ->
     [ case rpc:call(Node, riak_repl_stats, is_rt_dirty,[]) of
             false -> ok;
             _ -> riak_repl2_fscoordinator:node_dirty(Node)
-        end || Node <- Owners ].
+        end || {_Part, Node} <- Owners ].
 

From cfb3f1f22f1191c5dab299942a6c67b7a2b1246b Mon Sep 17 00:00:00 2001
From: Micah Warren <mwarren@basho.com>
Date: Tue, 9 Dec 2014 14:28:56 -0600
Subject: [PATCH 3/4] Added check so an fscoordinator won't call into itself.

---
 src/riak_repl2_fscoordinator.erl | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/riak_repl2_fscoordinator.erl b/src/riak_repl2_fscoordinator.erl
index 6a3a772d..39120553 100644
--- a/src/riak_repl2_fscoordinator.erl
+++ b/src/riak_repl2_fscoordinator.erl
@@ -195,7 +195,7 @@ node_dirty(Node) ->
         Leader ->
             Fullsyncs = riak_repl2_fscoordinator_sup:started(Leader),
             [riak_repl2_fscoordinator:node_dirty(Pid, Node) ||
-                {_, Pid} <- Fullsyncs]
+                {_, Pid} <- Fullsyncs, Pid =/= self()]
     end.
 
 node_dirty(Pid, Node) ->

From 82c567fb5f9dc9cdffb4d7a3758417c3de5c1c6a Mon Sep 17 00:00:00 2001
From: Micah Warren <mwarren@basho.com>
Date: Mon, 22 Dec 2014 14:02:19 -0600
Subject: [PATCH 4/4] Added reason for node dirty marking failure log warning.

---
 src/riak_repl_stats.erl | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/riak_repl_stats.erl b/src/riak_repl_stats.erl
index 093ad799..25c433c2 100644
--- a/src/riak_repl_stats.erl
+++ b/src/riak_repl_stats.erl
@@ -139,10 +139,10 @@ rt_dirty() ->
                       try
                         riak_repl2_fscoordinator:node_dirty(node())
                       catch
-                        _:_ ->
+                        W:Y ->
                          %% This could be triggered on startup if the
                          %% fscoordinator isn't running
-                         lager:warning("Failed to notify coordinator of rt_dirty status.")
+                         lager:warning("Failed to notify coordinator of rt_dirty status due to ~p:~p.", [W,Y])
                        end
             end),
             ok;