Merge branch 'gh177-staged-clustering' into jdm-merge-staged-clustering

Resolves merge conflicts between staged clustering and claim sim/claimv3 branches.

Conflicts:
	src/riak_core.erl
	src/riak_core_gossip.erl
	src/riak_core_ring.erl

Author: Jon Meredith

ebin/riak_core.app

@@ -20,6 +20,7 @@
              riak_core_app,
              riak_core_bucket,
              riak_core_cinfo_core,
+             riak_core_claimant,
              riak_core_claim,
              riak_core_claim_sim,
              riak_core_claim_util,
@@ -113,7 +114,7 @@
          {vnode_inactivity_timeout, 60000},
 
          %% Number of VNodes allowed to do handoff concurrently.
-         {handoff_concurrency, 1},
+         {handoff_concurrency, 2},
 
          %% Disable Nagle on HTTP sockets
          {disable_http_nagle, true},

src/riak_core.erl

@@ -20,8 +20,8 @@
 %%
 %% -------------------------------------------------------------------
 -module(riak_core).
--export([stop/0, stop/1, join/1, join/4, remove/1, down/1, leave/0,
-         remove_from_cluster/1]).
+-export([stop/0, stop/1, join/1, join/5, staged_join/1, remove/1, down/1,
+         leave/0, remove_from_cluster/1]).
 -export([vnode_modules/0]).
 -export([register/1, register/2, bucket_fixups/0, bucket_validators/0]).
 
@@ -54,19 +54,28 @@ stop(Reason) ->
 %%
 %% @doc Join the ring found on the specified remote node
 %%
-join(NodeStr) when is_list(NodeStr) ->
-    join(riak_core_util:str_to_node(NodeStr));
-join(Node) when is_atom(Node) ->
-    join(node(), Node).
+join(Node) ->
+    join(Node, true).
 
-join(Node, Node) ->
+%% @doc Join the remote cluster without automatically claiming ring
+%%      ownership. Used to stage a join in the newer plan/commit
+%%      approach to cluster administration. See {@link riak_core_claimant}
+staged_join(Node) ->
+    join(Node, false).
+
+join(NodeStr, Auto) when is_list(NodeStr) ->
+    join(riak_core_util:str_to_node(NodeStr), Auto);
+join(Node, Auto) when is_atom(Node) ->
+    join(node(), Node, Auto).
+
+join(Node, Node, _) ->
     {error, self_join};
-join(_, Node) ->
-    join(riak_core_gossip:legacy_gossip(), node(), Node, false).
+join(_, Node, Auto) ->
+    join(riak_core_gossip:legacy_gossip(), node(), Node, false, Auto).
 
-join(true, _, Node, _Rejoin) ->
+join(true, _, Node, _Rejoin, _Auto) ->
     legacy_join(Node);
-join(false, _, Node, Rejoin) ->
+join(false, _, Node, Rejoin, Auto) ->
     case net_adm:ping(Node) of
         pang ->
             {error, not_reachable};
@@ -78,7 +87,7 @@ join(false, _, Node, Rejoin) ->
                     %% Failure due to trying to join older node that
                     %% doesn't define legacy_gossip will be handled
                     %% in standard_join based on seeing a legacy ring.
-                    standard_join(Node, Rejoin)
+                    standard_join(Node, Rejoin, Auto)
             end
     end.
 
@@ -95,7 +104,7 @@ get_other_ring(Node) ->
             Error
     end.
 
-standard_join(Node, Rejoin) when is_atom(Node) ->
+standard_join(Node, Rejoin, Auto) when is_atom(Node) ->
     case net_adm:ping(Node) of
         pong ->
             case get_other_ring(Node) of
@@ -104,7 +113,7 @@ standard_join(Node, Rejoin) when is_atom(Node) ->
                         true ->
                             legacy_join(Node);
                         false ->
-                            standard_join(Node, Ring, Rejoin)
+                            standard_join(Node, Ring, Rejoin, Auto)
                     end;
                 _ -> 
                     {error, unable_to_get_join_ring}
@@ -113,7 +122,7 @@ standard_join(Node, Rejoin) when is_atom(Node) ->
             {error, not_reachable}
     end.
 
-standard_join(Node, Ring, Rejoin) ->
+standard_join(Node, Ring, Rejoin, Auto) ->
     {ok, MyRing} = riak_core_ring_manager:get_raw_ring(),
     SameSize = (riak_core_ring:num_partitions(MyRing) =:=
                 riak_core_ring:num_partitions(Ring)),
@@ -135,10 +144,16 @@ standard_join(Node, Ring, Rejoin) ->
                                                   gossip_vsn,
                                                   GossipVsn),
             {_, Ring5} = riak_core_capability:update_ring(Ring4),
-            riak_core_ring_manager:set_my_ring(Ring5),
+            Ring6 = maybe_auto_join(Auto, node(), Ring5),
+            riak_core_ring_manager:set_my_ring(Ring6),
             riak_core_gossip:send_ring(Node, node())
     end.
 
+maybe_auto_join(false, _Node, Ring) ->
+    Ring;
+maybe_auto_join(true, Node, Ring) ->
+    riak_core_ring:update_member_meta(Node, Ring, Node, '$autojoin', true).
+
 legacy_join(Node) when is_atom(Node) ->
     {ok, OurRingSize} = application:get_env(riak_core, ring_creation_size),
     case net_adm:ping(Node) of

src/riak_core_claim.erl

@@ -291,7 +291,8 @@ choose_claim_v2(Ring, Node, Params0) ->
     RingSize = riak_core_ring:num_partitions(Ring),
     NodeCount = erlang:length(Active),
     Avg = RingSize div NodeCount,
-    Deltas = [{Member, Avg - Count} || {Member, Count} <- Counts],
+    ActiveDeltas = [{Member, Avg - Count} || {Member, Count} <- Counts],
+    Deltas = add_default_deltas(Owners, ActiveDeltas, 0),
     {_, Want} = lists:keyfind(Node, 1, Deltas),
     TargetN = proplists:get_value(target_n_val, Params),
     AllIndices = lists:zip(lists:seq(0, length(Owners)-1),
@@ -672,6 +673,13 @@ get_counts(Nodes, Ring) ->
                          end, dict:from_list(Empty), Ring),
     dict:to_list(Counts).
 
+%% @private
+add_default_deltas(IdxOwners, Deltas, Default) ->
+    {_, Owners} = lists:unzip(IdxOwners),
+    Owners2 = lists:usort(Owners),
+    Defaults = [{Member, Default} || Member <- Owners2],
+    lists:usort(Deltas ++ Defaults).
+
 %% @private
 get_expected_partitions(Ring, Node) ->
     riak_core_ring:num_partitions(Ring) div get_member_count(Ring, Node).

src/riak_core_claim_sim.erl

@@ -272,7 +272,7 @@ dryrun1(Ring00, CmdsL, #simopts{wants = Wants,
                  cancel ->
                      riak_core_ring:cancel_transfers(Ring00);
                  finish ->
-                     riak_core_ring:finish_transfers(Ring00)
+                     riak_core_ring:future_ring(Ring00)
              end,
 
     Prepared(Ring01, Prepare),
@@ -287,13 +287,13 @@ dryrun1(Ring00, CmdsL, #simopts{wants = Wants,
                             PerCmd(RingAcc3, Cmd),
                             RingAcc3
                     end, RingAcc1, Cmds),
-              {_, NewRing2} = riak_core_ring:reassign_indices(NewRing),
+              {_, NewRing2} = riak_core_claimant:reassign_indices(NewRing),
               Pending = riak_core_ring:pending_changes(NewRing2),
               case Pending of
                   [] ->
                       NewRing3 = NewRing2;
                   _ ->
-                      NewRing3 = riak_core_ring:finish_transfers(NewRing2),
+                      NewRing3 = riak_core_ring:future_ring(NewRing2),
                       PostXfer(NewRing3)
               end,
               NewRing4 = run_rebalance(NewRing3, Wants, Choose, Rebalance),