- adjust gitignore

- make sure that the results dir is created
- ruggedize observer listen socket creation
- fix support for non-cuttlefish builds
- observer disk autodetection

.gitignore

@@ -16,4 +16,4 @@ doc/
 coverage
 tags
 riak-*
-results/
+*.png

results/.gitignore

@@ -0,0 +1,5 @@
+# Ignore everything in this directory
+*
+# Except this file
+!.gitignore
+

src/observer.erl

@@ -32,15 +32,22 @@ watch(Nodes, Collector) ->
     Pid.
 
 watcher(Master, Nodes, {_Host, Port, _Dir} = Collector) ->
-    {ok, LSock} = gen_tcp:listen(Port, [{active, false}, binary, {packet, 2}]),
-    Acceptor = spawn_link(?MODULE, lloop, [self(), LSock]),
-    monitor(process, Master),
-    Probes = [{Node, undefined} || Node <- Nodes],
-    W = #watcher{nodes=Nodes,
-		 acceptor=Acceptor,
-                 collector=Collector,
-                 probes=Probes},
-    watcher_loop(W).
+    case gen_tcp:listen(Port, [{active, false}, binary, 
+			       {packet, 2}]) of
+	{ok, LSock} ->
+	    Acceptor = spawn_link(?MODULE, lloop, [self(), LSock]),
+	    monitor(process, Master),
+	    Probes = [{Node, undefined} || Node <- Nodes],
+	    W = #watcher{nodes=Nodes,
+			 acceptor={Acceptor, LSock},
+			 collector=Collector,
+			 probes=Probes},
+	    watcher_loop(W);
+	{error, eaddrinuse} ->
+	    timer:sleep(100),
+	    watcher(Master, Nodes, Collector)
+	%% case_clause other errors
+    end.
 
 lloop(Master, LSock) ->
     {ok, Sock} = gen_tcp:accept(LSock),
@@ -49,7 +56,7 @@ lloop(Master, LSock) ->
     lloop(Master, LSock).
 
 watcher_loop(W=#watcher{probes=Probes,
-			acceptor=Acceptor,
+			acceptor={Acceptor,LSock},
 			collector={_,_,Dir}}) ->
     Missing = [Node || {Node, undefined} <- Probes],
     %% io:format("Missing: ~p~n", [Missing]),
@@ -87,7 +94,12 @@ watcher_loop(W=#watcher{probes=Probes,
 	    ?MODULE:watcher_loop(W2);
 	stop ->
 	    exit(Acceptor),
-	    [file:close(FD) || {_Sock, FD} <- get()]
+	    gen_tcp:close(LSock),
+	    [begin
+		 file:close(FD),
+		 gen_tcp:close(Sock)
+	     end
+	     || {Sock, FD} <- get()]
     end.
 
 
@@ -184,16 +196,19 @@ collect(H0) ->
     {_, P} = report_processes(H),
     {H2, N} = report_network(H),
 
-    %% this needs to figure out the active disks on its own, or report
-    %% for all disks
-    {H3, D} = report_disk2([{<<"dm-0">>, "dm-0"},
-			    {<<"dm-1">>, "dm-1"}], H2),
-    %% H3 = report_disk2([{<<"xvdb">>, "xvdb"},
-    %%                    {<<"xvdc">>, "xvdc"},
-    %%                    {<<"raid0">>, "md127"}], H2),
+    DiskList = 
+	case get(disks) of
+	    undefined ->
+		Disks = determine_disks(),
+		put(disks, Disks),
+		Disks;
+	    Disks ->
+		Disks
+	end,
+
+    {H3, D} = report_disk2(DiskList, H2),
     {_, V} = report_vmstat(H3),
     {_, M} = report_memory(H3),
-    %% H3 = try report_disk2(H2) catch _:_ -> H2 end,
     C = report_stats(riak_core_stat, all),
     R = report_stats(riak_kv_stat, all),
     Stats0 = L ++ Q ++ P ++ N ++ D ++ V ++ M ++ C ++ R,
@@ -203,6 +218,26 @@ collect(H0) ->
     %% catch print_down(Nodes),
     H3.
 
+%% this portion is meant to be run inside a VM instance running riak
+determine_disks() ->
+    DataDir = 
+	case application:get_env(riak_kv, storage_backend) of
+	    {ok, riak_kv_bitcask_backend} ->
+		{ok, Dir} = application:get_env(bitcask, data_root),
+		Dir;
+	    {ok, riak_kv_eleveldb_backend} ->
+		{ok, Dir} = application:get_env(eleveldb, data_root),
+		Dir;
+	    _ ->
+		error(unhandled_backend)
+	end,
+    Name0 = os:cmd("basename `df "++DataDir++
+		       " | tail -1 | awk '{print $1}'`"),
+    {Name, _} = lists:split(length(Name0)-1, Name0),
+    %% keep the old format just in case we need to extend this later.
+    [{Name, Name}].
+		
+		 
 report_queues(H) ->
     Max = lists:max([Len || Pid <- processes(),
                             {message_queue_len, Len} <- [process_info(Pid, message_queue_len)]]),
@@ -264,8 +299,8 @@ report_disk2(Name, Dev, LastStats, #history{rate=Rate}) ->
 	end,
     {Stats, Report}.
 
-append_atoms(Atom, Binary) ->
-    list_to_atom(binary_to_list(Binary) ++ 
+append_atoms(Atom, List) ->
+    list_to_atom(List ++ 
 		     "_" ++ atom_to_list(Atom)).
 
 report_memory(H) ->

src/rt_bench.erl

@@ -29,7 +29,7 @@ bench(Config, NodeList, TestName, Runners, Drop) ->
         _ -> ok
     end,
 
-    %% make a local config file, to be copied to a remote
+    %% make a lxoocal config file, to be copied to a remote
     %% loadgen. They're named separately because for simplicity, we
     %% use network operations even for local load generation
 
@@ -48,7 +48,7 @@ bench(Config, NodeList, TestName, Runners, Drop) ->
     BBDir = rt_config:get(basho_bench),
     GenList =
 	[begin
-	     G = lists:nth((C rem 2), LoadGens),
+	     G = lists:nth(C, LoadGens),
 	     {G, C}
 	 end
 	 || C <- lists:seq(1, Runners)],
@@ -70,7 +70,8 @@ bench(Config, NodeList, TestName, Runners, Drop) ->
                     lager:info("Spawning remote basho_bench w/ ~p on ~p",
                                [Cmd, LG]),
                     {0, R} = rtssh:ssh_cmd(LG, Cmd, false),
-                    lager:info("bench run finished, returned ~p", [R]),
+                    lager:info("bench run finished, on ~p returned ~p", 
+			       [LG, R]),
                     {0, _} = rtssh:ssh_cmd(LG, "rm -r "++BBTmp++"/"),
 		    Owner ! {done, ok}
                 catch

src/rtperf.erl

@@ -237,7 +237,7 @@ deploy_nodes(NodeConfig, Hosts) ->
             lists:zip(lists:zip(Nodes, Hosts), Configs)),
     timer:sleep(500),
 
-    case rt_config:get(cuttle, true) of
+    case rt_config:get(perf_cuttle, true) of
         false ->
             rt:pmap(fun({Node, Host}) ->
                             Config = [{riak_api,
@@ -391,11 +391,11 @@ maybe_prepop(Hosts, BinSize, SetSize) ->
 
 	    %% drop the cache
             rt_bench:bench(PrepopConfig, Hosts, PrepopName, 
-                           length(rt_config:get(perf_loadgens, [1])), true),
+                           1, true),
 
-            timer:sleep(timer:minutes(1)+timer:seconds(30)),
 	    stop_data_collectors(PPids),
-            collect_test_data(Hosts, PrepopName);
+            collect_test_data(Hosts, PrepopName),
+            timer:sleep(timer:minutes(1)+timer:seconds(30));
         false ->
             ok
     end.
@@ -423,7 +423,7 @@ standard_config(NodeCount) ->
 
 standard_config(NodeCount, AAE) ->
     Backend = rt_config:get(rt_backend, undefined),
-    Fish = rt_config:get(cuttle, true),
+    Fish = rt_config:get(perf_cuttle, true),
     RingSize = rt:nearest_ringsize(NodeCount),
     mk_std_conf(Backend, Fish, RingSize, AAE).
 
@@ -468,8 +468,8 @@ mk_std_conf(riak_kv_eleveldb_backend, true, Ring, AAE0) ->
      ]};
 mk_std_conf(_, false, Ring, AAE) ->
     [{riak_core,
-      {handoff_concurrency, 16},
-      [{ring_creation_size, Ring}]},
+      [{handoff_concurrency, 16},
+       {ring_creation_size, Ring}]},
      {riak_kv,
       [{anti_entropy,{AAE, []}}]}
     ];