changes for 2i

bcrunner-nocuttle.sh

@@ -1,26 +0,0 @@
-#!/bin/bash
-
-test_name=$1
-bin_size=$2
-version=$3
-
-if [ -z $version -o -z $test_name -o -z $bin_size ]; then
-    echo "out"
-    exit 1
-fi
-
-./riak_test -c rtperf2 -t get_put -- --restart true --prepop true \
-    --run-time 120 --target-pct 120 --ram-size 48 \
-    --bin-size $bin_size --name $test_name --bin-type exponential \
-    --version $version --cuttle false
-
-./riak_test -c rtperf2 -t get_put -- --restart true --prepop true \
-    --run-time 120 --target-pct 70 --ram-size 48 \
-    --bin-size $bin_size --name $test_name --bin-type exponential \
-    --version $version --cuttle false
-
-./riak_test -c rtperf2 -t get_put -- --restart true --prepop true \
-    --run-time 120 --target-pct 20 --ram-size 48 \
-    --bin-size $bin_size --name $test_name --bin-type exponential \
-    --version $version --cuttle false
-

bcrunner.sh

@@ -3,24 +3,41 @@
 test_name=$1
 bin_size=$2
 version=$3
+perf_test=$4
+backend=$5
+cuttle=$5
 
 if [ -z $version -o -z $test_name -o -z $bin_size ]; then
     echo "out"
     exit 1
 fi
 
-./riak_test -c rtperf2 -t get_put -- --restart true --prepop true \
+if [ -z $perf_test ]; then
+    perf_test="get_put"
+fi
+
+if [ -z $backend ]; then
+    backend="riak_kv_bitcask_backend"
+fi
+
+if [ ! -z $cuttle ]; then
+    cuttle="--cuttle false"
+else 
+    cuttle=""
+fi
+
+./riak_test -c rtperf2 -t $perf_test -b $backend -- --restart true --prepop true \
     --run-time 120 --target-pct 120 --ram-size 48 \
     --bin-size $bin_size --name $test_name --bin-type exponential \
-    --version $version
+    --version $version $cuttle
 
-./riak_test -c rtperf2 -t get_put -- --restart true --prepop true \
+./riak_test -c rtperf2 -t $perf_test -b $backend -- --restart true --prepop true \
     --run-time 120 --target-pct 70 --ram-size 48 \
     --bin-size $bin_size --name $test_name --bin-type exponential \
-    --version $version
+    --version $version $cuttle
 
-./riak_test -c rtperf2 -t get_put -- --restart true --prepop true \
+./riak_test -c rtperf2 -t $perf_test -b $backend -- --restart true --prepop true \
     --run-time 120 --target-pct 20 --ram-size 48 \
     --bin-size $bin_size --name $test_name --bin-type exponential \
-    --version $version
+    --version $version $cuttle
 

perf/2iperf.erl

@@ -0,0 +1,47 @@
+-module('2iperf').
+-compile(export_all).
+-include_lib("eunit/include/eunit.hrl").
+
+-define(HARNESS, (rt_config:get(rt_harness))).
+
+confirm() ->
+    lager:info("entering get_put:confirm()"),
+    HostList = rt_config:get(rt_hostnames),
+    Count = length(HostList),
+    BinSize = rt_config:get(perf_bin_size),
+
+    Config = rtperf:standard_config(Count),
+
+    ok = rtperf:build_cluster(Config),
+
+    SetSize = rtperf:target_size(rt_config:get(perf_target_pct),
+                                 BinSize,
+                                 rt_config:get(perf_ram_size),
+                                 Count),
+    LoadConfig = [],
+        %% rt_bench:config(
+        %%   50,
+        %%   rt_config:get(perf_duration),
+        %%   HostList,
+        %%   {int_to_bin_bigendian, {truncated_pareto_int, SetSize}},
+        %%   rt_bench:valgen(rt_config:get(perf_bin_type), BinSize),
+        %%   %% 4:1 get/put
+        %%   [{get, 3}, {update, 1}]
+        %%  ),
+
+    TwoIConfig = 
+	rt_bench:config(
+	  max,
+	  rt_config:get(perf_duration),
+	  HostList,
+          {truncated_pareto_int, SetSize},
+          rt_bench:valgen(rt_config:get(perf_bin_type), BinSize),
+	  [{{query_pb, 100}, 5}, {{query_pb, 1000}, 1},
+	   {{put_pb, 2}, 1}, {get_pb, 5}],
+	  <<"testbucket">>, '2i'
+	 ),
+
+    ok = rtperf:maybe_prepop(HostList, BinSize, SetSize),
+
+    ok = rtperf:run_test(HostList, TwoIConfig, LoadConfig),
+    pass.

priv/reporting/report-2i.sh

@@ -0,0 +1,15 @@
+#!/usr/bin/env bash
+set -e
+
+if [ ! -d "$1/" ]; then
+    echo "a directory must be specified"
+    exit 1
+fi
+if [ "x$2" == "xtrue" ]; then
+    (cd $1; rm -f *-digest; escript ~/bin/riak-digest.escript)
+fi
+
+D1=`basename "$1"`
+
+#generate our comparison graph
+gnuplot -e "dir1=\"$1\"; outfilename=\"${D1}-report.png\";" priv/reporting/summarize2i.gpl

priv/reporting/riak-digest.escript

@@ -131,6 +131,10 @@ avg_items(L, Names) ->
                           Dicts),
          case Name of
              %% vnode gets and puts are a per-minute rolling window
+	     index_fsm_create ->
+                 (lists:sum(Vals)/length(Vals)) / 60;
+	     index_fsm_create_error ->
+                 (lists:sum(Vals)/length(Vals)) / 60;
              vnode_gets ->
                  (lists:sum(Vals)/length(Vals)) / 60;
              vnode_puts ->
@@ -155,6 +159,7 @@ winnow(Data0) ->
 
 strip_stats(Glob) ->
     Filter = [
+	      index_fsm_create, index_fsm_create_error,
 	      node_gets, node_puts,
 	      vnode_gets, vnode_puts,
               node_get_fsm_time_median,

priv/reporting/summarize2i.gpl

@@ -0,0 +1,119 @@
+#clean up the environment for interactive use
+unset multiplot 
+reset
+
+set terminal png font "/usr/share/fonts/truetype/liberation/LiberationSans-Regular.ttf,9" size 850,1100
+set output outfilename
+
+#set term x11 size 850, 1100
+
+set multiplot
+set grid 
+
+#set key below
+set tics out 
+
+set lmargin 12
+set rmargin 10
+
+rd = dir1."/".'rstats-digest'
+
+## graph system stats
+set size .5, .315
+
+## graph latencies
+set origin 0, .315
+
+#set xtics 10
+
+#set yrange [500:150000]
+#set y2range [1000:1100000]
+#set y2tics nomirror
+
+set xlabel "5 second intervals"
+set ylabel "usec"
+unset y2label
+
+plot rd using "node_put_fsm_time_95" with lines, \
+     rd using "node_put_fsm_time_99" with lines, \
+     rd using "node_get_fsm_time_95" with lines, \
+     rd using "node_get_fsm_time_99" with lines
+
+unset y2tics
+unset y2range
+
+## graph ops performance 
+set origin 0, .63
+
+#ymax = median(rd) + 1000
+
+#set yrange [0:50000]
+set ylabel "operations per node" 
+set y2label "errors" 
+set y2tics nomirror
+
+plot rd using "index_fsm_create" with lines, \
+     rd using "vnode_gets" with lines axis x1y2, \
+     rd using "vnode_puts" with lines axis x1y2
+
+## graph system stats
+set origin .48, 0
+
+set xlabel "5 second intervals"
+set y2tics nomirror
+
+plot rd using "memory_page_dirty" with lines, \
+     rd using "memory_page_writeback" with lines axis x1y2
+
+unset y2tics
+unset y2range
+
+## graph latencies
+set origin .48, .315
+#set yrange [500:150000]
+#set y2range [1000:1100000]
+#set y2tics nomirror
+
+#set xtics 10
+
+set xlabel "5 second intervals"
+set ylabel "usec"
+unset y2label
+
+plot rd using "node_get_fsm_time_median" with lines, \
+     rd using "node_put_fsm_time_median" with lines
+
+unset y2tics
+unset y2range
+
+## graph ops performance 
+set origin .48, .63
+
+#set yrange [0:50000]
+set ylabel "operations per node" 
+
+#hack to set the title for the whole graph
+set label dir1 at screen 0.5,0.97 center front
+
+plot rd using "message_queue_max" with lines, \
+     rd using "dropped_vnode_requests_total" with lines
+
+set origin 0, 0
+
+#set xtics 60
+
+set yrange [0:100]
+
+set xlabel "5 second intervals"
+set ylabel "percentage"
+
+plot rd using "cpu_utilization" with lines, \
+     rd using "cpu_iowait" with lines, \
+     rd using "disk_utilization" with lines, \
+     rd using "memory_utilization" with lines
+
+unset yrange
+unset y2tics
+
+unset multiplot
+reset 

report-2i.sh

@@ -0,0 +1 @@
+priv/reporting/report-2i.sh

src/observer.erl

@@ -35,7 +35,7 @@ watcher(Master, Nodes, {_Host, Port, _Dir} = Collector) ->
     case gen_tcp:listen(Port, [{active, false}, binary,
                    {packet, 2}]) of
     {ok, LSock} ->
-        Acceptor = spawn_link(?MODULE, lloop, [self(), LSock]),
+        Acceptor = spawn(?MODULE, lloop, [self(), LSock]),
         monitor(process, Master),
         Probes = [{Node, undefined} || Node <- Nodes],
         W = #watcher{nodes=Nodes,
@@ -483,7 +483,9 @@ load_modules_on_nodes(Modules, Nodes) ->
     [case code:get_object_code(Module) of
          {Module, Bin, File} ->
              %% rpc:multicall(Nodes, code, purge, [Module]),
-             rpc:multicall(Nodes, code, load_binary, [Module, File, Bin]);
+             {Ret, []} = rpc:multicall(Nodes, code, 
+				       load_binary, [Module, File, Bin]),
+	     [{module, observer}] = lists:usort(Ret);
          error ->
              error({no_object_code, Module})
      end || Module <- Modules].

src/rt_bench.erl

@@ -29,7 +29,7 @@ bench(Config, NodeList, TestName, Runners, Drop) ->
         _ -> ok
     end,
 
-    %% make a lxoocal config file, to be copied to a remote
+    %% make a local 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
 
@@ -114,11 +114,16 @@ config(Rate, Duration, NodeList, KeyGen,
            <<"testbucket">>, riakc_pb).
 
 config(Rate, Duration, NodeList, KeyGen,
-       ValGen, Operations, Bucket, Driver) ->
-    DriverBucket = append_atoms(Driver, '_bucket'),
+       ValGen, Operations, Bucket, Driver0) ->
+    {Driver, DriverB} = 
+	case Driver0 of
+	    '2i' -> {pb, riakc_pb};
+	    _ -> {Driver0, Driver0}
+	end,
+    DriverBucket = append_atoms(DriverB, '_bucket'),
     DriverIps = append_atoms(Driver, '_ips'),
-    DriverReplies = append_atoms(Driver, '_replies'),
-    DriverName = append_atoms(basho_bench_driver_, Driver),
+    DriverReplies = append_atoms(DriverB, '_replies'),
+    DriverName = append_atoms(basho_bench_driver_, Driver0),
     [
      {mode, {rate, Rate}},
      {duration, Duration},

src/rtperf.erl

@@ -81,7 +81,7 @@ run_test(HostList, TestBenchConfig, BaseBenchConfig) ->
 
     TestName = test_name(),
 
-    Base = maybe_start_base_load(BaseBenchConfig),
+    Base = maybe_start_base_load(BaseBenchConfig, HostList, TestName),
 
     rt_bench:bench(TestBenchConfig, HostList, TestName,
                    length(rt_config:get(perf_loadgens, [1]))),
@@ -322,16 +322,23 @@ start_data_collectors(Hosts) ->
     PrepDir = "/tmp/perf-"++OSPid,
     file:make_dir(PrepDir),
     {ok, Hostname} = inet:gethostname(),
-    observer:watch(Nodes, {Hostname, 65001, PrepDir}).
+    P = observer:watch(Nodes, {Hostname, 65001, PrepDir}),
+    lager:info("started data collector: ~p", [P]),
+    P.
 
 stop_data_collectors(Collector) ->
     Collector ! stop,
     ok.
 
-maybe_start_base_load([]) ->
-    none.
+maybe_start_base_load([], _, _) ->
+    none;
+maybe_start_base_load(Config, HostList, TestName) ->
+    spawn(fun() ->
+		  rt_bench:bench(Config, HostList, TestName++"_base",
+				 length(rt_config:get(perf_loadgens, [1])))
+	  end).
 
-maybe_stop_base_load(none) ->
+maybe_stop_base_load(_) -> %% should be none, but benches aren't stoppable rn.
     ok.
 
 %% need more sensible test names.