Merge pull request #48 from basho/mra/merkle-cleanup

Add merkle library
This commit is contained in:
Scott Lystig Fritchie 2015-12-02 16:25:50 +09:00
commit e9b1134cd9
4 changed files with 375 additions and 0 deletions

1
.gitignore vendored
View file

@ -26,3 +26,4 @@ rel/machi
current_counterexample.eqc
foo*
typescript*
*.swp

View file

@ -0,0 +1,20 @@
%% machi merkle tree records
-record(naive, {
chunk_size = 1048576 :: pos_integer(), %% default 1 MB
recalc = true :: boolean(),
root :: 'undefined' | binary(),
lvl1 = [] :: [ binary() ],
lvl2 = [] :: [ binary() ],
lvl3 = [] :: [ binary() ],
leaves = [] :: [ { Offset :: pos_integer(),
Size :: pos_integer(),
Csum :: binary()} ]
}).
-record(mt, {
filename :: string(),
tree :: #naive{},
backend = 'naive' :: 'naive'
}).

156
src/machi_merkle_tree.erl Normal file
View file

@ -0,0 +1,156 @@
%% -------------------------------------------------------------------
%%
%% Copyright (c) 2007-2015 Basho Technologies, Inc. All Rights Reserved.
%%
%% This file is provided to you under the Apache License,
%% Version 2.0 (the "License"); you may not use this file
%% except in compliance with the License. You may obtain
%% a copy of the License at
%%
%% http://www.apache.org/licenses/LICENSE-2.0
%%
%% Unless required by applicable law or agreed to in writing,
%% software distributed under the License is distributed on an
%% "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
%% KIND, either express or implied. See the License for the
%% specific language governing permissions and limitations
%% under the License.
%%
%% -------------------------------------------------------------------
%% @doc Creates a Merkle tree per file based on the checksum data for
%% a given data file.
%%
%% The `naive' implementation representation is:
%%
%% `<<Length:64, Offset:32, 0>>' for unwritten bytes
%% `<<Length:64, Offset:32, 1>>' for trimmed bytes
%% `<<Length:64, Offset:32, Csum/binary>>' for written bytes
%%
%% The tree feeds these leaf nodes into hashes representing chunks of a minimum
%% size of at least 1024 KB (1 MB), but if the file size is larger, we will try
%% to get about 100 chunks for the first rollup "Level 1." We aim for around 10
%% hashes at level 2, and then 2 hashes level 3 and finally the root.
-module(machi_merkle_tree).
-include("machi.hrl").
-include("machi_merkle_tree.hrl").
-ifdef(TEST).
-compile(export_all).
-else.
-export([
open/2,
open/3,
tree/1,
filename/1,
diff/2
]).
-endif.
-define(TRIMMED, <<1>>).
-define(UNWRITTEN, <<0>>).
-define(NAIVE_ENCODE(Offset, Size, Data), <<Offset:64/unsigned-big, Size:32/unsigned-big, Data/binary>>).
-define(MINIMUM_CHUNK, 1048576). %% 1024 * 1024
-define(LEVEL_SIZE, 10).
-define(H, sha).
%% public API
open(Filename, DataDir) ->
open(Filename, DataDir, naive).
open(Filename, DataDir, Type) ->
Tree = load_filename(Filename, DataDir, Type),
{ok, #mt{ filename = Filename, tree = Tree, backend = Type}}.
tree(#mt{ tree = T, backend = naive }) ->
case T#naive.recalc of
true -> build_tree(T);
false -> T
end.
filename(#mt{ filename = F }) -> F.
diff(#mt{backend = naive, tree = T1}, #mt{backend = naive, tree = T2}) ->
case T1#naive.root == T2#naive.root of
true -> same;
false -> naive_diff(T1, T2)
end;
diff(_, _) -> error(badarg).
%% private
% @private
load_filename(Filename, DataDir, naive) ->
{Last, M} = do_load(Filename, DataDir, fun insert_csum_naive/2, []),
ChunkSize = max(?MINIMUM_CHUNK, Last div 100),
T = #naive{ leaves = lists:reverse(M), chunk_size = ChunkSize, recalc = true },
build_tree(T).
do_load(Filename, DataDir, FoldFun, AccInit) ->
CsumFile = machi_util:make_checksum_filename(DataDir, Filename),
{ok, T} = machi_csum_table:open(CsumFile, []),
Acc = machi_csum_table:foldl_chunks(FoldFun, {0, AccInit}, T),
ok = machi_csum_table:close(T),
Acc.
% @private
insert_csum_naive({Last, Size, _Csum}=In, {Last, MT}) ->
%% no gap
{Last+Size, update_acc(In, MT)};
insert_csum_naive({Offset, Size, _Csum}=In, {Last, MT}) ->
Hole = Offset - Last,
MT0 = update_acc({Last, Hole, unwritten}, MT),
{Offset+Size, update_acc(In, MT0)}.
% @private
update_acc({Offset, Size, unwritten}, MT) ->
[ {Offset, Size, ?NAIVE_ENCODE(Offset, Size, ?UNWRITTEN)} | MT ];
update_acc({Offset, Size, trimmed}, MT) ->
[ {Offset, Size, ?NAIVE_ENCODE(Offset, Size, ?TRIMMED)} | MT ];
update_acc({Offset, Size, <<_Tag:8, Csum/binary>>}, MT) ->
[ {Offset, Size, ?NAIVE_ENCODE(Offset, Size, Csum)} | MT ].
build_tree(MT = #naive{ leaves = L, chunk_size = ChunkSize }) ->
Lvl1s = build_level_1(ChunkSize, L, 1, [ crypto:hash_init(?H) ]),
Mod2 = length(Lvl1s) div ?LEVEL_SIZE,
Lvl2s = build_int_level(Mod2, Lvl1s, 1, [ crypto:hash_init(?H) ]),
Mod3 = length(Lvl2s) div 2,
Lvl3s = build_int_level(Mod3, Lvl2s, 1, [ crypto:hash_init(?H) ]),
Root = build_root(Lvl3s, crypto:hash_init(?H)),
MT#naive{ root = Root, lvl1 = Lvl1s, lvl2 = Lvl2s, lvl3 = Lvl3s, recalc = false }.
build_root([], Ctx) ->
crypto:hash_final(Ctx);
build_root([H|T], Ctx) ->
build_root(T, crypto:hash_update(Ctx, H)).
build_int_level(_Mod, [], _Cnt, [ Ctx | Rest ]) ->
lists:reverse( [ crypto:hash_final(Ctx) | Rest ] );
build_int_level(Mod, [H|T], Cnt, [ Ctx | Rest ]) when Cnt rem Mod == 0 ->
NewCtx = crypto:hash_init(?H),
build_int_level(Mod, T, Cnt + 1, [ crypto:hash_update(NewCtx, H), crypto:hash_final(Ctx) | Rest ]);
build_int_level(Mod, [H|T], Cnt, [ Ctx | Rest ]) ->
build_int_level(Mod, T, Cnt+1, [ crypto:hash_update(Ctx, H) | Rest ]).
build_level_1(_Size, [], _Multiple, [ Ctx | Rest ]) ->
lists:reverse([ crypto:hash_final(Ctx) | Rest ]);
build_level_1(Size, [{Pos, Len, Hash}|T], Multiple, [ Ctx | Rest ])
when ( Pos + Len ) > ( Size * Multiple ) ->
NewCtx = crypto:hash_init(?H),
build_level_1(Size, T, Multiple+1,
[ crypto:hash_update(NewCtx, Hash), crypto:hash_final(Ctx) | Rest ]);
build_level_1(Size, [{Pos, Len, Hash}|T], Multiple, [ Ctx | Rest ])
when ( Pos + Len ) =< ( Size * Multiple ) ->
build_level_1(Size, T, Multiple, [ crypto:hash_update(Ctx, Hash) | Rest ]).
naive_diff(#naive{lvl1 = L1}, #naive{lvl1=L2, chunk_size=CS2}) ->
Set1 = gb_sets:from_list(lists:zip(lists:seq(1, length(L1)), L1)),
Set2 = gb_sets:from_list(lists:zip(lists:seq(1, length(L2)), L2)),
%% The byte ranges in list 2 that do not match in list 1
%% Or should we do something else?
[ {(X-1)*CS2, CS2, SHA} || {X, SHA} <- gb_sets:to_list(gb_sets:subtract(Set1, Set2)) ].

View file

@ -0,0 +1,198 @@
%% -------------------------------------------------------------------
%%
%% Copyright (c) 2007-2015 Basho Technologies, Inc. All Rights Reserved.
%%
%% This file is provided to you under the Apache License,
%% Version 2.0 (the "License"); you may not use this file
%% except in compliance with the License. You may obtain
%% a copy of the License at
%%
%% http://www.apache.org/licenses/LICENSE-2.0
%%
%% Unless required by applicable law or agreed to in writing,
%% software distributed under the License is distributed on an
%% "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
%% KIND, either express or implied. See the License for the
%% specific language governing permissions and limitations
%% under the License.
%%
%% -------------------------------------------------------------------
-module(machi_merkle_tree_test).
-compile([export_all]).
-include("machi_merkle_tree.hrl").
-include_lib("eunit/include/eunit.hrl").
-include_lib("kernel/include/file.hrl").
-define(GAP_CHANCE, 0.10).
%% unit tests
basic_test() ->
random:seed(os:timestamp()),
Fsz = choose_size() * 1024,
Filesize = max(Fsz, 10*1024*1024),
ChunkSize = max(1048576, Filesize div 100),
N = make_leaf_nodes(Filesize),
D0 = #naive{ leaves = N, chunk_size = ChunkSize, recalc = true },
T1 = machi_merkle_tree:build_tree(D0),
D1 = #naive{ leaves = tl(N), chunk_size = ChunkSize, recalc = true },
T2 = machi_merkle_tree:build_tree(D1),
?assertNotEqual(T1#naive.root, T2#naive.root),
?assertEqual(1, length(machi_merkle_tree:naive_diff(T1, T2))).
make_leaf_nodes(Filesize) ->
lists:reverse(
lists:foldl(fun(T, Acc) -> machi_merkle_tree:update_acc(T, Acc) end,
[],
generate_offsets(Filesize, 1024, []))
).
choose_int(Factor) ->
random:uniform(1024*Factor).
small_int() ->
choose_int(10).
medium_int() ->
choose_int(1024).
large_int() ->
choose_int(4096).
generate_offsets(Filesize, Current, Acc) when Current < Filesize ->
Length0 = choose_size(),
Length = case Length0 + Current > Filesize of
false -> Length0;
true -> Filesize - Current
end,
Data = term_to_binary(os:timestamp()),
Checksum = machi_util:make_tagged_csum(client_sha, machi_util:checksum_chunk(Data)),
Gap = maybe_gap(random:uniform()),
generate_offsets(Filesize, Current + Length + Gap, [ {Current, Length, Checksum} | Acc ]);
generate_offsets(_Filesize, _Current, Acc) ->
lists:reverse(Acc).
random_from_list(L) ->
N = random:uniform(length(L)),
lists:nth(N, L).
choose_size() ->
F = random_from_list([fun small_int/0, fun medium_int/0, fun large_int/0]),
F().
maybe_gap(Chance) when Chance < ?GAP_CHANCE ->
choose_size();
maybe_gap(_) -> 0.
%% Define or remove these ifdefs if benchmarking is desired.
-ifdef(BENCH).
generate_offsets(FH, Filesize, Current, Acc) when Current < Filesize ->
Length0 = choose_size(),
Length = case Length0 + Current > Filesize of
false -> Length0;
true -> Filesize - Current
end,
{ok, Data} = file:pread(FH, Current, Length),
Checksum = machi_util:make_tagged_csum(client_sha, machi_util:checksum_chunk(Data)),
Gap = maybe_gap(random:uniform()),
generate_offsets(FH, Filesize, Current + Length + Gap, [ {Current, Length, Checksum} | Acc ]);
generate_offsets(_FH, _Filesize, _Current, Acc) ->
lists:reverse(Acc).
make_offsets_from_file(Filename) ->
{ok, Info} = file:read_file_info(Filename),
Filesize = Info#file_info.size,
{ok, FH} = file:open(Filename, [read, raw, binary]),
Offsets = generate_offsets(FH, Filesize, 1024, []),
file:close(FH),
Offsets.
choose_filename() ->
random_from_list([
"def^c5ea7511-d649-47d6-a8c3-2b619379c237^1",
"jkl^b077eff7-b2be-4773-a73f-fea4acb8a732^1",
"stu^553fa47a-157c-4fac-b10f-2252c7d8c37a^1",
"vwx^ae015d68-7689-4c9f-9677-926c6664f513^1",
"yza^4c784dc2-19bf-4ac6-91f6-58bbe5aa88e0^1"
]).
make_csum_file(DataDir, Filename, Offsets) ->
Path = machi_util:make_checksum_filename(DataDir, Filename),
filelib:ensure_dir(Path),
{ok, MC} = machi_csum_table:open(Path, []),
lists:foreach(fun({Offset, Size, Checksum}) ->
machi_csum_table:write(MC, Offset, Size, Checksum) end,
Offsets),
machi_csum_table:close(MC).
test() ->
test(100).
test(N) ->
{ok, F} = file:open("results.txt", [raw, write]),
lists:foreach(fun(X) -> format_and_store(F, run_test(X)) end, lists:seq(1, N)).
format_and_store(F, {OffsetNum, {MTime, MSize}, {NTime, NSize}}) ->
S = io_lib:format("~w\t~w\t~w\t~w\t~w\n", [OffsetNum, MTime, MSize, NTime, NSize]),
ok = file:write(F, S).
run_test(C) ->
random:seed(os:timestamp()),
OffsetFn = "test/" ++ choose_filename(),
O = make_offsets_from_file(OffsetFn),
Fn = "csum_" ++ integer_to_list(C),
make_csum_file(".", Fn, O),
Osize = length(O),
{MTime, {ok, M}} = timer:tc(fun() -> machi_merkle_tree:open(Fn, ".", merklet) end),
{NTime, {ok, N}} = timer:tc(fun() -> machi_merkle_tree:open(Fn, ".", naive) end),
?assertEqual(Fn, machi_merkle_tree:filename(M)),
?assertEqual(Fn, machi_merkle_tree:filename(N)),
MTree = machi_merkle_tree:tree(M),
MSize = byte_size(term_to_binary(MTree)),
NTree = machi_merkle_tree:tree(N),
NSize = byte_size(term_to_binary(NTree)),
?assertEqual(same, machi_merkle_tree:diff(N, N)),
?assertEqual(same, machi_merkle_tree:diff(M, M)),
{Osize, {MTime, MSize}, {NTime, NSize}}.
torture_test(C) ->
Results = [ run_torture_test() || _ <- lists:seq(1, C) ],
{ok, F} = file:open("torture_results.txt", [raw, write]),
lists:foreach(fun({MSize, MTime, NSize, NTime}) ->
file:write(F, io_lib:format("~p\t~p\t~p\t~p\n",
[MSize, MTime, NSize, NTime]))
end, Results),
ok = file:close(F).
run_torture_test() ->
{NTime, N} = timer:tc(fun() -> naive_torture() end),
MSize = byte_size(term_to_binary(M)),
NSize = byte_size(term_to_binary(N)),
{MSize, MTime, NSize, NTime}.
naive_torture() ->
N = lists:foldl(fun(T, Acc) -> machi_merkle_tree:update_acc(T, Acc) end, [], torture_generator()),
T = #naive{ leaves = lists:reverse(N), chunk_size = 10010, recalc = true },
machi_merkle_tree:build_tree(T).
torture_generator() ->
[ {O, 1, crypto:hash(sha, term_to_binary(now()))} || O <- lists:seq(1024, 1000000) ].
-endif. % BENCH