Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Newer
Older
100644 301 lines (268 sloc) 10.625 kB
90f6ce8 @argv0 initial import
argv0 authored
1 %% -------------------------------------------------------------------
2 %%
3 %% riak_core: Core Riak Application
4 %%
5 %% Copyright (c) 2007-2010 Basho Technologies, Inc. All Rights Reserved.
6 %%
7 %% This file is provided to you under the Apache License,
8 %% Version 2.0 (the "License"); you may not use this file
9 %% except in compliance with the License. You may obtain
10 %% a copy of the License at
11 %%
12 %% http://www.apache.org/licenses/LICENSE-2.0
13 %%
14 %% Unless required by applicable law or agreed to in writing,
15 %% software distributed under the License is distributed on an
16 %% "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
17 %% KIND, either express or implied. See the License for the
18 %% specific language governing permissions and limitations
19 %% under the License.
20 %%
21 %% -------------------------------------------------------------------
22
23 %% @doc A simple Erlang implementation of vector clocks as inspired by Lamport logical clocks.
24 %%
25 %% @reference Leslie Lamport (1978). "Time, clocks, and the ordering of events
26 %% in a distributed system". Communications of the ACM 21 (7): 558-565.
27 %%
28 %% @reference Friedemann Mattern (1988). "Virtual Time and Global States of
29 %% Distributed Systems". Workshop on Parallel and Distributed Algorithms:
30 %% pp. 215-226
31
32 -module(vclock).
33
34 -author('Justin Sheehy <justin@basho.com>').
35 -author('Andy Gross <andy@basho.com>').
36
37 -export([fresh/0,descends/2,merge/1,get_counter/2,get_timestamp/2,
97231e0 Add vclock:increment variant that accepts a timestamp.
Jon Meredith authored
38 increment/2,increment/3,all_nodes/1,equal/2,prune/3,timestamp/0]).
90f6ce8 @argv0 initial import
argv0 authored
39
ec82f3a @argv0 dialyzer: riak_core fixes
argv0 authored
40 -ifdef(TEST).
90f6ce8 @argv0 initial import
argv0 authored
41 -include_lib("eunit/include/eunit.hrl").
ec82f3a @argv0 dialyzer: riak_core fixes
argv0 authored
42 -endif.
90f6ce8 @argv0 initial import
argv0 authored
43
ec82f3a @argv0 dialyzer: riak_core fixes
argv0 authored
44 -export_type([vclock/0, timestamp/0, vclock_node/0]).
45
46 -opaque vclock() :: [vc_entry()].
90f6ce8 @argv0 initial import
argv0 authored
47 % The timestamp is present but not used, in case a client wishes to inspect it.
48 -type vc_entry() :: {vclock_node(), {counter(), timestamp()}}.
49
50 % Nodes can have any term() as a name, but they must differ from each other.
51 -type vclock_node() :: term().
8c5e75e @argv0 first pass at some better / more complete specs
argv0 authored
52 -type counter() :: non_neg_integer().
53 -type timestamp() :: non_neg_integer().
90f6ce8 @argv0 initial import
argv0 authored
54
55 % @doc Create a brand new vclock.
56 -spec fresh() -> vclock().
57 fresh() ->
58 [].
59
60 % @doc Return true if Va is a direct descendant of Vb, else false -- remember, a vclock is its own descendant!
ec82f3a @argv0 dialyzer: riak_core fixes
argv0 authored
61 -spec descends(Va :: vclock()|[], Vb :: vclock()|[]) -> boolean().
90f6ce8 @argv0 initial import
argv0 authored
62 descends(_, []) ->
63 % all vclocks descend from the empty vclock
64 true;
65 descends(Va, Vb) ->
66 [{NodeB, {CtrB, _T}}|RestB] = Vb,
67 CtrA =
68 case proplists:get_value(NodeB, Va) of
69 undefined ->
70 false;
71 {CA, _TSA} -> CA
72 end,
73 case CtrA of
74 false -> false;
75 _ ->
76 if
77 CtrA < CtrB ->
78 false;
79 true ->
80 descends(Va,RestB)
81 end
82 end.
83
84 % @doc Combine all VClocks in the input list into their least possible
85 % common descendant.
ec82f3a @argv0 dialyzer: riak_core fixes
argv0 authored
86 -spec merge(VClocks :: [vclock()]) -> vclock() | [].
90f6ce8 @argv0 initial import
argv0 authored
87 merge([]) -> [];
88 merge([SingleVclock]) -> SingleVclock;
89 merge([First|Rest]) -> merge(Rest, lists:keysort(1, First)).
90
91 merge([], NClock) -> NClock;
92 merge([AClock|VClocks],NClock) ->
93 merge(VClocks, merge(lists:keysort(1, AClock), NClock, [])).
94
95 merge([], [], AccClock) -> lists:reverse(AccClock);
96 merge([], [Left|Rest], AccClock) -> merge([], Rest, [Left|AccClock]);
97 merge(Left, [], AccClock) -> merge([], Left, AccClock);
98 merge(V=[{Node1,{Ctr1,TS1}}|VClock],
99 N=[{Node2,{Ctr2,TS2}}|NClock], AccClock) ->
100 if Node1 < Node2 ->
101 merge(VClock, N, [{Node1,{Ctr1,TS1}}|AccClock]);
102 Node1 > Node2 ->
103 merge(V, NClock, [{Node2,{Ctr2,TS2}}|AccClock]);
104 true ->
105 ({_Ctr,_TS} = C1) = if Ctr1 > Ctr2 -> {Ctr1,TS1};
106 true -> {Ctr2,TS2}
107 end,
108 merge(VClock, NClock, [{Node1,C1}|AccClock])
109 end.
110
111 % @doc Get the counter value in VClock set from Node.
112 -spec get_counter(Node :: vclock_node(), VClock :: vclock()) -> counter() | undefined.
113 get_counter(Node, VClock) ->
114 case proplists:get_value(Node, VClock) of
115 {Ctr, _TS} -> Ctr;
116 undefined -> undefined
117 end.
118
119 % @doc Get the timestamp value in a VClock set from Node.
120 -spec get_timestamp(Node :: vclock_node(), VClock :: vclock()) -> timestamp() | undefined.
121 get_timestamp(Node, VClock) ->
122 case proplists:get_value(Node, VClock) of
123 {_Ctr, TS} -> TS;
124 undefined -> undefined
125 end.
126
127 % @doc Increment VClock at Node.
128 -spec increment(Node :: vclock_node(), VClock :: vclock()) -> vclock().
129 increment(Node, VClock) ->
97231e0 Add vclock:increment variant that accepts a timestamp.
Jon Meredith authored
130 increment(Node, timestamp(), VClock).
131
132 % @doc Increment VClock at Node.
133 -spec increment(Node :: vclock_node(), IncTs :: timestamp(),
134 VClock :: vclock()) -> vclock().
135 increment(Node, IncTs, VClock) ->
90f6ce8 @argv0 initial import
argv0 authored
136 {{_Ctr, _TS}=C1,NewV} = case lists:keytake(Node, 1, VClock) of
137 false ->
97231e0 Add vclock:increment variant that accepts a timestamp.
Jon Meredith authored
138 {{1, IncTs}, VClock};
90f6ce8 @argv0 initial import
argv0 authored
139 {value, {_N, {C, _T}}, ModV} ->
97231e0 Add vclock:increment variant that accepts a timestamp.
Jon Meredith authored
140 {{C + 1, IncTs}, ModV}
90f6ce8 @argv0 initial import
argv0 authored
141 end,
142 [{Node,C1}|NewV].
143
97231e0 Add vclock:increment variant that accepts a timestamp.
Jon Meredith authored
144
90f6ce8 @argv0 initial import
argv0 authored
145 % @doc Return the list of all nodes that have ever incremented VClock.
146 -spec all_nodes(VClock :: vclock()) -> [vclock_node()].
147 all_nodes(VClock) ->
148 [X || {X,{_,_}} <- VClock].
149
97231e0 Add vclock:increment variant that accepts a timestamp.
Jon Meredith authored
150 % @doc Return a timestamp for a vector clock
151 -spec timestamp() -> timestamp().
90f6ce8 @argv0 initial import
argv0 authored
152 timestamp() ->
153 calendar:datetime_to_gregorian_seconds(erlang:universaltime()).
154
155 % @doc Compares two VClocks for equality.
156 % Not very fast.
157 -spec equal(VClockA :: vclock(), VClockB :: vclock()) -> boolean().
158 equal(VA,VB) ->
159 VSet1 = sets:from_list(VA),
160 VSet2 = sets:from_list(VB),
161 case sets:size(sets:subtract(VSet1,VSet2)) > 0 of
162 true -> false;
163 false ->
164 case sets:size(sets:subtract(VSet2,VSet1)) > 0 of
165 true -> false;
166 false -> true
167 end
168 end.
169
170 % @doc Possibly shrink the size of a vclock, depending on current age and size.
171 -spec prune(V::vclock(), Now::integer(), BucketProps::term()) -> vclock().
172 prune(V,Now,BucketProps) ->
724b951 @eriksoe vclock:prune(): In case of oldish vclock entries with identical times…
eriksoe authored
173 %% This sort need to be deterministic, to avoid spurious merge conflicts later.
174 %% We achieve this by using the node ID as secondary key.
175 SortV = lists:sort(fun({N1,{_,T1}},{N2,{_,T2}}) -> {T1,N1} < {T2,N2} end, V),
90f6ce8 @argv0 initial import
argv0 authored
176 prune_vclock1(SortV,Now,BucketProps).
177 % @private
178 prune_vclock1(V,Now,BProps) ->
179 case length(V) =< proplists:get_value(small_vclock,BProps) of
180 true -> V;
181 false ->
182 {_,{_,HeadTime}} = hd(V),
183 case (Now - HeadTime) < proplists:get_value(young_vclock,BProps) of
184 true -> V;
185 false -> prune_vclock1(V,Now,BProps,HeadTime)
186 end
187 end.
188 % @private
189 prune_vclock1(V,Now,BProps,HeadTime) ->
190 % has a precondition that V is longer than small and older than young
191 case length(V) > proplists:get_value(big_vclock,BProps) of
192 true -> prune_vclock1(tl(V),Now,BProps);
193 false ->
194 case (Now - HeadTime) > proplists:get_value(old_vclock,BProps) of
195 true -> prune_vclock1(tl(V),Now,BProps);
196 false -> V
197 end
198 end.
199
ec82f3a @argv0 dialyzer: riak_core fixes
argv0 authored
200 %% ===================================================================
201 %% EUnit tests
202 %% ===================================================================
203 -ifdef(TEST).
204
205 example_test() ->
206 A = vclock:fresh(),
207 B = vclock:fresh(),
208 A1 = vclock:increment(a, A),
209 B1 = vclock:increment(b, B),
210 true = vclock:descends(A1,A),
211 true = vclock:descends(B1,B),
212 false = vclock:descends(A1,B1),
213 A2 = vclock:increment(a, A1),
214 C = vclock:merge([A2, B1]),
215 C1 = vclock:increment(c, C),
216 true = vclock:descends(C1, A2),
217 true = vclock:descends(C1, B1),
218 false = vclock:descends(B1, C1),
219 false = vclock:descends(B1, A1),
220 ok.
221
90f6ce8 @argv0 initial import
argv0 authored
222 prune_small_test() ->
223 % vclock with less entries than small_vclock will be untouched
224 Now = riak_core_util:moment(),
225 OldTime = Now - 32000000,
226 SmallVC = [{<<"1">>, {1, OldTime}},
227 {<<"2">>, {2, OldTime}},
228 {<<"3">>, {3, OldTime}}],
229 Props = [{small_vclock,4}],
230 ?assertEqual(lists:sort(SmallVC), lists:sort(prune(SmallVC, Now, Props))).
231
232 prune_young_test() ->
233 % vclock with all entries younger than young_vclock will be untouched
234 Now = riak_core_util:moment(),
235 NewTime = Now - 1,
236 VC = [{<<"1">>, {1, NewTime}},
237 {<<"2">>, {2, NewTime}},
238 {<<"3">>, {3, NewTime}}],
239 Props = [{small_vclock,1},{young_vclock,1000}],
240 ?assertEqual(lists:sort(VC), lists:sort(prune(VC, Now, Props))).
241
242 prune_big_test() ->
243 % vclock not preserved by small or young will be pruned down to
244 % no larger than big_vclock entries
245 Now = riak_core_util:moment(),
246 NewTime = Now - 1000,
247 VC = [{<<"1">>, {1, NewTime}},
248 {<<"2">>, {2, NewTime}},
249 {<<"3">>, {3, NewTime}}],
250 Props = [{small_vclock,1},{young_vclock,1},
251 {big_vclock,2},{old_vclock,100000}],
252 ?assert(length(prune(VC, Now, Props)) =:= 2).
253
254 prune_old_test() ->
255 % vclock not preserved by small or young will be pruned down to
256 % no larger than big_vclock and no entries more than old_vclock ago
257 Now = riak_core_util:moment(),
258 NewTime = Now - 1000,
259 OldTime = Now - 100000,
260 VC = [{<<"1">>, {1, NewTime}},
261 {<<"2">>, {2, OldTime}},
262 {<<"3">>, {3, OldTime}}],
263 Props = [{small_vclock,1},{young_vclock,1},
264 {big_vclock,2},{old_vclock,10000}],
265 ?assert(length(prune(VC, Now, Props)) =:= 1).
266
7ce0ed9 @jonmeredith Added test for determinstic vclock pruning for nodes with same timest…
jonmeredith authored
267 prune_order_test() ->
268 % vclock with two nodes of the same timestamp will be pruned down
269 % to the same node
270 Now = riak_core_util:moment(),
271 OldTime = Now - 100000,
272 VC1 = [{<<"1">>, {1, OldTime}},
273 {<<"2">>, {2, OldTime}}],
274 VC2 = lists:reverse(VC1),
275 Props = [{small_vclock,1},{young_vclock,1},
276 {big_vclock,2},{old_vclock,10000}],
277 ?assertEqual(prune(VC1, Now, Props), prune(VC2, Now, Props)).
278
90f6ce8 @argv0 initial import
argv0 authored
279 accessor_test() ->
280 VC = [{<<"1">>, {1, 1}},
281 {<<"2">>, {2, 2}}],
282 ?assertEqual(1, get_counter(<<"1">>, VC)),
283 ?assertEqual(1, get_timestamp(<<"1">>, VC)),
284 ?assertEqual(2, get_counter(<<"2">>, VC)),
285 ?assertEqual(2, get_timestamp(<<"2">>, VC)),
286 ?assertEqual(undefined, get_counter(<<"3">>, VC)),
287 ?assertEqual(undefined, get_timestamp(<<"3">>, VC)),
288 ?assertEqual([<<"1">>, <<"2">>], all_nodes(VC)).
289
290 merge_test() ->
291 VC1 = [{<<"1">>, {1, 1}},
292 {<<"2">>, {2, 2}},
293 {<<"4">>, {4, 4}}],
294 VC2 = [{<<"3">>, {3, 3}},
295 {<<"4">>, {3, 3}}],
296 ?assertEqual([], merge(vclock:fresh())),
297 ?assertEqual([{<<"1">>,{1,1}},{<<"2">>,{2,2}},{<<"3">>,{3,3}},{<<"4">>,{4,4}}],
298 merge([VC1, VC2])).
ec82f3a @argv0 dialyzer: riak_core fixes
argv0 authored
299
300 -endif.
Something went wrong with that request. Please try again.