syn_consistency_SUITE.erl 18 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508
  1. %% ==========================================================================================================
  2. %% Syn - A global process registry.
  3. %%
  4. %% Copyright (C) 2015, Roberto Ostinelli <roberto@ostinelli.net>.
  5. %% All rights reserved.
  6. %%
  7. %% The MIT License (MIT)
  8. %%
  9. %% Copyright (c) 2015 Roberto Ostinelli
  10. %%
  11. %% Permission is hereby granted, free of charge, to any person obtaining a copy
  12. %% of this software and associated documentation files (the "Software"), to deal
  13. %% in the Software without restriction, including without limitation the rights
  14. %% to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  15. %% copies of the Software, and to permit persons to whom the Software is
  16. %% furnished to do so, subject to the following conditions:
  17. %%
  18. %% The above copyright notice and this permission notice shall be included in
  19. %% all copies or substantial portions of the Software.
  20. %%
  21. %% THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  22. %% IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  23. %% FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  24. %% AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  25. %% LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  26. %% OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  27. %% THE SOFTWARE.
  28. -module(syn_consistency_SUITE).
  29. %% callbacks
  30. -export([all/0]).
  31. -export([init_per_suite/1, end_per_suite/1]).
  32. -export([groups/0, init_per_group/2, end_per_group/2]).
  33. -export([init_per_testcase/2, end_per_testcase/2]).
  34. %% tests
  35. -export([
  36. two_nodes_netsplit_when_there_are_no_conflicts/1,
  37. two_nodes_netsplit_kill_resolution_when_there_are_conflicts/1,
  38. two_nodes_netsplit_callback_resolution_when_there_are_conflicts/1
  39. ]).
  40. -export([
  41. three_nodes_netsplit_kill_resolution_when_there_are_conflicts/1
  42. ]).
  43. %% internal
  44. -export([process_reply_main/0]).
  45. -export([conflicting_process_callback_dummy/3]).
  46. -export([process_exit_callback_dummy/4]).
  47. %% include
  48. -include_lib("common_test/include/ct.hrl").
  49. %% ===================================================================
  50. %% Callbacks
  51. %% ===================================================================
  52. %% -------------------------------------------------------------------
  53. %% Function: all() -> GroupsAndTestCases | {skip,Reason}
  54. %% GroupsAndTestCases = [{group,GroupName} | TestCase]
  55. %% GroupName = atom()
  56. %% TestCase = atom()
  57. %% Reason = term()
  58. %% -------------------------------------------------------------------
  59. all() ->
  60. [
  61. {group, two_nodes_netsplits},
  62. {group, three_nodes_netsplits}
  63. ].
  64. %% -------------------------------------------------------------------
  65. %% Function: groups() -> [Group]
  66. %% Group = {GroupName,Properties,GroupsAndTestCases}
  67. %% GroupName = atom()
  68. %% Properties = [parallel | sequence | Shuffle | {RepeatType,N}]
  69. %% GroupsAndTestCases = [Group | {group,GroupName} | TestCase]
  70. %% TestCase = atom()
  71. %% Shuffle = shuffle | {shuffle,{integer(),integer(),integer()}}
  72. %% RepeatType = repeat | repeat_until_all_ok | repeat_until_all_fail |
  73. %% repeat_until_any_ok | repeat_until_any_fail
  74. %% N = integer() | forever
  75. %% -------------------------------------------------------------------
  76. groups() ->
  77. [
  78. {two_nodes_netsplits, [shuffle], [
  79. two_nodes_netsplit_when_there_are_no_conflicts,
  80. two_nodes_netsplit_kill_resolution_when_there_are_conflicts,
  81. two_nodes_netsplit_callback_resolution_when_there_are_conflicts
  82. ]},
  83. {three_nodes_netsplits, [shuffle], [
  84. three_nodes_netsplit_kill_resolution_when_there_are_conflicts
  85. ]}
  86. ].
  87. %% -------------------------------------------------------------------
  88. %% Function: init_per_suite(Config0) ->
  89. %% Config1 | {skip,Reason} |
  90. %% {skip_and_save,Reason,Config1}
  91. %% Config0 = Config1 = [tuple()]
  92. %% Reason = term()
  93. %% -------------------------------------------------------------------
  94. init_per_suite(Config) ->
  95. %% init
  96. SlaveNodeShortName = syn_slave,
  97. %% start slave
  98. {ok, SlaveNode} = syn_test_suite_helper:start_slave(SlaveNodeShortName),
  99. %% config
  100. [
  101. {slave_node_short_name, SlaveNodeShortName},
  102. {slave_node, SlaveNode}
  103. | Config
  104. ].
  105. %% -------------------------------------------------------------------
  106. %% Function: end_per_suite(Config0) -> void() | {save_config,Config1}
  107. %% Config0 = Config1 = [tuple()]
  108. %% -------------------------------------------------------------------
  109. end_per_suite(Config) ->
  110. %% get slave node name
  111. SlaveNodeShortName = proplists:get_value(slave_node_short_name, Config),
  112. %% stop slave
  113. syn_test_suite_helper:stop_slave(SlaveNodeShortName).
  114. %% -------------------------------------------------------------------
  115. %% Function: init_per_group(GroupName, Config0) ->
  116. %% Config1 | {skip,Reason} |
  117. %% {skip_and_save,Reason,Config1}
  118. %% GroupName = atom()
  119. %% Config0 = Config1 = [tuple()]
  120. %% Reason = term()
  121. %% -------------------------------------------------------------------
  122. init_per_group(three_nodes_netsplits, Config) ->
  123. %% init
  124. SlaveNode2ShortName = syn_slave_2,
  125. %% start slave 2
  126. {ok, SlaveNode2} = syn_test_suite_helper:start_slave(SlaveNode2ShortName),
  127. %% config
  128. [
  129. {slave_node_2_short_name, SlaveNode2ShortName},
  130. {slave_node_2, SlaveNode2}
  131. | Config
  132. ];
  133. init_per_group(_GroupName, Config) -> Config.
  134. %% -------------------------------------------------------------------
  135. %% Function: end_per_group(GroupName, Config0) ->
  136. %% void() | {save_config,Config1}
  137. %% GroupName = atom()
  138. %% Config0 = Config1 = [tuple()]
  139. %% -------------------------------------------------------------------
  140. end_per_group(three_nodes_netsplits, Config) ->
  141. %% get slave node 2 name
  142. SlaveNode2ShortName = proplists:get_value(slave_node_2_short_name, Config),
  143. %% stop slave
  144. syn_test_suite_helper:stop_slave(SlaveNode2ShortName);
  145. end_per_group(_GroupName, _Config) -> ok.
  146. % ----------------------------------------------------------------------------------------------------------
  147. % Function: init_per_testcase(TestCase, Config0) ->
  148. % Config1 | {skip,Reason} | {skip_and_save,Reason,Config1}
  149. % TestCase = atom()
  150. % Config0 = Config1 = [tuple()]
  151. % Reason = term()
  152. % ----------------------------------------------------------------------------------------------------------
  153. init_per_testcase(_TestCase, Config) ->
  154. %% get slave
  155. SlaveNode = proplists:get_value(slave_node, Config),
  156. %% set schema location
  157. application:set_env(mnesia, schema_location, ram),
  158. rpc:call(SlaveNode, mnesia, schema_location, [ram]),
  159. %% return
  160. Config.
  161. % ----------------------------------------------------------------------------------------------------------
  162. % Function: end_per_testcase(TestCase, Config0) ->
  163. % void() | {save_config,Config1} | {fail,Reason}
  164. % TestCase = atom()
  165. % Config0 = Config1 = [tuple()]
  166. % Reason = term()
  167. % ----------------------------------------------------------------------------------------------------------
  168. end_per_testcase(_TestCase, Config) ->
  169. %% get slave
  170. SlaveNode = proplists:get_value(slave_node, Config),
  171. syn_test_suite_helper:clean_after_test(SlaveNode).
  172. %% ===================================================================
  173. %% Tests
  174. %% ===================================================================
  175. two_nodes_netsplit_when_there_are_no_conflicts(Config) ->
  176. %% get slave
  177. SlaveNode = proplists:get_value(slave_node, Config),
  178. CurrentNode = node(),
  179. %% start syn
  180. ok = syn:start(),
  181. ok = syn:init(),
  182. ok = rpc:call(SlaveNode, syn, start, []),
  183. ok = rpc:call(SlaveNode, syn, init, []),
  184. timer:sleep(100),
  185. %% start processes
  186. LocalPid = syn_test_suite_helper:start_process(),
  187. SlavePidLocal = syn_test_suite_helper:start_process(SlaveNode),
  188. SlavePidSlave = syn_test_suite_helper:start_process(SlaveNode),
  189. %% register
  190. ok = syn:register(local_pid, LocalPid),
  191. ok = syn:register(slave_pid_local, SlavePidLocal), %% slave registered on local node
  192. ok = rpc:call(SlaveNode, syn, register, [slave_pid_slave, SlavePidSlave]), %% slave registered on slave node
  193. timer:sleep(100),
  194. %% check tables
  195. 3 = mnesia:table_info(syn_processes_table, size),
  196. 3 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  197. LocalActiveReplicas = mnesia:table_info(syn_processes_table, active_replicas),
  198. 2 = length(LocalActiveReplicas),
  199. true = lists:member(SlaveNode, LocalActiveReplicas),
  200. true = lists:member(CurrentNode, LocalActiveReplicas),
  201. SlaveActiveReplicas = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, active_replicas]),
  202. 2 = length(SlaveActiveReplicas),
  203. true = lists:member(SlaveNode, SlaveActiveReplicas),
  204. true = lists:member(CurrentNode, SlaveActiveReplicas),
  205. %% simulate net split
  206. syn_test_suite_helper:disconnect_node(SlaveNode),
  207. timer:sleep(1000),
  208. %% check tables
  209. 1 = mnesia:table_info(syn_processes_table, size),
  210. [CurrentNode] = mnesia:table_info(syn_processes_table, active_replicas),
  211. %% reconnect
  212. syn_test_suite_helper:connect_node(SlaveNode),
  213. timer:sleep(1000),
  214. %% check tables
  215. 3 = mnesia:table_info(syn_processes_table, size),
  216. 3 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  217. LocalActiveReplicas2 = mnesia:table_info(syn_processes_table, active_replicas),
  218. 2 = length(LocalActiveReplicas2),
  219. true = lists:member(SlaveNode, LocalActiveReplicas2),
  220. true = lists:member(CurrentNode, LocalActiveReplicas2),
  221. SlaveActiveReplicas2 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, active_replicas]),
  222. 2 = length(SlaveActiveReplicas2),
  223. true = lists:member(SlaveNode, SlaveActiveReplicas2),
  224. true = lists:member(CurrentNode, SlaveActiveReplicas2),
  225. %% check processes
  226. LocalPid = syn:find_by_key(local_pid),
  227. SlavePidLocal = syn:find_by_key(slave_pid_local),
  228. SlavePidSlave = syn:find_by_key(slave_pid_slave),
  229. LocalPid = rpc:call(SlaveNode, syn, find_by_key, [local_pid]),
  230. SlavePidLocal = rpc:call(SlaveNode, syn, find_by_key, [slave_pid_local]),
  231. SlavePidSlave = rpc:call(SlaveNode, syn, find_by_key, [slave_pid_slave]),
  232. %% kill processes
  233. syn_test_suite_helper:kill_process(LocalPid),
  234. syn_test_suite_helper:kill_process(SlavePidLocal),
  235. syn_test_suite_helper:kill_process(SlavePidSlave).
  236. two_nodes_netsplit_kill_resolution_when_there_are_conflicts(Config) ->
  237. %% get slave
  238. SlaveNode = proplists:get_value(slave_node, Config),
  239. CurrentNode = node(),
  240. %% set process callback env variable
  241. ok = application:set_env(syn, process_exit_callback, [syn_consistency_SUITE, process_exit_callback_dummy]),
  242. ok = rpc:call(SlaveNode, application, set_env, [syn, process_exit_callback, [syn_consistency_SUITE, process_exit_callback_dummy]]),
  243. ok = application:set_env(syn, is_test, true),
  244. ok = rpc:call(SlaveNode, application, set_env, [syn, is_test, true]),
  245. %% start syn
  246. ok = syn:start(),
  247. ok = syn:init(),
  248. ok = rpc:call(SlaveNode, syn, start, []),
  249. ok = rpc:call(SlaveNode, syn, init, []),
  250. timer:sleep(100),
  251. %% register global process
  252. ResultPid = self(),
  253. global:register_name(syn_consistency_SUITE_result, ResultPid),
  254. %% start processes
  255. LocalPid = syn_test_suite_helper:start_process(),
  256. SlavePid = syn_test_suite_helper:start_process(SlaveNode),
  257. %% register
  258. ok = syn:register(conflicting_key, SlavePid),
  259. timer:sleep(100),
  260. %% check tables
  261. 1 = mnesia:table_info(syn_processes_table, size),
  262. 1 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  263. %% check process
  264. SlavePid = syn:find_by_key(conflicting_key),
  265. %% simulate net split
  266. syn_test_suite_helper:disconnect_node(SlaveNode),
  267. timer:sleep(1000),
  268. %% check tables
  269. 0 = mnesia:table_info(syn_processes_table, size),
  270. [CurrentNode] = mnesia:table_info(syn_processes_table, active_replicas),
  271. %% now register the local pid with the same key
  272. ok = syn:register(conflicting_key, LocalPid),
  273. %% check process
  274. LocalPid = syn:find_by_key(conflicting_key),
  275. %% reconnect
  276. syn_test_suite_helper:connect_node(SlaveNode),
  277. timer:sleep(1000),
  278. %% check tables
  279. 1 = mnesia:table_info(syn_processes_table, size),
  280. 1 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  281. %% check process
  282. FoundPid = syn:find_by_key(conflicting_key),
  283. true = lists:member(FoundPid, [LocalPid, SlavePid]),
  284. KilledPid = lists:nth(1, lists:delete(FoundPid, [LocalPid, SlavePid])),
  285. receive
  286. {exited, undefined, KilledPid, undefined, killed} -> ok
  287. after 3000 ->
  288. ok = process_exit_callback_was_not_called_from_local_node
  289. end,
  290. %% kill processes
  291. syn_test_suite_helper:kill_process(LocalPid),
  292. syn_test_suite_helper:kill_process(SlavePid),
  293. %% unregister
  294. global:unregister_name(syn_consistency_SUITE_result).
  295. two_nodes_netsplit_callback_resolution_when_there_are_conflicts(Config) ->
  296. %% get slave
  297. SlaveNode = proplists:get_value(slave_node, Config),
  298. CurrentNode = node(),
  299. %% load configuration variables from syn-test.config => this sets the conflicting_process_callback option
  300. syn_test_suite_helper:set_environment_variables(),
  301. syn_test_suite_helper:set_environment_variables(SlaveNode),
  302. %% start syn
  303. ok = syn:start(),
  304. ok = syn:init(),
  305. ok = rpc:call(SlaveNode, syn, start, []),
  306. ok = rpc:call(SlaveNode, syn, init, []),
  307. timer:sleep(100),
  308. %% start processes
  309. LocalPid = syn_test_suite_helper:start_process(fun process_reply_main/0),
  310. SlavePid = syn_test_suite_helper:start_process(SlaveNode, fun process_reply_main/0),
  311. %% register global process
  312. ResultPid = self(),
  313. global:register_name(syn_consistency_SUITE_result, ResultPid),
  314. %% register
  315. Meta = {some, meta, data},
  316. ok = syn:register(conflicting_key, SlavePid, Meta),
  317. timer:sleep(100),
  318. %% check tables
  319. 1 = mnesia:table_info(syn_processes_table, size),
  320. 1 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  321. %% check process
  322. SlavePid = syn:find_by_key(conflicting_key),
  323. %% simulate net split
  324. syn_test_suite_helper:disconnect_node(SlaveNode),
  325. timer:sleep(1000),
  326. %% check tables
  327. 0 = mnesia:table_info(syn_processes_table, size),
  328. [CurrentNode] = mnesia:table_info(syn_processes_table, active_replicas),
  329. %% now register the local pid with the same key
  330. ok = syn:register(conflicting_key, LocalPid, Meta),
  331. %% check process
  332. LocalPid = syn:find_by_key(conflicting_key),
  333. %% reconnect
  334. syn_test_suite_helper:connect_node(SlaveNode),
  335. timer:sleep(1000),
  336. %% check tables
  337. 1 = mnesia:table_info(syn_processes_table, size),
  338. 1 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  339. %% check process
  340. FoundPid = syn:find_by_key(conflicting_key),
  341. true = lists:member(FoundPid, [LocalPid, SlavePid]),
  342. %% check message received from killed pid
  343. KilledPid = lists:nth(1, lists:delete(FoundPid, [LocalPid, SlavePid])),
  344. receive
  345. {exited, KilledPid, Meta} -> ok
  346. after 2000 ->
  347. ok = conflicting_process_did_not_receive_message
  348. end,
  349. %% kill processes
  350. syn_test_suite_helper:kill_process(LocalPid),
  351. syn_test_suite_helper:kill_process(SlavePid),
  352. %% unregister
  353. global:unregister_name(syn_consistency_SUITE_result).
  354. three_nodes_netsplit_kill_resolution_when_there_are_conflicts(Config) ->
  355. %% get slaves
  356. SlaveNode = proplists:get_value(slave_node, Config),
  357. SlaveNode2 = proplists:get_value(slave_node_2, Config),
  358. CurrentNode = node(),
  359. %% start syn
  360. ok = syn:start(),
  361. ok = syn:init(),
  362. ok = rpc:call(SlaveNode, syn, start, []),
  363. ok = rpc:call(SlaveNode, syn, init, []),
  364. ok = rpc:call(SlaveNode2, syn, start, []),
  365. ok = rpc:call(SlaveNode2, syn, init, []),
  366. timer:sleep(100),
  367. %% start processes
  368. LocalPid = syn_test_suite_helper:start_process(),
  369. SlavePid = syn_test_suite_helper:start_process(SlaveNode),
  370. Slave2Pid = syn_test_suite_helper:start_process(SlaveNode2),
  371. %% register
  372. ok = syn:register(conflicting_key, SlavePid),
  373. ok = syn:register(slave_2_process, Slave2Pid),
  374. timer:sleep(100),
  375. %% check tables
  376. 2 = mnesia:table_info(syn_processes_table, size),
  377. 2 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  378. 2 = rpc:call(SlaveNode2, mnesia, table_info, [syn_processes_table, size]),
  379. %% check process
  380. SlavePid = syn:find_by_key(conflicting_key),
  381. %% simulate net split
  382. syn_test_suite_helper:disconnect_node(SlaveNode),
  383. timer:sleep(1000),
  384. %% check tables
  385. 1 = mnesia:table_info(syn_processes_table, size),
  386. 1 = rpc:call(SlaveNode2, mnesia, table_info, [syn_processes_table, size]),
  387. ActiveReplicaseDuringNetsplit = mnesia:table_info(syn_processes_table, active_replicas),
  388. true = lists:member(CurrentNode, ActiveReplicaseDuringNetsplit),
  389. true = lists:member(SlaveNode2, ActiveReplicaseDuringNetsplit),
  390. %% now register the local pid with the same conflicting key
  391. ok = syn:register(conflicting_key, LocalPid),
  392. %% check process
  393. LocalPid = syn:find_by_key(conflicting_key),
  394. %% reconnect
  395. syn_test_suite_helper:connect_node(SlaveNode),
  396. timer:sleep(1000),
  397. %% check tables
  398. 2 = mnesia:table_info(syn_processes_table, size),
  399. 2 = rpc:call(SlaveNode, mnesia, table_info, [syn_processes_table, size]),
  400. 2 = rpc:call(SlaveNode2, mnesia, table_info, [syn_processes_table, size]),
  401. %% check processes
  402. FoundPid = syn:find_by_key(conflicting_key),
  403. true = lists:member(FoundPid, [LocalPid, SlavePid]),
  404. Slave2Pid = syn:find_by_key(slave_2_process),
  405. Slave2Pid = rpc:call(SlaveNode, syn, find_by_key, [slave_2_process]),
  406. Slave2Pid = rpc:call(SlaveNode2, syn, find_by_key, [slave_2_process]),
  407. %% kill processes
  408. syn_test_suite_helper:kill_process(LocalPid),
  409. syn_test_suite_helper:kill_process(SlavePid),
  410. syn_test_suite_helper:kill_process(Slave2Pid).
  411. %% ===================================================================
  412. %% Internal
  413. %% ===================================================================
  414. process_reply_main() ->
  415. receive
  416. {shutdown, Meta} ->
  417. timer:sleep(500), %% wait for global processes to propagate
  418. global:send(syn_consistency_SUITE_result, {exited, self(), Meta})
  419. end.
  420. conflicting_process_callback_dummy(_Key, Pid, Meta) ->
  421. Pid ! {shutdown, Meta}.
  422. process_exit_callback_dummy(Key, Pid, Meta, Reason) ->
  423. global:send(syn_consistency_SUITE_result, {exited, Key, Pid, Meta, Reason}).