mod_offline_SUITE:messages_after_relogin sometimes fails

[arcusfelis]

MongooseIM version: b076e4a
Installed from: source
Erlang/OTP version: 20

in messages_after_relogin

====== Test name: messages_after_relogin                                                                                          
====== Reason:    {test_case_failed,                                                                                              
                      {has_stanzas_but_shouldnt,                                                                                  
                          {client,<<"alicE@localhost/res1">>,escalus_tcp,                                                         
                              <0.7743.0>,                                                                                         
                              [{event_manager,<0.7740.0>},       
                               {server,<<"localhost">>},                                                                          
                               {username,<<"alicE">>},                                                                            
                               {resource,<<"res1">>}],           
                              [{event_client,                                                                                     
                                   [{event_manager,<0.7740.0>},  
                                    {server,<<"localhost">>},                                                                     
                                    {username,<<"alicE">>},                                                                       
                                    {resource,<<"res1">>}]},     
                               {resource,<<"res1">>},            
                               {username,<<"alicE">>},           
                               {server,<<"localhost">>},         
                               {host,<<"localhost">>},           
                               {port,5222},                      
                               {auth,{escalus_auth,auth_plain}}, 
                               {wspath,undefined},               
                               {username,<<"alicE">>},           
                               {server,<<"localhost">>},         
                               {password,<<"matygrysa">>},       
                               {stream_id,<<"DA4E27366225FAC2">>}]},                                                              
                          [{xmlel,<<"presence">>,                
                               [{<<"from">>,<<"alicE@localhost/res1">>},                                                          
                                {<<"to">>,<<"alice@localhost/res1">>},                                                            
                                {<<"type">>,<<"unavailable">>}], 
                               []}]}} 

detected, during ODBC refactoring.

This version of test would work.

messages_after_relogin(Config) ->                                                    
    %% given                                                                         
    escalus:story(                                                                   
        Config, [{alice, 1}, {bob, 1}],                                              
        fun(User1, User2) ->                                                         
            user_blocks(User1, [User2])                                              
        end),                                                                        
    %% XXX Because alice can receive presence unavalable from alice                  
    %% XXX It's a potential bug, please test it.                                     
    %% XXX has_stanzas_but_shouldnt                                                  
    mongoose_helper:kick_everyone(),                                                 
    escalus:story(                                                                   
        Config, [{alice, 1}, {bob, 1}],                                              
        fun(User1, User2) ->                                                         
            message_is_not_delivered(User2, [User1], <<"Hey alice, are you there?">>),
            message_is_delivered(User1, [User1], <<"Hey bob, carpe diem!">>)         
        end).

But we actually need to fix bug, because this presence unavailable should be not pushed to the client.
I suspect this presence is sent from old alice process to a new one.

[arcusfelis]

Another similar bug even in 596fbc5

====== Suite FAILED: mod_blocking_SUITE (1 of 20 tests failed)
====== Test name: messages_arrive_after_unblock_and_relogin
====== Reason:    {test_case_failed,
                      {has_stanzas_but_shouldnt,
                          {client,<<"alicE@localhost/res1">>,escalus_tcp,
                              <0.8752.3>,
                              [{event_manager,<0.8751.3>},
                               {server,<<"localhost">>},
                               {username,<<"alicE">>},
                               {resource,<<"res1">>}],
                              [{event_client,
                                   [{event_manager,<0.8751.3>},
                                    {server,<<"localhost">>},
                                    {username,<<"alicE">>},
                                    {resource,<<"res1">>}]},
                               {resource,<<"res1">>},
                               {username,<<"alicE">>},
                               {server,<<"localhost">>},
                               {host,<<"localhost">>},
                               {port,5222},
                               {auth,{escalus_auth,auth_plain}},
                               {wspath,undefined},
                               {username,<<"alicE">>},
                               {server,<<"localhost">>},
                               {password,<<"matygrysa">>},
                               {stream_id,<<"55B52BA9532F2F00">>}]},
                          [{xmlel,<<"presence">>,
                               [{<<"from">>,<<"alicE@localhost/res1">>},
                                {<<"to">>,<<"alice@localhost/res1">>},
                                {<<"type">>,<<"unavailable">>}],
                               []}]}}

[arcusfelis]

One way to fix it is in mongoose to force keeping sids (session identifiers) in state of ejabberd_c2s together with presence subscription lists. But it would require more space. In theory, it would allow faster presence update broadcasts (woohoo, without reading session manager). Still. it would require more space to store (i.e. not cool).

Another way is "just" force to use mongoose_helper:kick_everyone() in tests. We can modify escalus story function, to check that resource was released properly. I.e. call mongooseim and wait for the old session process to die. But we would also need a subscription list of the old process (heh).

Another solution just for this set of tests is to use a different resource for each story. If we use "alice@localhost/res1_storyid", than we would avoid the old presences to begin with. Here "storyid" should be a unique value for each call of escalus:story/1. We can implement this "escalus"-wide, but in THEORY it can break some of our tests, that contain resources hardcoded. And some of the tests, that assume, that the resources are always the same.

[michalwski]

@arcusfelis I believe this was fixed already is that right? At least I didn't see it failing for a month already.

[fenek]

Haven't seen it failing as well. Closing for now.