Commit graph

91 commits

Author SHA1 Message Date
Scott Lystig Fritchie
7433f6cf99 WIP: deprecate proj_proposed 2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
6fb3f55fee WIP: minor refactoring 2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
89f81e45b1 Add machi_partition_simulator:always_these_partitions(Parts), try a hard scenario
So, this is an interesting case where an asymmetric network parittion
can cause the current algorithm to cycle for several seconds, then one
participant X becomes less active (I'm not sure why), the other two
participants slowly come to an agreement, then X seems to wake up and
return everyone to the cycle/flapping loop.

    SET always_last_partitions ON ... we should see convergence to correct chains.
    16:35:03.986 c uses: [{epoch,321},{author,b},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:04.118 b uses: [{epoch,323},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{repair_airquote_done,{we_agree,321}},{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:04.492 c uses: [{epoch,324},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:04.520 b uses: [{epoch,325},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:04.583 a uses: [{epoch,326},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:04.894 c uses: [{epoch,327},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:04.922 b uses: [{epoch,328},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:05.291 a uses: [{epoch,329},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:05.296 c uses: [{epoch,330},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:05.324 b uses: [{epoch,331},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:05.830 c uses: [{epoch,332},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:06.023 a uses: [{epoch,333},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:06.128 b uses: [{epoch,334},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:06.342 c uses: [{epoch,335},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:06.530 b uses: [{epoch,336},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:06.734 a uses: [{epoch,337},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:06.746 c uses: [{epoch,338},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:06.932 b uses: [{epoch,339},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:07.267 c uses: [{epoch,340},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:07.334 b uses: [{epoch,341},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:07.460 a uses: [{epoch,342},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:07.669 c uses: [{epoch,343},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:07.736 b uses: [{epoch,344},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:08.165 a uses: [{epoch,345},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:08.194 c uses: [{epoch,346},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:08.541 b uses: [{epoch,347},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:08.702 c uses: [{epoch,348},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:08.894 a uses: [{epoch,349},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:08.944 b uses: [{epoch,350},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:09.212 c uses: [{epoch,351},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:09.346 b uses: [{epoch,352},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:09.598 a uses: [{epoch,353},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:09.614 c uses: [{epoch,354},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:09.748 b uses: [{epoch,355},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:10.135 c uses: [{epoch,356},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:10.150 b uses: [{epoch,357},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
    16:35:10.329 a uses: [{epoch,358},{author,a},{upi,[a]},{repair,[c]},{down,[b]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,c]}]},{d2,[]}]
    16:35:10.537 c uses: [{epoch,359},{author,c},{upi,[b]},{repair,[a,c]},{down,[]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[a,b,c]}]},{d2,[]}]
    16:35:10.552 b uses: [{epoch,360},{author,b},{upi,[b,c]},{repair,[]},{down,[a]},{d,[{author_proc,react},{ps,[{b,a}]},{nodes_up,[b,c]}]},{d2,[]}]
2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
6893d8af52 Re-introduce the 1-way partition generation scheme of olde, default=oneway_partitions
This is a return to the old, possibly asymmetric/unidirectional network
partition simulation scheme.  PULSE testing so far for the
symmetric/bidirectional partitioning scheme (via the "islands" approach)
appears to be very stable, yay.

So, let's go back to the harder environment and see what happens!
2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
098524ea2d Change verbosity of react_to_env_C120() 2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
61d3aafccb Fix non-TEST compilation errors 2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
e0d4dce8af Fix PULSE model to work around after-the-fact/retrospective sanity check limitation 2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
2471e61cc7 Fix PULSE model problem, yay! 2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
72adece716 WIP: EUnit and PULSE test fixing 2015-03-02 20:20:20 +09:00
Scott Lystig Fritchie
5d86b23851 PULSE: try to avoid false positives, add verbosity, tighten shrinking 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
6ded565f26 Add extra history to exception tuple in projection_transition_is_sane() 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
502afc7c19 Fix error in projection_transition_is_sane() 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
96f5b329c9 Tweaks for PULSE 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
39bee01936 Initial PULSE test for chain manager is done and ready for punishment 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
ed900c2a5f Fix up pid vs. atom name usage in chain manager 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
16a45660cc WIP: Initial PULSE test for chain manager 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
ac2af6d1ae Minor changes for initial PULSE testing 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
e72f239905 Remove spammy message but keep the TODO string in place 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
367f17154d Done with WIP: we have a fully automatable (though non-deterministic) test case 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
873aa749f9 WIP: stuck, fix think-o in unanimous_report() 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
c69a206039 WIP: stuck, need to add even MORE repairing list, before continuing 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
a36f23ee7a WIP: stuck, need to add repairing list before continuing with projection sanity check 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
8d9cabd214 Bring flowchart & code back into sync, yay! 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
30f5a84cea Clean up cruft, add more comments 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
32cfcccf34 First part of larger sanity test is now prototyped.
This is some brute-force-and-not-subtle hackery, but it looks like I've
got the basis for a test that a model checker (QuickCheck or Concuerror
or something else) can use for a good/bad check.

The following properties are examined (but not enforced):

* At each epoch, are each of the chains disjoint?  I.e. no single FLU
  is a member of different chains at the same epoch.

  This is a safety/sanity check.

* For each unique chain UPI list at each epoch, are all of the FLUs in that
  chain unanimous in their agreement:
    agreed_membership: all UPI FLUs agree about the UPI list
    not_agreed: the membership algorithm has not yet agreed on
                the UPI list

  This is not a safety/sanity check per se, but it can be useful input
  into a good safety check.

Some examples:

* At epoch 0, there is no agreement on UPI membership of the one [a,b,c]
  chain.
* At epoch 1, there is full agreement,
* At epoch 4, we're back to no agreement.
* At epoch 17, there's agreement on a small chain with UPI list=[a].
  (This agreement continues until epoch 216, but that history is omitted
  here.)

   [{0,
     {ok_disjoint,[{[a,b,c],
                    not_unique,0,
                    [<<159,215,105,140,29,151,142,2,162,90,225,209,10,102,119,
                       193,110,72,75,245>>,
                     <<213,46,129,248,23,50,210,247,145,68,65,112,232,101,28,56,
                       239,12,78,227>>,
                     <<230,146,66,183,10,218,57,29,233,166,108,176,118,109,
                       226,186,190,56,174,108>>]}]}},
    {1,{ok_disjoint,[{agreed_membership,[a,b,c]}]}},
    {4,
     {ok_disjoint,[{not_unique,[a,b,c],
                               [not_in_this_epoch,
                                <<208,227,221,233,254,160,36,134,252,106,
                                  124,192,101,171,168,68,169,55,2,54>>]}]}},
    {6,
     {ok_disjoint,[{not_unique,[a,b,c],
                               [not_in_this_epoch,
                                <<191,47,203,143,195,230,71,162,39,132,188,
                                  128,64,39,18,9,73,148,207,220>>]}]}},
    {17,{ok_disjoint,[{agreed_membership,[a]}]}},
    {24,{ok_disjoint,[{agreed_membership,[a]}]}},
    [...]

Starting at epoch 419, the network stabilized, but not fully,
into two "islands" of servers, a alone and b&c together.
At epoch 486, the network is fully stabilized with the same network
partition.  We see rapid convergence to two chains, [a] and [b,c].

    {419,{ok_disjoint,[{agreed_membership,[a]}]}},
    {425,{ok_disjoint,[{agreed_membership,[b]}]}},
    {436,{ok_disjoint,[{agreed_membership,[b]}]}},
    {442,{ok_disjoint,[{agreed_membership,[b]}]}},
    {444,{ok_disjoint,[{agreed_membership,[b]}]}},
    {454,{ok_disjoint,[{agreed_membership,[b]}]}},
    {456,{ok_disjoint,[{agreed_membership,[b]}]}},
    {458,{ok_disjoint,[{agreed_membership,[b]}]}},
    {463,{ok_disjoint,[{agreed_membership,[b]}]}},
    {468,{ok_disjoint,[{agreed_membership,[b]}]}},
    {479,{ok_disjoint,[{agreed_membership,[b]}]}},
    {482,{ok_disjoint,[{agreed_membership,[b]}]}},
    {486,{ok_disjoint,[{agreed_membership,[a]}]}},
    {488,{ok_disjoint,[{agreed_membership,[b]}]}},
    {490,{ok_disjoint,[{agreed_membership,[b,c]}]}},
    {492,{ok_disjoint,[{agreed_membership,[b,c]}]}}]

foo
2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
4def1ad026 Move test code from machi_chain_manager1.erl -> machi_chain_manager1_test.erl 2015-03-02 20:20:19 +09:00
Scott Lystig Fritchie
a2f087181e Change C210's sleep time to be proportional/ranked to member's rank 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
c0ef199c6f Hey, I think this is finally working, hooray! 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
258bce84d1 WIP: remove lots of debugging cruft 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
4fac90f5a9 Fix end-of-repair logic by querying repair target for epoch sync 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
0a6b8268fb Fix major error in rank_projection(), silly me 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
fdca511385 Fix broken machi_partition_simulator.erl, derp 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
0b88a12c16 WIP: Debugging cruft, egadz, but improving (see below)
So, this still pops up occasionally:

    % rebar skip_deps=true -v eunit suites=machi_flu0_test,machi_chain_manager1
    [...]
    a private: [{epoch,223},{author,a},{upi,[a,b]},{repair,[]},{down,[c]},{d,[{author_proc,react},{nodes_up,[a,b]}]},{d2,[{up_nodz,[a,b]},{hooray,{v2,{2014,11,3},{20,19,57}}}]}]
    b private: [{epoch,224},{author,b},{upi,[b,a]},{repair,[]},{down,[c]},{d,[{author_proc,react},{nodes_up,[a,b]}]},{d2,[{up_nodz,[a,b]},{hooray,{v2,{2014,11,3},{20,19,57}}}]}]
    c private: [{epoch,191},{author,c},{upi,[c]},{repair,[]},{down,[a,b]},{d,[{author_proc,react},{nodes_up,[c]}]},{d2,[{up_nodz,[c]},{hooray,{v2,{2014,11,3},{20,19,57}}}]}]

The mis-ordering between [a,b] and [b,a] happens after the partition settled
on the islands of [a,b] and [c].

    { c100 , ? LINE , _AnyOtherReturnValue } {c100,734,
                                          {err,error,
                                           {badmatch,[a,b]},
                                           from,
                                           [{epoch,70},
                                            {author,a},
                                            {upi,[a]},
                                            {repair,[b]},
                                            {down,[c]},
                                            {d,
                                             [{author_proc,react},
                                              {nodes_up,[a,b]}]},
                                            {d2,[]}],
                                           to,
                                           [{epoch,194},
                                            {author,b},
                                            {upi,[b,a]},
                                            {repair,[]},
                                            {down,[c]},
                                            {d,
                                             [{author_proc,react},
                                              {nodes_up,[a,b]}]},
                                            {d2,[]}],
                                           relative_to,a,stack,[...]
2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
5d0eed865a Duh, fix really stupid think-o bug in perhaps_call(), oi oi oi 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
a94374cc8c Add machi_partition_simulator.erl + refactor to use it 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
4d3a9ed757 WIP: Per notes: change unanimous test @ A20. Improvement! 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
201108ec5b Temp WIP: Change network partitions to be bi-directional only 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
6ef46a3464 Temp WIP: I am going to sever the connection between the flowchart and the code. TODO
That diagram is really valuable, but it also takes a long time
to make any kind of edit; the process is too slow.  This is a todo
item a reminder that the flowchart is important documentation and
must be brought back into sync with the code soon.
2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
9404e954e7 WIP: chain mgr clutter, trying to debug infinite loop 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
fd1b4363b9 WIP: chain manager getting better, but occasionally gets infinite loop (II) 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
67f94d1cff WIP: chain manager getting better, but occasionally gets infinite loop 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
2f54525422 Fix chain mgmt flowchart A40 conditions (II) 2015-03-02 20:20:18 +09:00
Scott Lystig Fritchie
c1fd3df35d Fix chain mgmt flowchart A40 conditions 2015-03-02 20:20:17 +09:00
Scott Lystig Fritchie
12d8a94497 Add reminder about chain manager init bootstrapping TODO 2015-03-02 20:20:17 +09:00
Scott Lystig Fritchie
064b637d81 Remove docs/machi/flowchart-machi-chain-mgmt1.jpg 2015-03-02 20:20:17 +09:00
Scott Lystig Fritchie
83e4937658 Chain manager projection store flowchart implemented & passes smoke test! 2015-03-02 20:20:17 +09:00
Scott Lystig Fritchie
8faa1404c6 Remove unused prev_epoch_num and prev_epoch_csum 2015-03-02 20:20:17 +09:00
Scott Lystig Fritchie
9af576d753 WIP: broken, don't use 2015-03-02 20:20:17 +09:00
Scott Lystig Fritchie
cbc5260e93 WIP: Chain manager projection store flowchart goop draft 2 2015-03-02 20:20:17 +09:00
Scott Lystig Fritchie
dbcc87b4a4 WIP: Chain manager projection store flowchart goop 2015-03-02 20:20:17 +09:00