/
PeerSelection.hs
1638 lines (1450 loc) · 67.2 KB
/
PeerSelection.hs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
{-# LANGUAGE BangPatterns #-}
{-# LANGUAGE DeriveFunctor #-}
{-# LANGUAGE DerivingVia #-}
{-# LANGUAGE DerivingStrategies #-}
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE GeneralisedNewtypeDeriving #-}
{-# LANGUAGE LambdaCase #-}
{-# LANGUAGE NamedFieldPuns #-}
{-# LANGUAGE NumericUnderscores #-}
{-# LANGUAGE RecordWildCards #-}
{-# LANGUAGE ScopedTypeVariables #-}
{-# OPTIONS_GHC -Wno-orphans #-}
module Test.Ouroboros.Network.PeerSelection (tests) where
import qualified Data.ByteString.Char8 as BS
import Data.Function (on)
import Data.List (groupBy, foldl')
import Data.Maybe (listToMaybe, isNothing, fromMaybe)
import Data.Set (Set)
import qualified Data.Set as Set
import Data.Map.Strict (Map)
import qualified Data.Map.Strict as Map
import Data.Void (Void)
import qualified Data.Signal as Signal
import Data.Signal (Signal, Events, E(E), TS(TS))
import qualified Data.OrdPSQ as PSQ
import Control.Monad.Class.MonadTime
import Control.Tracer (Tracer (..))
import qualified Network.DNS as DNS (defaultResolvConf)
import Network.Socket (SockAddr)
import Network.Mux.Timeout
import Ouroboros.Network.PeerSelection.Governor hiding
(PeerSelectionState (..))
import qualified Ouroboros.Network.PeerSelection.Governor as Governor
import qualified Ouroboros.Network.PeerSelection.KnownPeers as KnownPeers
import qualified Ouroboros.Network.PeerSelection.LocalRootPeers as LocalRootPeers
import qualified Ouroboros.Network.PeerSelection.EstablishedPeers as EstablishedPeers
import Ouroboros.Network.PeerSelection.RootPeersDNS
import Test.Ouroboros.Network.PeerSelection.Instances
import qualified Test.Ouroboros.Network.PeerSelection.LocalRootPeers
import Test.Ouroboros.Network.PeerSelection.MockEnvironment hiding (tests)
import qualified Test.Ouroboros.Network.PeerSelection.MockEnvironment
import Test.Ouroboros.Network.PeerSelection.PeerGraph
import Test.QuickCheck
import Test.QuickCheck.Signal
import Test.Tasty (TestTree, testGroup, after, DependencyType(..))
import Test.Tasty.QuickCheck (testProperty)
tests :: TestTree
tests =
testGroup "Ouroboros.Network.PeerSelection"
[ Test.Ouroboros.Network.PeerSelection.LocalRootPeers.tests
, Test.Ouroboros.Network.PeerSelection.MockEnvironment.tests
, testGroup "basic"
[ testProperty "has output" prop_governor_hasoutput
, testProperty "no failure" prop_governor_nofail
, testProperty "no livelock" prop_governor_nolivelock
]
-- The no livelock property is needed to ensure other tests terminate
, after AllSucceed "Ouroboros.Network.PeerSelection.basic" $
testGroup "safety"
[ testProperty "no excess busyness" prop_governor_nobusyness
, testProperty "event coverage" prop_governor_trace_coverage
, testProperty "connection status" prop_governor_connstatus
]
-- The no livelock property is needed to ensure other tests terminate
, after AllSucceed "Ouroboros.Network.PeerSelection.basic" $
testGroup "progress"
[ testProperty "gossip reachable" prop_governor_gossip_1hr
-- , testProperty "progresses towards public root peers target (from below)"
-- prop_governor_target_publicroots
, testProperty "progresses towards known peers target (from below)"
prop_governor_target_known_below
, testProperty "progresses towards known peers target (from above)"
(expectFailure prop_governor_target_known_above)
--TODO: for the moment this correctly fails because it finds
-- a nice corner case in the space, that we've not yet
-- decided how we wish to resolve.
, testProperty "progresses towards established peers target (from below)"
prop_governor_target_established_below
, testProperty "progresses towards established peers target (from above)"
prop_governor_target_established_above
, testProperty "progresses towards active peers target (from below)"
prop_governor_target_active_below
, testProperty "progresses towards active peers target (from above)"
prop_governor_target_active_above
]
]
--
-- QuickCheck properties
--
-- We start with basic properties in the style of "never does bad things"
-- and progress to properties that check that it "eventually does good things".
--
-- In the "never does bad things" category we have:
--
-- * A basic state space exploration property that checks we don't encounter
-- internal errors. This includes some limited checking that we get adequate
-- coverage of the different actions, by looking for coverage of all the
-- trace events. The coverage checks here are useful to give us confidence
-- about coverage for some of the other properties.
--
-- * A no-livelock property. This checks that the governor does not get stuck
-- doing too many steps at a single moment in (virtual) time. It's quite easy
-- to write bugs that don't cause the governor to fail, but cause it to go
-- into a busy cycle. See also the "no excessive busyness" property for a
-- more advanced version.
--
-- * A "no excessive busyness" property. This checks that the governor does not
-- remain too busy for too long. It's quite easy to write bugs that don't
-- cause the governor to fail, but cause it to go into fairly-busy cycles.
--
-- * A state consistency property that the governor's view of part of the state
-- and the "true" state of the mock environment are maintained in an
-- appropriate correspondence.
--
-- In the "eventually does good things" category we have:
--
-- * A basic property to check the governor does produce non-trivial traces.
--
-- * A cold peer gossip "reachable" property: that the governor either hits
-- its target for the number of cold peers, or finds all the reachable peers.
--
-- * A known peer target progress property: that the governor makes progress
-- within a bounded time towards its known peers target, from below and above.
--
-- * An established peer target property: the same as above but for established
-- peers.
--
-- * An active peer target property: the same as above but for active peers.
--
-- Properties that we would like to have:
--
-- * A public root peers target property: that the governor hits its target for
-- for the number of public root peers (or as near as possible), and does
-- not "grossly" overshoot. Since the public roots is a one sided target, but
-- we don't want to overshoot excessively.
--
-- * A local root peers target property: that the governor hits its target for
-- getting all its local root peers into the established state, and a target
-- number of them into the active state (or as near as possible).
--
-- Other properties we might like to think about
--
-- * time to find new nodes after a graph change is ok
-- * targets or root peer set dynamic
-- | As the most basic property we run the governor and check that it produces
-- any trace output at all. It should elicit some activity, unless the test
-- environment is actually empty.
--
prop_governor_hasoutput :: GovernorMockEnvironment -> Bool
prop_governor_hasoutput env =
let trace = selectPeerSelectionTraceEvents $
runGovernorInMockEnvironment env
in hasOutput env (selectGovernorEvents trace)
hasOutput :: GovernorMockEnvironment
-> [(Time, TracePeerSelection PeerAddr)]
-> Bool
hasOutput _ (_:_) = True
hasOutput env [] = isEmptyEnv env
isEmptyEnv :: GovernorMockEnvironment -> Bool
isEmptyEnv GovernorMockEnvironment {
localRootPeers,
publicRootPeers,
targets
} =
(LocalRootPeers.null localRootPeers
|| all (\(t,_) -> targetNumberOfKnownPeers t == 0) targets)
&& (Set.null publicRootPeers
|| all (\(t,_) -> targetNumberOfRootPeers t == 0) targets)
-- | As a basic property we run the governor to explore its state space a bit
-- and check it does not throw any exceptions (assertions such as invariant
-- violations).
--
-- We do /not/ assume freedom from livelock for this property, so we run the
-- governor for a maximum number of trace events rather than for a fixed
-- simulated time.
--
prop_governor_nofail :: GovernorMockEnvironment -> Bool
prop_governor_nofail env =
let trace = take 5000 .
selectPeerSelectionTraceEvents $
runGovernorInMockEnvironment env
in foldl' (flip seq) True
[ assertPeerSelectionState st ()
| (_, GovernorDebug (TraceGovernorState _ _ st)) <- trace ]
-- | It is relatively easy to write bugs where the governor is stuck in a tight
-- cycle of continuous activity. Due to the way the I\/O simulator manages
-- virtual time, these bugs exhibits themselves by infinite trace activity
-- without time advancing.
--
-- It is important to catch these bugs early in the set of tests, since it is
-- hard to write many of the other more interesting properties if there are
-- these kinds of livelock bugs. Or to put it another way, the other properties
-- can be expressed more simply if they can assume within event traces that the
-- time always advances after some finite number of events.
--
prop_governor_nolivelock :: GovernorMockEnvironment -> Property
prop_governor_nolivelock env =
let trace = take 5000 .
selectGovernorEvents .
selectPeerSelectionTraceEvents $
runGovernorInMockEnvironment env
in case tooManyEventsBeforeTimeAdvances 1000 trace of
Nothing -> property True
Just (t, es) ->
counterexample
("over 1000 events at time: " ++ show t ++ "\n" ++
"first 50 events: " ++ (unlines . map show . take 50 $ es)) $
property False
-- | Scan the trace and return any occurrence where we have at least threshold
-- events before virtual time moves on. Return the tail of the trace from that
-- point on.
--
tooManyEventsBeforeTimeAdvances :: Int -> [(Time, e)] -> Maybe (Time, [e])
tooManyEventsBeforeTimeAdvances _ [] = Nothing
tooManyEventsBeforeTimeAdvances threshold trace0 =
go [ (t, diffTime t' t, e)
| ((t, e), (t', _)) <- zip trace0 (tail trace0) ]
where
go [] = Nothing
go trace@((t,_,_):_) = case countdown threshold trace of
Just es' -> go es'
Nothing -> Just (t, trace')
where
trace' = take threshold [ e | (_,_,e) <- trace ]
countdown 0 _ = Nothing
countdown _ [] = Just []
countdown n ((_t,dt,_e):es)
| dt == 0 = countdown (n-1) es
| otherwise = Just es
-- | It is easy to get bugs where the governor is stuck in a cycle, working but
-- not making progress. This kind of bug would result in the governor thread
-- being excessively busy, so it might not be easily noticed.
--
-- This is more subtle and general than a simple livelock test that just checks
-- we don't get completely stuck. This property is about the possibility that
-- the governor is excessively busy over some period of time. This includes
-- "slow" livelocks where time advances during some of the steps in the cycle.
-- More interestingly this is also about a failure to converge and return to
-- being idle sufficiently quickly.
--
-- For example the governor could gets stuck in a cycle promoting and demoting
-- a peer once a second. In such a failure mode it will have a continuous level
-- of activity and will not return to being idle (perhaps ever or perhaps for
-- an extended period until some other environment perturbation gets us out of
-- the cycle).
--
-- The approach we take is based on the observation that the governor can
-- (quite reasonably) start with a big burst of activity (e.g. as it gossips
-- to discover a big graph) but that in the long term it settles down and only
-- has small bursts of activity in reaction to perturbations in the environment
-- such as failures or changes in targets.
--
-- The approach we take is to look at spans of busy activity followed by
-- periods of idleness. If the spans of busy activity are too long then we
-- fail. So this counts the time of busyness not the number of events. We
-- account for activity in the environment that the governor needs to respond
-- to by counting \"perturbation credits"\: more credits means we allow longer
-- spans of busyness.
--
-- More specifically: we look at runs of events where the time between events
-- is less than a threshold. This implies there follows a threshold level of
-- idleness. Starting or within that run of events there can be environment
-- events. These are the perturbations from the environment that we expect to
-- trigger a series of responses from the governor. So we expect longer periods
-- of business for bigger perturbations. We sum all the perturbations credits
-- included in a run of events. We use a formula that relates the credits to
-- the permitted time span of the run. If the run is within the permitted time
-- span then it is ok, otherwise it is a failure (and the run is the
-- counterexample).
--
-- TODO: This test uses static root peers, but we should move to dynamic root
-- peers.
--
prop_governor_nobusyness :: GovernorMockEnvironment -> Property
prop_governor_nobusyness env =
let trace = selectPeerSelectionTraceEvents $
runGovernorInMockEnvironment env
in case tooBusyForTooLong (takeFirstNHours 10 trace) of
Nothing -> property True
Just (busyStartTime, busyEndTime, credits, trace') ->
counterexample
("busy span too long\n" ++
"start time: " ++ show busyStartTime ++ "\n" ++
"end time: " ++ show busyEndTime ++ "\n" ++
"span credits: " ++ show credits ++ "\n" ++
"first 50 events:\n" ++
(unlines . map show . take 50 $ trace')) $
property False
--
tooBusyForTooLong :: [(Time, TestTraceEvent)]
-> Maybe (Time, Time, DiffTime,
[(Time, TestTraceEvent)])
tooBusyForTooLong trace0 =
-- Pass in each timed event, with the diff-time to the next event
idle [ (t, diffTime t' t, e)
| ((t, e), (t', _)) <- zip trace0 (tail trace0) ]
where
-- How long between events before we say it's the end of a busy span
sameSpanThreshold :: DiffTime
sameSpanThreshold = 45
-- Starting credits for a busy span, even if there are no triggering
-- environment events. The value chosen here takes account of the normal
-- exponential backoff is 2+4+8+16+32 = 62, before a gap of 64 that's
-- bigger than the sameSpanThreshold of 45.
initialEventCredits :: DiffTime
initialEventCredits = 65
-- We process the event trace linearly, flipping between two states: idle
-- and busy. In the idle state, the next (non-debug) event flips us into
-- the busy state, starting with some minimal initial credits.
idle :: [(Time, DiffTime, TestTraceEvent)]
-> Maybe (Time, Time, DiffTime, [(Time, TestTraceEvent)])
idle [] = Nothing
idle ((_, _, GovernorDebug{}):trace') = idle trace'
idle trace@((busyStartTime,_,_):_) =
case busy busyStartTime initialEventCredits trace of
Right trace' -> idle trace'
Left (busyEndTime, credits) ->
Just (busyStartTime, busyEndTime, credits, trace')
where
trace' = [ (t, e)
| (t,_dt, e) <-
takeWhile (\(t,_,_) -> t <= busyEndTime) trace
, case e of
GovernorDebug{} -> False
_ -> True
]
busy :: Time -> DiffTime -> [(Time, DiffTime, TestTraceEvent)]
-> Either (Time, DiffTime) [(Time, DiffTime, TestTraceEvent)]
-- For normal governor events we check if the length of the busy time span
-- is now too big (adjusted for any perturbation credits). If so we've
-- found a violation.
busy !busyStartTime !credits ((busyEndTime, _dt, GovernorEvent{}) : _trace')
| busySpanLength > credits = Left (busyEndTime, credits)
where
busySpanLength = diffTime busyEndTime busyStartTime
-- We also look at how long it is to the next event to see if this is the
-- last event in the busy span, and if so we return to idle.
busy !_busyStartTime !_credits ((_t, dt, _event) : trace')
| dt > sameSpanThreshold = Right trace'
-- For environment events we calculate the perturbation credits this
-- contributes and add it to our running total.
busy !busyStartTime !credits ((_, _, MockEnvEvent e) : trace') =
busy busyStartTime (credits + fromIntegral (envEventCredits e)) trace'
-- Otherwise we move on to the next event, updating the length of this busy
-- time span.
busy !busyStartTime !credits (_ : trace') =
busy busyStartTime credits trace'
-- running out of events before we find a violation is ok
busy !_ !_ [] = Right []
envEventCredits :: TraceMockEnv -> Int
envEventCredits (TraceEnvAddPeers peerGraph) = 80 * 5 + length adjacency * 5
where
PeerGraph adjacency = peerGraph
envEventCredits (TraceEnvSetLocalRoots peers) = LocalRootPeers.size peers
envEventCredits (TraceEnvSetPublicRoots peers) = Set.size peers
envEventCredits TraceEnvPublicRootTTL = 60
envEventCredits (TraceEnvGossipTTL _) = 30
envEventCredits (TraceEnvSetTargets PeerSelectionTargets {
targetNumberOfRootPeers = _,
targetNumberOfKnownPeers,
targetNumberOfEstablishedPeers,
targetNumberOfActivePeers
}) = 80
+ 10 * (targetNumberOfKnownPeers
+ targetNumberOfEstablishedPeers
+ targetNumberOfActivePeers)
envEventCredits (TraceEnvPeersDemote Noop _) = 10
envEventCredits (TraceEnvPeersDemote ToWarm _) = 30
envEventCredits (TraceEnvPeersDemote ToCold _) = 30
envEventCredits TraceEnvPeersStatus{} = 0
-- These events are visible in the environment but are the result of actions
-- initiated by the governor, hence the get no credit.
envEventCredits TraceEnvRootsResult{} = 0
envEventCredits TraceEnvGossipRequest{} = 0
envEventCredits TraceEnvGossipResult{} = 0
-- | A coverage property, much like 'prop_governor_nofail' but we look to see
-- that we get adequate coverage of the state space. We look for all the trace
-- events that the governor can produce, and tabules which ones we see.
--
prop_governor_trace_coverage :: GovernorMockEnvironment -> Property
prop_governor_trace_coverage env =
let trace = take 5000 .
selectPeerSelectionTraceEvents $
runGovernorInMockEnvironment env
traceNumsSeen = collectTraces trace
traceNamesSeen = allTraceNames `Map.restrictKeys` traceNumsSeen
in coverTable "trace events" [ (n, 1) | n <- Map.elems allTraceNames ] $
tabulate "trace events" (Map.elems traceNamesSeen)
True
--TODO: use cover to check we do indeed get them all. There are a few
-- cases we do not cover yet. These should be fixed first.
collectTraces :: [(Time, TestTraceEvent)] -> Set Int
collectTraces trace =
Set.fromList [ traceNum e | (_, GovernorEvent e) <- trace ]
traceNum :: TracePeerSelection peeraddr -> Int
traceNum TraceLocalRootPeersChanged{} = 00
traceNum TraceTargetsChanged{} = 01
traceNum TracePublicRootsRequest{} = 02
traceNum TracePublicRootsResults{} = 03
traceNum TracePublicRootsFailure{} = 04
traceNum TraceGossipRequests{} = 05
traceNum TraceGossipResults{} = 06
traceNum TraceForgetColdPeers{} = 07
traceNum TracePromoteColdPeers{} = 08
--traceNum TracePromoteColdLocalPeers{} = 09
traceNum TracePromoteColdFailed{} = 10
traceNum TracePromoteColdDone{} = 11
traceNum TracePromoteWarmPeers{} = 12
--traceNum TracePromoteWarmLocalPeers{} = 13
traceNum TracePromoteWarmFailed{} = 14
traceNum TracePromoteWarmDone{} = 15
traceNum TraceDemoteWarmPeers{} = 16
traceNum TraceDemoteWarmFailed{} = 17
traceNum TraceDemoteWarmDone{} = 18
traceNum TraceDemoteHotPeers{} = 19
traceNum TraceDemoteHotFailed{} = 20
traceNum TraceDemoteHotDone{} = 21
traceNum TraceDemoteAsynchronous{} = 22
traceNum TraceGovernorWakeup{} = 23
allTraceNames :: Map Int String
allTraceNames =
Map.fromList
[ (00, "TraceLocalRootPeersChanged")
, (01, "TraceTargetsChanged")
, (02, "TracePublicRootsRequest")
, (03, "TracePublicRootsResults")
, (04, "TracePublicRootsFailure")
, (05, "TraceGossipRequests")
, (06, "TraceGossipResults")
, (07, "TraceForgetColdPeers")
, (08, "TracePromoteColdPeers")
-- , (09, "TracePromoteColdLocalPeers")
, (10, "TracePromoteColdFailed")
, (11, "TracePromoteColdDone")
, (12, "TracePromoteWarmPeers")
-- , (13, "TracePromoteWarmLocalPeers")
, (14, "TracePromoteWarmFailed")
, (15, "TracePromoteWarmDone")
, (16, "TraceDemoteWarmPeers")
, (17, "TraceDemoteWarmFailed")
, (18, "TraceDemoteWarmDone")
, (19, "TraceDemoteHotPeers")
, (20, "TraceDemoteHotFailed")
, (21, "TraceDemoteHotDone")
, (22, "TraceDemoteAsynchronous")
, (23, "TraceGovernorWakeup")
]
-- | Run the governor for up to 1 hour (simulated obviously) and look at the
-- set of known peers it has selected. This uses static targets and root peers.
--
-- As a basic correctness property, the peers the governor selects must be a
-- subset of those that are in principle reachable in the mock network
-- environment.
--
-- More interestingly, we expect the governor to find enough peers. Either it
-- must find all the reachable ones, or if the target for the number of known
-- peers to find is too low then it should at least find the target number.
--
prop_governor_gossip_1hr :: GovernorMockEnvironment -> Property
prop_governor_gossip_1hr env@GovernorMockEnvironment {
peerGraph,
localRootPeers,
publicRootPeers,
targets
} =
let trace = selectPeerSelectionTraceEvents $
runGovernorInMockEnvironment env {
targets = singletonScript (targets', NoDelay)
}
Just found = knownPeersAfter1Hour trace
reachable = firstGossipReachablePeers peerGraph
(LocalRootPeers.keysSet localRootPeers <> publicRootPeers)
in subsetProperty found reachable
.&&. bigEnoughProperty found reachable
where
-- This test is only about testing gossiping,
-- so do not try to establish connections:
targets' :: PeerSelectionTargets
targets' = (fst (scriptHead targets)) {
targetNumberOfEstablishedPeers = 0,
targetNumberOfActivePeers = 0
}
knownPeersAfter1Hour :: [(Time, TestTraceEvent)] -> Maybe (Set PeerAddr)
knownPeersAfter1Hour trace =
listToMaybe
[ KnownPeers.toSet (Governor.knownPeers st)
| (_, GovernorDebug (TraceGovernorState _ _ st))
<- reverse (takeFirstNHours 1 trace)
]
-- The ones we find should be a subset of the ones possible to find
subsetProperty found reachable =
counterexample ("reachable: " ++ show reachable ++ "\n" ++
"found: " ++ show found) $
property (found `Set.isSubsetOf` reachable)
-- We expect to find enough of them, either the target number or the
-- maximum reachable.
bigEnoughProperty found reachable
-- But there's an awkward corner case: if the number of public roots
-- available is bigger than the target then we will likely not get
-- all the roots (but which subset we get is random), but if we don't
-- get all the roots then the set of peers actually reachable is
-- incomplete, so we cannot expect to reach the usual target.
--
-- But we can at least expect to hit the target for root peers.
| Set.size publicRootPeers > targetNumberOfRootPeers targets'
= property (Set.size found >= targetNumberOfRootPeers targets')
| otherwise
= counterexample ("reachable : " ++ show reachable ++ "\n" ++
"found : " ++ show found ++ "\n" ++
"found # : " ++ show (Set.size found) ++ "\n" ++
"expected #: " ++ show expected) $
property (Set.size found == expected)
where
expected = Set.size reachable `min` targetNumberOfKnownPeers targets'
-- | Check the governor's view of connection status does not lag behind reality
-- by too much.
--
prop_governor_connstatus :: GovernorMockEnvironment -> Property
prop_governor_connstatus env =
let trace = takeFirstNHours 1
. selectPeerSelectionTraceEvents $
runGovernorInMockEnvironment env
--TODO: check any actually get a true status output and try some deliberate bugs
in conjoin (map ok (groupBy ((==) `on` fst) trace))
where
-- We look at events when the environment's view of the state of all the
-- peer connections changed, and check that before simulated time advances
-- the governor's view of the same state was brought in sync.
--
-- We do that by finding the env events and then looking for the last
-- governor state event before time moves on.
ok :: [(Time, TestTraceEvent)] -> Property
ok trace =
counterexample ("last few events:\n" ++ (unlines . map show) trace) $
case (lastEnvStatus, lastGovStatus) of
(Nothing, _) -> property True
(Just envStatus, Just govStatus) -> envStatus === govStatus
(Just envStatus, Nothing) -> envStatus === Map.empty
where
lastEnvStatus =
listToMaybe
[ Map.filter (not . isPeerCold) status
| (_, MockEnvEvent (TraceEnvPeersStatus status)) <- reverse trace ]
isPeerCold PeerCold = True
isPeerCold _ = False
lastGovStatus =
listToMaybe
[ Governor.establishedPeersStatus st
| (_, GovernorDebug (TraceGovernorState _ _ st)) <- reverse trace ]
--
-- Progress properties
--
-- | The main progress property for known peers: that we make progress towards
-- the target for known peers from below. See 'prop_governor_target_known_above'
-- for the (simpler) corresponding property for hitting the target from above.
--
-- Intuitively the property we want is that the governor either hits its target
-- for the number of known peers, or gets as close as reasonably possible. The
-- environment may be such that it prevents the governor from reaching its
-- target, e.g. because the target is too high, or not all peers may be
-- reachable by the gossip graph.
--
-- We approach this property as the conjunction of several simpler properties.
-- We take this approach for three main reasons.
--
-- 1. Firstly modularity help us break down a complex problem into simpler
-- problems. Overall this progress idea turns out to be quite subtle and
-- tricky to express precisely in a way that actually works.
-- 2. Secondly, modularity can give us opportunities to reuse code in other
-- properties and we want to have progress properties for all the governor
-- targets.
-- 3. Thirdly, it turns out to be hard to dictate in a universal way precisely
-- what the governor can be expected to do. It is simpler to specify looser
-- constraints on what it must and must not do. We can then argue informally
-- that the combination of properties must lead to the kinds of outcomes we
-- intend.
--
-- We decompose the progress property into the following (informally stated)
-- properties:
--
-- 1. The set of peers the governor knows about is a subset of the peers the
-- environment has told the governor about.
--
-- This is a weak property since it simply says that the governor does not
-- invent things out of thin air. We might expect that we could strengthen
-- this property to require that the subset be maximal in some sense however
-- such a property is violated by dynamic targets. There are also timing
-- issues which would complicate such a strengthened property: the governor
-- has legitimate reasons to update its internal state some time after the
-- environment informs it about new peers.
--
-- 2. If the governor is below target and has the opportunity to gossip then
-- within a bounded time it should perform a gossip with one of its known
-- peers.
--
-- This is the primary progress property. It is a relatively weak property:
-- we do not require that progress is actually made, just that opportunities
-- for progress are taken when available. We cannot always demand actual
-- progress since there are environments where it is not possible to make
-- progress, even though opportunities for gossip remain available. Examples
-- include environments where the total set of peers in the graph is less
-- than the target for known peers.
--
-- 3. The governor should not gossip too frequently with any individual peer,
-- except when the governor forgets known peers.
--
-- This is both useful in its own right, but it also helps to strengthen the
-- primary property by helping to ensure that the choices of which peers to
-- gossip with are reasonable. In the primary property we do not require that
-- the peer the the governor chooses to gossip with is one of the
-- opportunities as defined by the property. We do not require this because
-- the set of opportunities is a lower bound not an upper bound, and trying
-- to make it a tight bound becomes complex and over-specifies behaviour.
-- There is the danger however that the governor could appear to try to make
-- progress by gossiping but always picking useless choices that avoid making
-- actual progress. By requiring that the governor not gossip with any
-- individual peer too often we can shrink the set of peers the governor can
-- choose and thus force the governor to eventually pick other peers to
-- gossip with, which should mean the governor eventually picks peers that
-- can enable progress.
--
-- 4. When the governor does perform a gossip, within a bounded time it should
-- include the results into its known peer set, or the known peer set should
-- reach its target size.
--
-- This helps to strengthen the primary progress property by ensuring the
-- results of gossip are used to make progress when that is possible.
--
-- 5. The governor should not shrink its known peer set except when it is above
-- the target size.
--
-- This also helps to strengthen the second property by ensuring monotonic
-- progress, except when we overshoot targets or when targets are reduced.
--
-- The overall progress argument is then an semi-formal argument, structured
-- much like classic proofs about loops. A classic loop proof has two parts: 1.
-- if the loop does terminate it gets the right result, and 2. it must
-- eventually terminate because it makes progress in some measure that is
-- bounded.
--
-- Of course in our setting there is no termination, but we can reach a goal
-- and remain in a steady state until the environment changes. Our argument is
-- that the governor makes progress to increase the size of its set of known
-- peers until either it hits its target number of known peers, or it reaches a
-- maximal possible set. As a corollary if the targets do not change too
-- frequently then it will eventually hit the target or reach a maximal set.
--
-- Property number 1 above tells us that if we do reach our goal condition that
-- we will have a correct result, as property 1 tells us that all the governors'
-- known peers are ones supplied by the environment.
--
-- Progress from below relies on the combination of property 2, 3, 4 and 5.
-- Property 2 tells us that we eventually do some gossip with some peer, but
-- does not by itself establish that we make progress in a bounded measure.
-- Property 3 gives us the bounded measure. Property 3 gives us a set of peers
-- that we have not gossiped with recently. When the governor does gossip with
-- a peer then it is removed from this set (but scheduled to be added back some
-- time later). So the measure is the size of this set of peers. It is clearly
-- bounded below by the empty set. So the combination of 2 and 3 tells us we
-- make progress in this bounded measure, but that does not directly translate
-- into increasing the size of the known peers set. Properties 4 and 5 tell us
-- that progress with gossiping will eventually translate into increasing the
-- size of the known peers set if that is possible.
--
-- There is one known wrinkle to this argument to do with property 3 that when
-- the governor gossips with a peer it is removed from the tracking set however
-- it gets added back some time later. If they get added back too soon then it
-- would undermine the progress argument because it would break the argument
-- about decreasing the bounded measure. This is readily solved however: we
-- simply need to make sure the time scale for gossip frequency is relatively
-- long, and the other progress bounds are relatively short.
--
prop_governor_target_known_below :: GovernorMockEnvironment -> Property
prop_governor_target_known_below env =
prop_governor_target_known_1_valid_subset env
.&&. prop_governor_target_known_2_opportunity_taken env
.&&. prop_governor_target_known_3_not_too_chatty env
.&&. prop_governor_target_known_4_results_used env
.&&. prop_governor_target_known_5_no_shrink_below env
-- | The set of peers the governor knows about is a subset of the peers the
-- environment has told the governor about.
--
-- We derive a number of signals:
--
-- 1. A signal of the accumulation of all the peers the environment has ever
-- told the governor about, based on the environment trace.
--
-- 2. A signal of the set of known peers in the governor state.
--
-- Based on these signals we check:
--
-- * That the governor known peers is a subset of the accumulated environment
-- known peers.
--
prop_governor_target_known_1_valid_subset :: GovernorMockEnvironment
-> Property
prop_governor_target_known_1_valid_subset env =
let events = Signal.eventsFromListUpToTime (Time (60 * 60))
. selectPeerSelectionTraceEvents
. runGovernorInMockEnvironment
$ env
envKnownPeersSig :: Signal (Set PeerAddr)
envKnownPeersSig =
Signal.nubBy ((==) `on` Set.size)
. Signal.scanl Set.union Set.empty
. Signal.fromChangeEvents Set.empty
. Signal.selectEvents
(\case
TraceEnvSetLocalRoots x -> Just (LocalRootPeers.keysSet x)
TraceEnvRootsResult x -> Just (Set.fromList x)
TraceEnvGossipResult _ x -> Just (Set.fromList x)
_ -> Nothing
)
. selectEnvEvents
$ events
govKnownPeersSig :: Signal (Set PeerAddr)
govKnownPeersSig =
selectGovState (KnownPeers.toSet . Governor.knownPeers) events
validState :: Set PeerAddr -> Set PeerAddr -> Bool
validState knownPeersEnv knownPeersGov =
knownPeersGov `Set.isSubsetOf` knownPeersEnv
in counterexample
"Signal key: (environment known peers, governor known peers)" $
signalProperty 20 show (uncurry validState) $
(,) <$> envKnownPeersSig
<*> govKnownPeersSig
-- | If the governor is below target and has the opportunity to gossip then
-- within a bounded time it should perform a gossip with one of its known peers.
--
-- We derive a number of signals:
--
-- 1. A signal of the target for known peers from the environment
--
-- 2. A signal of the set of known peers in the governor state.
--
-- 3. A signal of the set of peers with which the governor has gossiped
-- recently, based on the requests to the environment
--
-- 4. Based on 2 and 3, a signal of the set of gossip opportunities: the
-- current known peers that are not in the recent gossip set.
--
-- 5. A signal of the environment gossip request events.
--
-- 6. Based on 1, 2, 4 and 5, a signal that becomes False if for 30 seconds:
-- the number of known peers is below target; the set of opportunities is
-- non empty; and no gossip request event has occurred.
--
-- Based on these signals we check:
--
-- * That the signal 6 remains True at all times.
--
prop_governor_target_known_2_opportunity_taken :: GovernorMockEnvironment
-> Property
prop_governor_target_known_2_opportunity_taken env =
let events = Signal.eventsFromListUpToTime (Time (10 * 60 * 60))
. selectPeerSelectionTraceEvents
. runGovernorInMockEnvironment
$ env
envTargetsSig :: Signal Int
envTargetsSig =
selectEnvTargets targetNumberOfKnownPeers events
govKnownPeersSig :: Signal (Set PeerAddr)
govKnownPeersSig =
selectGovState (KnownPeers.toSet . Governor.knownPeers) events
envGossipUnavailableSig :: Signal (Set PeerAddr)
envGossipUnavailableSig =
Signal.keyedLinger
-- peers are unavailable for gossip for at least an
-- hour after each gossip interaction
(60 * 60)
(maybe Set.empty Set.singleton)
. Signal.fromEvents
. Signal.selectEvents
(\case TraceEnvGossipRequest peer _ -> Just peer
_ -> Nothing)
. selectEnvEvents
$ events
-- We define the governor's gossip opportunities at any point in time
-- to be the governor's set of known peers, less the ones we can see
-- that it has gossiped with recently.
--
gossipOpportunitiesSig :: Signal (Set PeerAddr)
gossipOpportunitiesSig =
(Set.\\) <$> govKnownPeersSig <*> envGossipUnavailableSig
-- Note that we only require that the governor try to gossip, it does
-- not have to succeed.
envGossipsEventsAsSig :: Signal (Maybe PeerAddr)
envGossipsEventsAsSig =
Signal.fromEvents
. Signal.selectEvents
(\case TraceEnvGossipRequest addr _ -> Just addr
_ -> Nothing)
. selectEnvEvents
$ events
-- The signal of all the things of interest for this property.
-- This is used to compute the final predicate, and is also what
-- we want to report if there is a property violation.
combinedSig :: Signal (Int,
Set PeerAddr,
Set PeerAddr,
Maybe PeerAddr)
combinedSig =
(,,,) <$> envTargetsSig
<*> govKnownPeersSig
<*> gossipOpportunitiesSig
<*> envGossipsEventsAsSig
-- This is the ultimate predicate signal
gossipOpportunitiesOkSig :: Signal Bool
gossipOpportunitiesOkSig =
Signal.truncateAt (Time (60 * 60 * 10)) $
governorEventuallyTakesGossipOpportunities combinedSig
in counterexample
"Signal key: (target, known peers, opportunities, gossip event)" $
-- Check the predicate signal but for failures report the input signal
signalProperty 20 (show . snd) fst $
(,) <$> gossipOpportunitiesOkSig
<*> combinedSig
governorEventuallyTakesGossipOpportunities
:: Signal (Int, Set PeerAddr, Set PeerAddr, Maybe PeerAddr)
-> Signal Bool
governorEventuallyTakesGossipOpportunities =
-- Time out and fail after 30 seconds if we enter and remain in a bad state
fmap not
. Signal.timeout timeLimit badState
where
timeLimit :: DiffTime
timeLimit = 30
badState (target, govKnownPeers, gossipOpportunities, gossipEvent) =
-- A bad state is one where we are below target;
Set.size govKnownPeers < target
-- where we do have opportunities; and
&& not (Set.null gossipOpportunities)
-- are not performing an action to take the opportunity.
&& isNothing gossipEvent
-- Note that if a gossip does take place, we do /not/ require the gossip
-- target to be a member of the gossipOpportunities. This is because
-- the gossip opportunities set is a lower bound not an upper bound.
-- There is a separate property to check that we do not gossip too
-- frequently with any individual peer.
-- | The governor should not gossip too frequently with any individual peer,
-- except when the governor forgets known peers.
--
-- We derive a number of signals:
--
-- Based on these signals we check:
--
prop_governor_target_known_3_not_too_chatty :: GovernorMockEnvironment
-> Property
prop_governor_target_known_3_not_too_chatty env =
let events = Signal.eventsFromListUpToTime (Time (10 * 60 * 60))
. selectPeerSelectionTraceEvents
. runGovernorInMockEnvironment
$ env
gossipOk Nothing _ = True
gossipOk (Just peers) unavailable =
Set.null (peers `Set.intersection` unavailable)
in signalProperty 20 show (uncurry gossipOk) $
recentGossipActivity 3600 events
recentGossipActivity :: DiffTime
-> Events TestTraceEvent
-> Signal (Maybe (Set PeerAddr), Set PeerAddr)
recentGossipActivity d =
Signal.fromChangeEvents (Nothing, Set.empty)
. Signal.primitiveTransformEvents (go Set.empty PSQ.empty)
--TODO: we should be able to avoid primitiveTransformEvents and express
-- this as some combo of keyed linger and keyed until.
where
go :: Set PeerAddr
-> PSQ.OrdPSQ PeerAddr Time ()
-> [E TestTraceEvent]
-> [E (Maybe (Set PeerAddr), Set PeerAddr)]
go !recentSet !recentPSQ txs@(E (TS t _) _ : _)
| Just (k, t', _, recentPSQ') <- PSQ.minView recentPSQ
, t' <= t
, let recentSet' = Set.delete k recentSet
= E (TS t' 0) (Nothing, recentSet')
: go recentSet' recentPSQ' txs
-- When we see a gossip request we add it to the recent set and schedule
-- it to be removed again at time d+t. We arrange for the change in the
-- recent set to happen after the gossip event.
go !recentSet !recentPSQ
(E (TS t i) (GovernorEvent (TraceGossipRequests _ _ _ addrs)) : txs) =
let recentSet' = recentSet <> addrs
recentPSQ' = foldl' (\q a -> PSQ.insert a t' () q) recentPSQ addrs
t' = d `addTime` t
in E (TS t i) (Just addrs, recentSet)
: E (TS t (i+1)) (Nothing, recentSet') -- updated in next change at same time
: go recentSet' recentPSQ' txs
-- When the governor is forced to forget known peers, we drop it from
-- the recent activity tracking, which means if it is added back again