#349 - Test Failure - error: in "fc_crypto/rand_test": check rc < E + sigma has failed - Peerplays_Blockchain/peerplays_migrated

rilesdun commented

2022-04-20 19:18:09 +00:00

(Migrated from gitlab.com)

I am seeing a test failure from @bobinson's latest merge from beatrice branch into master branch. I have collected the following logs but the full log can be found here - https://gitlab.com/PBSA/peerplays/-/jobs/2356660496

3348825ms th_a       real128_test.cpp:50           test_method          ] ten: 10. two: 2. twenty: 20. 
3348826ms th_a       real128_test.cpp:51           test_method          ] real128("12345.6789"): 12345.6789 
3348826ms th_a       real128_test.cpp:52           test_method          ] ten/3*3: 9.999999999999999999 
3348828ms th_a       serialization_test.cpp:54     operator()           ] Creating nested object with 100 level(s)
3348829ms th_a       serialization_test.cpp:59     operator()           ] Creating level 100
3348829ms th_a       serialization_test.cpp:54     operator()           ] Creating nested object with 150 level(s)
3348830ms th_a       serialization_test.cpp:59     operator()           ] Creating level 100
3348831ms th_a       serialization_test.cpp:54     operator()           ] Creating nested object with 150 level(s)
3348831ms th_a       serialization_test.cpp:59     operator()           ] Creating level 100
*** 1 failure is detected in the test module "AllTests"
Cleaning up project directory and file based variables
00:00
ERROR: Job failed: exit status 1

I am seeing a test failure from @bobinson's latest merge from beatrice branch into master branch. I have collected the following logs but the full log can be found here - https://gitlab.com/PBSA/peerplays/-/jobs/2356660496 ``` 3348825ms th_a real128_test.cpp:50 test_method ] ten: 10. two: 2. twenty: 20. 3348826ms th_a real128_test.cpp:51 test_method ] real128("12345.6789"): 12345.6789 3348826ms th_a real128_test.cpp:52 test_method ] ten/3*3: 9.999999999999999999 3348828ms th_a serialization_test.cpp:54 operator() ] Creating nested object with 100 level(s) 3348829ms th_a serialization_test.cpp:59 operator() ] Creating level 100 3348829ms th_a serialization_test.cpp:54 operator() ] Creating nested object with 150 level(s) 3348830ms th_a serialization_test.cpp:59 operator() ] Creating level 100 3348831ms th_a serialization_test.cpp:54 operator() ] Creating nested object with 150 level(s) 3348831ms th_a serialization_test.cpp:59 operator() ] Creating level 100 *** 1 failure is detected in the test module "AllTests" Cleaning up project directory and file based variables 00:00 ERROR: Job failed: exit status 1 ```

rilesdun commented

2022-04-20 19:18:10 +00:00

(Migrated from gitlab.com)

assigned to @bobinson and @serkixenos

rilesdun commented

2022-04-20 19:18:55 +00:00

(Migrated from gitlab.com)

mentioned in issue #348

serkixenos commented

2022-04-27 12:28:50 +00:00

(Migrated from gitlab.com)

assigned to @pavel.baykov

pavel.baykov commented

2022-04-27 21:38:21 +00:00

(Migrated from gitlab.com)

that’s was solved before in develop branch, however just to recap:

1.we have different sources of RNG across code base , they don’t have direct tests
2. this specific test was not implemented correctly,
3. based on discussion with @serkixenos, the best way to check RNG is DieHarder test, the original test was written in Fortran by later on ported to C, i found one of the alive project: https://github.com/rurban/dieharder
executed on Ubuntu 20.04, to test default RNG:

#=============================================================================#
#          dieharder version 3.31.2beta Copyright 2003 Robert G. Brown        #
#=============================================================================#
   rng_name    |rands/second|   Seed   |  k ints/sec|k doubles/sec|
        mt19937|  7.65e+07  | 106531697|     76509  |      72453  |
#=============================================================================#
        test_name   |ntup| tsamples |psamples|  p-value |Assessment
#=============================================================================#

most of items passing, but there are failures too
...

rgb_permutations| 5| 100000| 100|0.99809606| WEAK

that’s was solved before in develop branch, however just to recap: 1.we have different sources of RNG across code base , they don’t have direct tests 2. this specific test was not implemented correctly, 3. based on discussion with @serkixenos, the best way to check RNG is DieHarder test, the original test was written in Fortran by later on ported to C, i found one of the alive project: https://github.com/rurban/dieharder executed on Ubuntu 20.04, to test default RNG: ``` #=============================================================================# # dieharder version 3.31.2beta Copyright 2003 Robert G. Brown # #=============================================================================# rng_name |rands/second| Seed | k ints/sec|k doubles/sec| mt19937| 7.65e+07 | 106531697| 76509 | 72453 | #=============================================================================# test_name |ntup| tsamples |psamples| p-value |Assessment #=============================================================================# ``` most of items passing, but there are failures too ... rgb_permutations| 5| 100000| 100|0.99809606| WEAK

pavel.baykov commented

2022-04-27 21:42:49 +00:00

(Migrated from gitlab.com)

if need i can check using Dieharder every source of RNG we have,

pavel.baykov commented

2022-04-27 21:44:27 +00:00

(Migrated from gitlab.com)

but as a good reference is Bitcoin code where, RNG sources are rotating, that’s separate problem besides of quality of RNG in terms of p-values

hirunda commented

2022-05-04 15:02:19 +00:00

(Migrated from gitlab.com)

assigned to @hirunda

serkixenos commented

2022-05-04 16:11:27 +00:00

(Migrated from gitlab.com)

unassigned @bobinson, @serkixenos, and @pavel.baykov

serkixenos commented

2022-05-04 16:12:33 +00:00

(Migrated from gitlab.com)

@hirunda We agreed to remove anything related to fc RNGm from the fc library and peerplays code.

serkixenos commented

2022-05-04 16:13:34 +00:00

(Migrated from gitlab.com)

We will need two merge requests for this, one for fc library, and another for peerplays.

serkixenos commented

2022-05-16 19:26:43 +00:00

(Migrated from gitlab.com)

assigned to @prandnum

hirunda commented

2022-05-16 19:57:44 +00:00

(Migrated from gitlab.com)

Hi Kiran,

Please,

Run fc library test executable "./build/libraries/fc/tests/all_tests" and
observe CI/CD logs, and verify that rng test failure is not reported

Thanks,
Davor

Hi Kiran, Please, Run fc library test executable "./build/libraries/fc/tests/all_tests" and observe CI/CD logs, and verify that rng test failure is not reported Thanks, Davor

prandnum commented

2022-05-19 19:56:43 +00:00

(Migrated from gitlab.com)

no errors seen. logs attached. fc_tests.log

no errors seen. logs attached. [fc_tests.log](/uploads/413b696d92079a37756b912c5197d6ec/fc_tests.log)

prandnum (Migrated from gitlab.com) closed this issue

2022-05-19 19:56:44 +00:00

Test Failure - error: in "fc_crypto/rand_test": check rc < E + sigma has failed #349