tests: Fix flaky TestAccountSelected test

[ohill]

Summary

Go 1.20 introduced a change to math/rand that seeds the global RNG with a random value. This test TestAccountSelected was previously relying on the fixed-seed global RNG for deterministic behavior. There was already a non-global random generator used for this test in testingEnvMoreKeys, but it was not used for two usages of the global RNG to so different values would be produced for multiple iterations calling testingEnv.

Running with GODEBUG=randautoseed=0 and extra logging revealed the difference between the old and new behavior.

Test Plan

Existing tests should pass, and the number of leaders chosen in this test is unchanged, the same as when GODEBUG=randautoseed=0 is set.

Hardened these tests further by asserting the leader & committee sizes checked by these tests match the deterministic values used since #716 and #1094 (the last consensus-gated change to this package). The values being asserted came from extra logging added to checkouts of ec4d9b5 and 566855e on Ubuntu 18.04 with Go 1.12.17, and boost 1.65.1. These values were reproduced in this PR after adding a rand.New(rand.NewSource(1)) to generate the sequence of seeds, matching the previous behavior before Go 1.20.

[codecov]

Codecov Report

Merging #5788 (2a88dbd) into master (d3df476) will increase coverage by 0.01%.
Report is 3 commits behind head on master.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #5788      +/-   ##
==========================================
+ Coverage   55.53%   55.55%   +0.01%     
==========================================
  Files         473      473              
  Lines       66815    66815              
==========================================
+ Hits        37105    37117      +12     
+ Misses      27192    27188       -4     
+ Partials     2518     2510       -8     

see 13 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[jasonpaulos]

Makes sense!

[ohill]

Hardened these tests further by asserting the leader & committee sizes checked by these tests match the deterministic values used since #716 and #1094 (the last consensus-gated change to this package). The values being asserted came from extra logging added to checkouts of ec4d9b5 and 566855e on Ubuntu 18.04 with Go 1.12.17, and boost 1.65.1 (running each test individually to avoid reuse of the global RNG between tests). These values were reproduced in this PR after adding a rand.New(rand.NewSource(1)) to generate the sequence of seeds, matching the previous behavior before Go 1.20.

[jasonpaulos]

LGTM