fix: improve processEffectiveBalanceUpdates

[twoeths]

Motivation

@nflaig found that epoch transition is 10x slower as seen in altair processEpoch - mainnet_e81889 benchmark. This is because we broke the optimization in #2902

Description

• cache validators in beforeProcessEpoch()
• cache new compounding validators in processPendingConsolidations
• update processEffectiveBalanceUpdates(): use the above to compute effectiveBalanceLimit to avoid unnecessary updates
• also track it in numEffectiveBalanceUpdates metric

[codecov]

Codecov Report

Attention: Patch coverage is 6.89655% with 27 lines in your changes missing coverage. Please review.

Project coverage is 49.37%. Comparing base (499cf57) to head (275f82e).
Report is 1 commits behind head on electra-fork-rebasejul30.

Additional details and impacted files

@@                     Coverage Diff                      @@
##           electra-fork-rebasejul30    #7043      +/-   ##
============================================================
- Coverage                     49.39%   49.37%   -0.03%     
============================================================
  Files                           589      589              
  Lines                         39217    39233      +16     
  Branches                       2250     2243       -7     
============================================================
- Hits                          19372    19371       -1     
- Misses                        19804    19821      +17     
  Partials                         41       41              

[ensi321]

Wonder why this isn't caught by the benchmark check.

This branch previously had a bad benchmark on processEffectiveBalanceUpdates but #7018 fixed it.

[twoeths]

Wonder why this isn't caught by the benchmark check.

This branch previously had a bad benchmark on processEffectiveBalanceUpdates but #7018 fixed it.

@ensi321 the benchmark did not fail on processEffectiveBalanceUpdates(), instead it failed on processEpoch(), specifically on state.hashTreeRoot()

the current version of processEffectiveBalanceUpdates() introduced redundant updates on validator.effectiveBalance with the same value and it caused pressure on the state.hashTreeRoot(). This is fixed by (effectiveBalance < effectiveBalanceLimit && effectiveBalance + UPWARD_THRESHOLD < balance) condition

[twoeths]

spec tests passed, benchmark was confirmed in local env

this branch

 altair processEpoch - mainnet_e81889
    ✔ altair processEpoch - mainnet_e81889                                3.343826 ops/s    299.0586 ms/op        -         12 runs   4.24 s

which is 10x faster than electra branch

altair processEpoch - mainnet_e81889
    ✔ altair processEpoch - mainnet_e81889                               0.3105313 ops/s    3.220287  s/op        -         12 runs   41.8 s

e2e consistently failed but it's the common issue of electra branch, see #6986 (comment)

[nflaig]

LGTM

[wemeetagain]

🎉 This PR is included in v1.22.0 🎉