EF test fails on process_sync_aggregate

[syjn99]

Reproduction

1. Add test for process_sync_aggregate at testing/ef-tests/tests/tests.rs:

test_operation!(
    sync_aggregate,
    SyncAggregate,
    "sync_aggregate",
    process_sync_aggregate
);

2. Run test via make test. In my case assert_eq messed up my shell, so I used make test &> test.log

Result

test tests_process_sync_aggregate::test_operation ... FAILED

...

---- tests_process_sync_aggregate::test_operation stdout ----
Testing case: random_misc_balances_and_half_participation_with_duplicates
Testing case: invalid_signature_infinite_signature_with_single_participant
Testing case: random_only_one_participant_with_duplicates
Testing case: sync_committee_rewards_duplicate_committee_max_effective_balance_only_participate_second_one
Testing case: sync_committee_with_nonparticipating_withdrawable_member
Testing case: invalid_signature_no_participants
Testing case: sync_committee_rewards_duplicate_committee_max_effective_balance_only_participate_first_one
Testing case: random_high_participation_with_duplicates
Testing case: random_with_exits_with_duplicates
thread 'tests_process_sync_aggregate::test_operation' panicked at testing/ef-tests/tests/tests.rs:101:1:
assertion `left == right` failed: Post state mismatch in case random_with_exits_with_duplicates
...

Comment

Balance mismatch at index 0: actual=30999064780, expected=30999064790
Balance mismatch at index 1: actual=31997830296, expected=31997830306
Balance mismatch at index 2: actual=31997184677, expected=31997184667
Balance mismatch at index 3: actual=31997184677, expected=31997184667
Balance mismatch at index 4: actual=31997184677, expected=31997184667
Balance mismatch at index 5: actual=31997184677, expected=31997184667
Balance mismatch at index 6: actual=31997183349, expected=31997183359
Balance mismatch at index 7: actual=31999519537, expected=31999519527
Balance mismatch at index 8: actual=31997184677, expected=31997184667
Balance mismatch at index 9: actual=31999519537, expected=31999519527
...

Hacking with debugger, I found two interesting facts:

1. This test only fails for random_with_exits_with_duplicates. (Link for consensus-specs)
2. It differs about 10 gwei for every balance entry.