Initial nested cache Context StateDB implementation

[drklee3]

Description

• Use CacheContext for StateDB snapshots
• Update tests & benchmarks to use actual x/evm keeper instead of mock keeper

This will be later moved to the Kava repo when it's fully de-coupled from x/evm

Benchmarks

                                    │  journal.txt  │              nested-ctx.txt              │
                                    │    sec/op     │     sec/op       vs base                   │
TokenTransfer-12                       128.5µ ±  1%      167.6µ ±  2%     +30.44% (p=0.000 n=10)
EmitLogs-12                            3.128m ±  1%      3.247m ±  1%      +3.79% (p=0.000 n=10)
TokenTransferFrom-12                   127.4µ ±  0%      192.0µ ±  1%     +50.74% (p=0.000 n=10)
TokenMint-12                           123.1µ ±  1%      161.2µ ±  1%     +31.03% (p=0.000 n=10)
MessageCall-12                         49.50m ±  0%     384.17m ±  2%    +676.11% (p=0.000 n=10)
SetParams-12                           1.449µ ±  1%      1.480µ ±  1%      +2.14% (p=0.000 n=10)
GetParams-12                           2.309µ ±  0%      2.350µ ±  1%      +1.80% (p=0.000 n=10)
ApplyTransaction-12                    68.69µ ±  1%      85.50µ ±  1%     +24.47% (p=0.000 n=10)
ApplyTransactionWithLegacyTx-12        69.42µ ±  1%      86.44µ ±  1%     +24.52% (p=0.000 n=10)
ApplyTransactionWithDynamicFeeTx-12    69.34µ ±  0%      85.70µ ±  1%     +23.60% (p=0.000 n=10)
ApplyMessage-12                        15.48µ ±  0%      34.52µ ±  1%    +122.96% (p=0.000 n=10)
ApplyMessageWithLegacyTx-12            15.56µ ±  0%      34.59µ ±  0%    +122.25% (p=0.000 n=10)
ApplyMessageWithDynamicFeeTx-12        15.61µ ±  1%      34.23µ ±  1%    +119.29% (p=0.000 n=10)
CreateAccountNew-12                    2.313µ ±  2%     13.758µ ±  2%    +494.68% (p=0.000 n=10)
CreateAccountExisting-12               168.2n ±  4%    10293.0n ±  1%   +6021.32% (p=0.000 n=10)
AddBalance-12                          153.2n ±  2%    23670.0n ±  1%  +15350.39% (p=0.000 n=10)
SetCode-12                             565.2n ±  1%    11346.0n ±  1%   +1907.61% (p=0.000 n=10)
SetState-12                            565.9n ±  1%    11400.0n ±  2%   +1914.67% (p=0.000 n=10)
AddLog-12                             107.30n ± 13%      82.33n ±  7%     -23.28% (p=0.000 n=10)
Snapshot-12                            968.4n ±  0%     9351.5n ±  3%    +865.62% (p=0.000 n=10)
SubBalance-12                          155.2n ±  2%    27699.0n ±  1%  +17753.05% (p=0.000 n=10)
SetNonce-12                            60.87n ±  6%   10371.50n ±  2%  +16937.37% (p=0.000 n=10)
AddRefund-12                          30.010n ± 12%      4.748n ± 27%     -84.18% (p=0.000 n=10)
Suicide-12                             643.4n ±  4%    72174.0n ±  1%  +11118.47% (p=0.000 n=10)
geomean                                5.448µ            26.82µ          +392.20%

[nddeluca]

Benchmarks are quite a bit slower -- I assume some of those now hit keeper methods instead of appending to journal/adding a map entry? My guess is some of those benchmarks become a little closer if the time taken in Commit is considered.

Could we design a benchmark that snapshots at various depths (maybe 10, 1000, 10000) with some changes at each and calls Commit at end? Be useful to get a better measure of performance penalty of nested cache contexts

[drklee3]

A lot slower at high number of layers with this benchmark:

ethermint/x/evm/statedb/benchmark_test.go

Lines 39 to 66 in cf42160

	func BenchmarkNestedSnapshot(b *testing.B) {
	benches := []int{1, 4, 10, 100, 1000, 10000}
	for _, layers := range benches {
	b.Run(fmt.Sprintf("%d layers", layers), func(b *testing.B) {
	suite := GetTestSuite(b)
	for i := 0; i < b.N; i++ {
	b.StopTimer()
	db := statedb.New(suite.Ctx, suite.App.EvmKeeper, emptyTxConfig)
	// Create layers of nested snapshots
	for i := 0; i < layers; i++ {
	db.Snapshot()
	// Some state change each snapshot
	key := common.BigToHash(big.NewInt(int64(i + 1)))
	value := common.BigToHash(big.NewInt(int64(i + 1)))
	db.SetState(address, key, value)
	}
	b.StartTimer()
	require.NoError(b, db.Commit())
	}
	})
	}
	}

Benchmark takes forever to run since the setup with paused bench timer is much longer than the actual timed Commit() part

goos: darwin
goarch: arm64
pkg: github.com/evmos/ethermint/x/evm/statedb
                               │ BenchmarkNestedSnapshot-journal.txt │          BenchmarkNestedSnapshot-ctx.txt           │
                               │               sec/op                │       sec/op         vs base                       │
NestedSnapshot/1_layers-12                              556.3n ± ∞ ¹         5375.0n ± ∞ ¹       +866.21% (p=0.008 n=5)
NestedSnapshot/4_layers-12                              534.2n ± ∞ ¹        12952.0n ± ∞ ¹      +2324.56% (p=0.008 n=5)
NestedSnapshot/10_layers-12                             568.9n ± ∞ ¹        34901.0n ± ∞ ¹      +6034.82% (p=0.008 n=5)
NestedSnapshot/100_layers-12                            621.3n ± ∞ ¹      1540187.0n ± ∞ ¹    +247797.47% (p=0.008 n=5)
NestedSnapshot/1000_layers-12                           776.4n ± ∞ ¹    150212761.0n ± ∞ ¹  +19347241.71% (p=0.008 n=5)
NestedSnapshot/10000_layers-12                          1.414µ ± ∞ ¹   17185876.541µ ± ∞ ¹              ~ (p=0.095 n=2+5)
geomean                                                 697.6n                1.459m          +209092.25%
¹ need >= 6 samples for confidence interval at level 0.95

                               │ BenchmarkNestedSnapshot-journal.txt │           BenchmarkNestedSnapshot-ctx.txt            │
                               │                B/op                 │         B/op          vs base                        │
NestedSnapshot/1_layers-12                               25.00 ± ∞ ¹          6619.00 ± ∞ ¹      +26376.00% (p=0.008 n=5)
NestedSnapshot/4_layers-12                               24.00 ± ∞ ¹         17025.00 ± ∞ ¹      +70837.50% (p=0.008 n=5)
NestedSnapshot/10_layers-12                              25.00 ± ∞ ¹         43000.00 ± ∞ ¹     +171900.00% (p=0.008 n=5)
NestedSnapshot/100_layers-12                             25.00 ± ∞ ¹       1694623.00 ± ∞ ¹    +6778392.00% (p=0.008 n=5)
NestedSnapshot/1000_layers-12                            25.00 ± ∞ ¹     149449860.00 ± ∞ ¹  +597799340.00% (p=0.008 n=5)
NestedSnapshot/10000_layers-12                           30.00 ± ∞ ¹   14602752152.00 ± ∞ ¹               ~ (p=0.095 n=2+5)
geomean                                                  25.60                1.543Mi          +6319793.96%
¹ need >= 6 samples for confidence interval at level 0.95

                               │ BenchmarkNestedSnapshot-journal.txt │           BenchmarkNestedSnapshot-ctx.txt           │
                               │              allocs/op              │      allocs/op       vs base                        │
NestedSnapshot/1_layers-12                               1.000 ± ∞ ¹          98.000 ± ∞ ¹       +9700.00% (p=0.008 n=5)
NestedSnapshot/4_layers-12                               1.000 ± ∞ ¹         260.000 ± ∞ ¹      +25900.00% (p=0.008 n=5)
NestedSnapshot/10_layers-12                              1.000 ± ∞ ¹         644.000 ± ∞ ¹      +64300.00% (p=0.008 n=5)
NestedSnapshot/100_layers-12                             1.000 ± ∞ ¹       15990.000 ± ∞ ¹    +1598900.00% (p=0.008 n=5)
NestedSnapshot/1000_layers-12                            1.000 ± ∞ ¹     1071292.000 ± ∞ ¹  +107129100.00% (p=0.008 n=5)
NestedSnapshot/10000_layers-12                           1.000 ± ∞ ¹   101818490.000 ± ∞ ¹               ~ (p=0.095 n=2+5)
geomean                                                  1.000                17.49k          +1748852.62%
¹ need >= 6 samples for confidence interval at level 0.95

[evgeniy-scherbina]

LGTM