[config change] Improve BlocksReExecutor implementation

[ganeshvanahalli]

This PR changes the implementation of blocks re-execution mainly to use a separate stateDB during each block's re-execution and to verify that the resulting state successfully commits.

Config option --blocks-reexecutor.blocks-per-thread is renamed to --blocks-reexecutor.min-blocks-per-thread.
A new config option trie-clean-limit (memory allowance in MB to use for caching trie nodes in memory) is added.

Pulls geth PR- OffchainLabs/go-ethereum#364
Resolves NIT-2810

[eljobe]

LGTM

[magicxyyz]

looks good, but we should configure cleans cache + we potentially can simplify / clean up arbitrum.AdvanceStateUpToBlock and follow the release func pattern

[magicxyyz]

hashdb.Defaults have CleanCacheSize = 0 (cleans cache disabled). We should use cleans cache here to have better perfomance, similarly to:

nitro/cmd/staterecovery/staterecovery.go

Lines 39 to 44 in 7a3d6e6

	hashConfig := *hashdb.Defaults
	hashConfig.CleanCacheSize = cacheConfig.TrieCleanLimit * 1024 * 1024
	trieConfig := &triedb.Config{
	Preimages: false,
	HashDB: &hashConfig,
	}

[ganeshvanahalli]

Done. I've added a new config field trie-clean-limit and have set the default value for this as 600 sourcing from here let me know if it should be adjusted

[magicxyyz]

how about following the release function pattern?

• we could return arbitrum.StateReleaseFunc that will call Dereference. That can be something like func() {triedb.Dereference(root)})

• we would need to move the call to release func returned from arbitrum.FindLastAvailableState to the thread started with LaunchThread so as it would be called after AdvanceStateUpToBlock or even better - we could pass the startStateRelease to advanceStateUpToBlock and call it after we reference next state

• arbitrum.AdvanceStateUpToBlock was used only here, so we can modify it to commit statedb after each reexecuted block and check the hashes OR we can just remove arbitrum.AdvanceStateUpToBlock

• we can return release func also from advanceStateUpToBlock and have something like:

startState, startHeader, startStateRelease, err := arbitrum.FindLastAvailableState(ctx, s.blockchain, s.stateFor, s.blockchain.GetHeaderByNumber(start), nil, -1)
...
s.LaunchThread(func(ctx context.Context) {
    ...
    _, release, err := advanceStateUpToBlock(..., startHeader, startStateRelease, ...)
    ...
    defer release()
    ...
}

[magicxyyz]

note to startStateRelease - we should make sure that it will be called also in case of an error

[ganeshvanahalli]

Agreed, its a great idea to utilize StateReleaseFuncs here!
Also removed AdvanceStateUpToBlock from geth.