Node is really slow to process transactions after fork

[MexicanAce]

🐛 Bug Report for zkSync Era In-Memory Node

📝 Description

After starting up the node with era_test_node fork testnet, deploying a smart contract and processing transactions seem to be really slow (6+ seconds). Even the make test-e2e command practically times out attempting to mine 100 blocks.

🔄 Reproduction Steps

1. Run era_test_node fork testnet
2. Run make test-e2e
3. Wait until it finishes

🤔 Expected Behavior

This should be almost as fast as era_test_node run, and complete all end-to-end tests in less than 2 seconds.

😯 Current Behavior

make test-e2e takes 11+ seconds just for two tests, and times out when attempting to mine too many blocks.

📎 Log Output

$ hardhat test --network zkSyncTestnet


  evm_mine
    ✔ Should mine one block (6256ms)

  evm_increaseTime
    ✔ Should increase current timestamp of the node (5142ms)

  evm_setNextBlockTimestamp
    ✔ Should set current timestamp of the node to specific value (2909ms)

  evm_setTime
    ✔ Should set current timestamp of the node to specific value (2890ms)

  hardhat_setBalance
    ✔ Should update the balance of an account (118ms)

  hardhat_setNonce
    ✔ Should update the nonce of an account (650ms)

  hardhat_mine
^C

[nbaztec]

I looked into it and it seems in fork mode, there are many storage read calls that are made (which isn't done on the run mode) especially during block creation/mining. As of now I do not see how to not have this behavior (with the exception of pre-fetching the entire state). Caching will not help since most of these storage keys are only requested once but there are ~20 of those and each is a network call.