[RPC] witness generation causes RPC node latest block/batch information to drift

[hexoscott]

It appears that asking an RPC node for latest block/batch information whilst an unwind is happening is causing the responses to also roll back.

This is happening when the network is halted at the sequencer so there are no new blocks being produced.

[hexoscott]

I have attempted the following local setup:

• 1 sequencer
• 1 RPC node consuming DS from sequencer
• Spammer script to keep sending some EOA transfers to the sequencer
• Bash script to keep calling eth_blockNumber and zkevm_batchNumber on repeat

With this running I can call for the witness for batch 2 which performs a full unwind of a number of batches and the script reporting the latest block and batch doesn't alter the output during this process.

[hexoscott]

After inspecting the logs of the RPC node from this test where it was first observed , this is related to a problem that is being fixed elsewhere that causes the RPC to go into an unwind loop when the sequencer is halted.

[hexoscott]

This was something else, the sequencer was reporting one block lower than the actual datastream for zkevm_getLatestDataStreamBlock which caused the RPC node to go into an unwind loop whilst a network was halted. The issue on the sequencer side has since been resolved and was proved off in the mainnet upgrade to fork12.