Nonce values stuck too low when using Nonce Tracker

We’ve noticed that sometimes a getTransactionCount hitting an INFURA node can return a value that is invalid when used as a nonce in a transaction.

{"code":-32010,"message":"Transaction nonce is too low. Try incrementing the nonce."}

When the Provider Engine Nonce tracker is used, one can end up in a loop of invalid nonces as the value from getTransactionCount is now cached. The nonce is never cleared from an error during sendTransaction, even if the nonce is the reason the sendTransaction failed.

[roderik]

I believe this might be a specific issue with Parity PoA and/or short blocktime setups and not Infura. We are running into the same issues on a permissioned setup that looks a lot like Kovan.

[danfinlay]

I’m not sure our new nonce logic could live in provider-engine by itself: It depends on having a list of sent/pending transactions, so it can increment over the highest nonce between that and the network’s latest confirmed tx for that account.

I think nonce tracking with provider-engine is probably going to continue to be inadequate for as long as it doesn’t also track sent/pending transactions.