bug(jerigon): prover panicking onmpt_read_hash_node from Jerigon payload

[Nashtare]

I'm currently running a prover in witness-only mode to catch up with Ethereum mainnet on top of v0.6.0 (commit f531584), with a Jerigon node running on commit 83e0f2fa8c8f6632370e20fef7bbc8a4991c73c8 (@praetoriansentry to confirm).

Two batches failed out of 600, each with a KernelPanic on one txn when hitting mpt_read_hash_node, which shouldn't happen with the Jerigon tracer (but may happen in some edge cases uncovered by the native tracer).

Attached are the extracted block witnesses from each batch (originally packing 20 blocks per batch), as well as the isolated GenerationInputs of the failing txns.

block payloads:

• 20342679.json
• 20342681.json

individual inputs:

• b20342679_txn_88_input.json
• b20342681_txn_54_input.json

[Nashtare]

Regarding txn 88 in block 20342679 (txn hash 0x33770a49...), the txn is failing on mainnet because of an OOG issue during the execution of a CALL. The other txn is exactly the same, just repeated 2 blocks after.

I suspect missing account information because of the aborted execution. The address passed as arg of the CALL is 0xc126624ca2ad52db3abc417fa0a21da90303745e which the prover is trying to access whether the associated account is dead or not.

IIRC, @frisitano and @wborgeaud had done tweaks on CALL gas charges for related failure cases. Not sure whether this is a regression or another edge case that wasn't covered yet.

[praetoriansentry]

a Jerigon node running on commit 83e0f2fa8c8f6632370e20fef7bbc8a4991c73c8

Confirmed