Aptos responded to yesterday's "network outage": it was not a transaction load issue, but caused by non-deterministic code, and a fix has been deployed

ChainCatcher news, Aptos released a report on the network outage that occurred yesterday, stating that the Aptos network began experiencing transaction delays around 16:15 PDT on October 18 (07:15 Beijing time on October 19). The transaction load was not the issue in this incident; submitted transactions were not lost, and no forks occurred. Non-deterministic code caused the problem, and a fix has been deployed. The issue was resolved around 12:30 Beijing time on October 19.

On August 22, the Aptos core codebase submitted performance-centric code changes, and on October 16, the FeeStatement feature went live, detailing transaction fees/refunds. The initial code changes introduced non-determinism that was revealed only by FeeStatement. Specifically, validators consistently believed that the gas budget for transactions was insufficient to execute them, and due to the non-determinism introduced in the August code changes, they could not reach a consensus on the amount of gas used up to that point.

After identifying the actual event output differences in non-deterministic transaction execution, the issue was traced back in the code to the FeeStatement event and code changes. Meanwhile, a developer began running transaction simulations through code changes to restore mapping changes, executing repeatedly to ensure consistent results. Additionally, the recovery from the code submission has been implemented, and Docker builds for validator operators have begun, followed by the release of a new version.