Fault-tolerant system and method of managing transaction failures in hierarchies

摘要

A fault-tolerant technique for managing cross-hierarchy failure of a process of transactions is disclosed. The process has blocks of transactions and one or more of the transactions may be subtransaction of others. The technique includes enabling a block having one or more transactions in the process to notice, e.g., the failure in a transaction not in the process, wherein the noticed failure will cause a failure of a transaction (FFT) in the process. The technique also includes, upon the failure of the transaction FFT in the process, determining an in-progress closest recoverable ancestor (ICRA) in a block in the process. The ICRA is the nearest transaction to the failed transaction FFT, in the block on an ancestor tree of the FFT, that is in- block-recoverable and in- progress. The technique further includes recovering from the first process failed transaction by undoing from the ICRA to that failed transaction. The undoing step includes aborting or compensating one or more transactions, or both. In the case that the closest recoverable ancestor (CRA), i.e., the nearest transaction to the FFT is not in- progress, the closest in-progress ancestor (CIP) of the CRA and the closest recoverable in-progress ancestor (ERA) of the CIP are identified. The CIP is the nearest transaction to the CRA, on the ancestor tree of the FFT, that is neither committed nor aborted as viewed from outside the first process. The ERA is the nearest in-block recoverable transaction to the CIP on the ancestor tree of the FFT and in progress in a block containing the CIP.