A non-transitory computer readable medium comprising a plurality of machine-readable instructions which when executed by one or more processors of a server are adapted to cause the server to perform a method comprising:
- receiving, by a main coordinator, a distributed transaction comprising a plurality of processes for processing, wherein the processing uses a two-phase commit protocol;
- preparing, by main coordinator, at least one participant comprising at least one of the plurality of processes to prepare for execution by the at least one participant on at least one resource;
- receiving, by the main coordinator, information comprising prepared processes of the at least one of the plurality of processes prior to execution of the prepared processes on the at least one resource by the at least one participant
- receiving an initial state for each of the at least one resource prior to the prepared processes for the at least one resource;
- generating a transaction log for the distributed transaction for the at least one participant using the information, wherein the transaction log comprises the prepared processes and the initial state for the each of the at least one resource;
- storing, by the main coordinator, the transaction log to a semi-durable storage of the main coordinator, wherein the semi-durable storage is remote and separate from the at least one participant, and wherein a first write speed of the semi-durable storage is faster than a second write speed of a durable storage of the main coordinator;
- replicating the transaction log on at least one additional semi-durable storage for each of the at least one participant, wherein the at least one additional semi-durable storage is further configured to retain in the transaction log for a time after a system failure;
- determining, during processing the distributed transaction, the executed processes in the update information;
- removing the executed processes from the transaction log in the semi-durable storage and the transaction log replicated on the at least one additional semi-durable storage to generate a culled transaction log in the semi-durable storage of the main coordinator and a participant log in the each of the at least one additional semi-durable storage of the each of the at least one participant;
- in response to determining the system failure, executing a flush process to store the culled transaction log to the durable storage of the main coordinator, wherein the each of the at least one participant further execute the flush process to store the culled transaction log on at least one additional durable storage of the at least one participant;
- retrieving the culled transaction log from the durable storage, wherein the culled transaction log comprises a commit state for the at least one participant;
- determining that the at least one participant has recovered after the system failure;
- requesting a status of the at least one participant based on the participant log stored to the at least one additional durable storage of the at least one participant;
- and
- determining whether to rollback or commit the prepared processes for the each of the at least one resource based on the culled transaction log and the participant log for the each of the at least one participant.