API Doc
Search Docs...
⌘ K
ACOSABSAVEANSABDR
  • ABDR synchronous replication and recovery technical whitepaper

How synchronous replication works

Synchronous replication creates a replica on the target side whose data is consistent with the source side in real time, achieving the data protection objective of RPO = 0. The process is centrally scheduled by controller-manager. Through efficient coordination between asynchronous replication jobs and ABS CDP jobs, the replication state evolves from initial alignment to real-time synchronization.

Initialization and replica creation

Synchronous replication first establishes a data baseline through a full asynchronous replication job. The system obtains source virtual machine configuration information and ABS volume snapshots, which record the full state at the current point in time, and generates the replica virtual machine in the target cluster through the storage network. At this point, the data state of the replica virtual machine is consistent with the state of the source side when the snapshot was created.

Data catch-up

During full initialization, service I/O continues on the source virtual machine. To eliminate the incremental data difference generated during initialization, the system enters the data catch-up phase and uses incremental asynchronous replication jobs and the ABS CDP mechanism to achieve dynamic alignment:

  1. Track (incremental tracking): Before incremental asynchronous replication starts, the system starts accurately recording all data changes that occur on the source side.

  2. Delta Copy (incremental alignment): After the incremental asynchronous replication job is complete, the system synchronizes the incremental data generated after the incremental snapshot to the replica side based on the data changes recorded in the Track phase. This step is designed to quickly close the data gap while services are running, so that the replica virtual machine and the source virtual machine become consistent for the first time.

Real-time synchronization

After historical incremental data is synchronized, the system enters the stable synchronous replication state. At this point, real-time data synchronization is implemented through the Mirror (dual-write) mechanism. Write I/O generated by the source virtual machine is written to both the source side and the target side, and the I/O returns success only after writes are complete on both sides. Through the dual-write mechanism, the system ensures that any subsequent writes remain consistent between the source side and the target side, achieving real-time replication with RPO = 0.