In iSCSI mode, ABS ensures minimal impact on I/O access during node, network, or service anomalies through multiple reliability mechanisms using Access VIP (Virtual IP) and iSCSI Redirector.
VIP and iSCSI Redirector ensure the client continuously connects to the currently available Access. All Access nodes are equivalent and ensure stable and consistent data access services. Therefore, the client can connect to any Access to achieve reliable data access.
In the I/O path, if the following role nodes encounter an issue, the impact on data I/O services is as follows:
VIP server (iSCSI redirector) node anomaly
ABS quickly switches the VIP service to another available node through the cluster state detection mechanism. The client does not need to be reconfigured; new probing path requests are automatically directed to the new VIP server. Established I/O paths do not require the participation of the VIP server, and therefore remain unaffected.
Access node anomaly
The Access node connected to a virtual machine is typically the node where the virtual machine resides. If the node experiences an overall anomaly, the virtual machine will trigger HA to rebuild on another node after a certain period. If there is only an issue with the Access service on the node, the iSCSI initiator detects a brief disconnection (the application does not perceive an I/O error, but an abrupt high latency instead) and requests a new Access address from the VIP server (iSCSI Redirector). After the client establishes a connection with the new Access, it can continue to provide service.
ChunkServer anomaly
Meta rapidly executes shard eviction and removes faulty ChunkServers when necessary to ensure I/O continuity while initiating recovery tasks to repair data.
Meta node anomaly
The ZooKeeper cluster automatically elects a new Meta leader node to quickly restore metadata services.
