Cache Node Fault/Recovery

BoostIO are deployed on compute nodes to provide distributed read/write cache services. The methods for troubleshooting the compute nodes are described as follows.

**Table 1** Cache node fault scenarios
Scenario	Impact	Handling Method	Remarks
Cache node fault	Copy data in the write cache is lost. Object data in the read cache is lost. The SDK fails to distribute requests.	The ZooKeeper heartbeat detects that the cache process is faulty and instructs the cluster management module to update the view and then release the new view. Partition data affected by the process fault is forcibly evicted to the back-end storage.	If a cache node is temporarily faulty, only the statuses of the node view and partition view are changed. If a cache node is permanently faulty, the cluster management module removes the node from the cluster and changes both the node view and partition view. The time window for determining temporary faults is configurable.
Cache node recovery	Read/write cache functions need to be restored.	Temporary fault: The ZooKeeper heartbeat detects that the cache process is recovered and instructs the cluster management module to update the view and then release the new view. Permanent fault: The removed node is added to the cluster as a new node for expansion.	None

Parent topic: Troubleshooting