 | ||
| Data node resiliency | ||
|---|---|---|
| Prev | Compute node and data node resiliency | Next |
The HyperScale data node acts as a secondary storage tier for storing vDisk versions. vDisk versions are periodically synchronized with the data node. If a compute node fails, the vDisk version is read from the data node. For a vDisk having the VRTSSilver flavor and above, data is reflected from one data node (primary) to another data node (secondary).
HyperScale provides resiliency at data plane in case of following data node failures. There is minimal impact on applications and data management services such as snapshots and recoveries. Alerts are raised for all these failures in the Alerts/Events area of the HyperScale dashboard ().
If a data node's solid-state drive (SSD) or hard disk drive (HDD) fails, ongoing snapshots are aborted, and the data node goes into the Down state. You must remove the data node from the configuration. After you replace the faulty disk, re-add the same data node in the configuration.
See Removing a data node.
See Adding a data node.
If the data node's data network card or data network cable fails, the data node cannot communicate with the controller. The data node goes into the Down state. After you fix the network communication issue, remove the data node from the configuration and add it again.
If the storage service (hyperscale-dmld) fails, the data node goes into the Down state. You can remove failed data node and add a new one or fix the existing node and add it again.
If the compute node cannot send episodic data synchronization messages to the primary data node.
In each of these failures, the vDisks associated with the failed data node are marked as resiliency degraded and its ownership is transferred automatically to a peer data node. The peer data node starts serving the next snapshots and recovery reads. Any ongoing requests at the time of failure restart on the new primary data node after the ownership change.
To restore the reflection back for degraded vDisks, you must remove the failed data node, replace it, and perform a data node synchronization.