Teradata has many features to protect you against a potential data loss. Each of this features protects your data in a different way, and on a different level of granularity. Default activates some of the features; others have to be turned on explicitly.
1. The Teradata Transient Journal – Transaction Failure Protection
All changes of a transaction are recorded in the so-called transaction log or Teradata transient journal.
While the majority of all Teradata tables are distributed across all AMPs by the Primary Index values, the transient journal is not. Each AMP keeps its transient journal.
Whenever a row is changed during a transaction, a copy of the original row is stored in the transient journal and ensures data integrity. It's enabling the system to do a roll back of the table to the initial state, for example, in the case of error.
The rows are only temporarily stored in the transient journal, and deleted as soon as the transaction is successfully committed. The parsing engine takes care that all AMPs successfully finished their work before any transaction is committed.
2. Teradata Fallback – AMP Failure Protection
The transient journal protects your data only against transaction failures. In the case of an AMP failure, is possible that your data is not accessible anymore until the AMP is coming back online. The fallback protection strategy was implemented, to ensure data accessibility in such cases,
Fallback protection means that each row is stored twice. A second hash map is used to determine for each row an AMP, which will save a copy of the row. Fallback protected tables require twice as much space.
The advantage of fallback protection is that your data may be accessible in case of an AMP failure, as the auxiliary AMP will take over its tasks until the failed AMP comes back online.
Only in the unlikely event, that the backup AMP fails as well, your table ultimately will become inaccessible.
To achieve maximum protection, AMP’s are grouped together in clusters. Main AMP and fallback AMP always belong to the same cluster and protect each other.
Furthermore, the main AMP and the fallback AMP are never stored together physically at the same node, which is a prudent design choice on the nature of hardware failures. Even if a complete node fails, the fallback protection still allows us to access the data!
3. Raid 1 – Disk Failure Protection
Fallback protection helps in case of the failure of the Unix processes (VPROC, AMP), belonging to the Teradata System.
Teradata uses the public RAID 1 protection (mirroring) If you want to protect the data against disk failures. Although each AMP has exactly one virtual disk assigned (VDISK), several physical disks make up this VDISK. Half of the discs keep the mirrored data.
As in the case of fallback protection, the cost is the usage of twice as much as disk space.
The RAID 1 configuration ensures enhanced data accessibility and security. If a single disk fails, data will serve from the mirror disk. Only if both the primary disk and the mirror disk fail, and no fallback protection was enabled, the data would be lost.
4. The Teradata Clique – Node Failure Protection
This protection mechanism adds another level of security to the Teradata System. Clique configuration protects against complete node failures. Cliques are grouping nodes together like clusters group the AMPs together.
The AMPs of the failed node will be migrated to another node belonging to the same clique and stays fully functional. The node taking over the AMPs will have more work to do, which will negatively impact the performance.
Nevertheless, to overcome this restriction, Teradata offers hot standby nodes for the sole purpose of taking over the AMPs of a failed node. As these AMPs are not engaged in routine operations in the ordinary course of business, no performance degradation occurs.
Keep in mind that before the AMP migration takes place, a system restart is required, followed by another restart as soon as the failed node goes online again.