Snapshot of HBase table

by Chongxin Li for Facebook

Snapshot of HBase table is useful in production. Although data stored on Hadoop has replicas to protect from failed servers, this does not prevent use from software bugs or misoperations that might delete or alter data in ways we did not plan. To create a snapshot, two problems should be taken into account: snapshot request should be returned as soon as possible and snapshot must have a low impact on the cluster. This proposal provides a preliminary design of snapshot implementation on HBase.