This topic describes the architecture and storage mechanism of data backup.
Data backup architecture
Data backup preferentially uses a follower replica as the backup target. The data backup architecture is shown in the following figure.
Storage mechanism of data backup files
The native NFS protocol performs poorly in handling a large number of small files (for example, tens of millions of 2 MB files) in data backup. To address this, OceanBase Database introduced data aggregation storage in V4.x. Specifically, data backup files are packaged and stored in batches to reduce the number of small files. In this mechanism, data backup files include SSTable metadata, tablet metadata, and macroblock objects. By default, 2,048 data backup files (not exceeding 2 MB each) are aggregated into one batch. You can modify the number of data backup files in a batch by using the tenant-level parameter backup_data_file_size. The sizes of aggregated files vary because the sizes of metadata and macroblocks are different.