BR Tool Overview

When BR performs a backup operation, it first obtains the following information from PD:

• The current TS (timestamp) as the time of the backup snapshot
• The TiKV node information of the current cluster

According to these information, BR starts a TiDB instance internally to obtain the database or table information corresponding to the TS, and filters out the system databases (information_schema, performance_schema, mysql) at the same time.

According to the backup sub-command, BR adopts the following two types of backup logic:

• Full backup: BR traverses all the tables and constructs the KV range to be backed up according to each table.
• Single table backup: BR constructs the KV range to be backed up according a single table.

Finally, BR collects the KV range to be backed up and sends the complete backup request to the TiKV node of the cluster.

The structure of the request:

BackupRequest{
    ClusterId,      // The cluster ID.
    StartKey,       // The starting key of the backup (backed up).
    EndKey,         // The ending key of the backup (not backed up).
    StartVersion,   // The version of the last backup snapshot, used for the incremental backup.
    EndVersion,     // The backup snapshot time.
    StorageBackend, // The path where backup files are stored.
    RateLimit,      // Backup speed (MB/s).
}

After receiving the backup request, the TiKV node traverses all Region leaders on the node to find the Regions that overlap with the KV ranges in this request. The TiKV node backs up some or all of the data within the range, and generates the corresponding SST file.

After finishing backing up the data of the corresponding Region, the TiKV node returns the metadata to BR. BR collects the metadata and stores it in the backupmeta file which is used for restoration.

If StartVersion is not 0, the backup is seen as an incremental backup. In addition to KVs, BR also collects DDLs between [StartVersion, EndVersion). During data restoration, these DDLs are restored first.

If checksum is enabled when you execute the backup command, BR calculates the checksum of each backed up table for data check.

Types of backup files

Two types of backup files are generated in the path where backup files are stored:

• The SST file: stores the data that the TiKV node backed up.
• The backupmeta file: stores the metadata of this backup operation, including the number, the key range, the size, and the Hash (sha256) value of the backup files.

The format of the SST file name

The SST file is named in the format of storeID_regionID_regionEpoch_keyHash_cf, where

• storeID is the TiKV node ID;
• regionID is the Region ID;
• regionEpoch is the version number of the Region;
• keyHash is the Hash (sha256) value of the startKey of a range, which ensures the uniqueness of a key;
• cf indicates the Column Family of RocksDB (default or write by default).

During the data restoration process, BR performs the following tasks in order:

1. It parses the backupmeta file in the backup path, and then starts a TiDB instance internally to create the corresponding databases and tables based on the parsed information.

2. It aggregates the parsed SST files according to the tables.

3. It pre-splits Regions according to the key range of the SST file so that every Region corresponds to at least one SST file.

4. It traverses each table to be restored and the SST file corresponding to each tables.

5. It finds the Region corresponding to the SST file and sends a request to the corresponding TiKV node for downloading the file. Then it sends a request for loading the file after the file is successfully downloaded.

After TiKV receives the request to load the SST file, TiKV uses the Raft mechanism to ensure the strong consistency of the SST data. After the downloaded SST file is loaded successfully, the file is deleted asynchronously.

After the restoration operation is completed, BR performs a checksum calculation on the restored data to compare the stored data with the backed up data.