All Products
Document Center

How the backup process works

Last Updated: Aug 25, 2020

DBS offers features such as full backup, real-time incremental backup, and data recovery. This section describes the impact on database performance during the backup process.


  1. When performing database backups, DBS needs to read data from your database, which affects database performance.
  2. If you have a standby database, we recommend that you perform backups on the standby database to avoid any performance issues on the primary database.
  3. If you use RDS databases, you can only perform backups on the primary database.
  4. We recommend that you perform full backups during off-peak hours to minimize the impact on your service.
  5. The beta release of DBS provides a fixed number of backup threads and maintains a stable backup speed. The official release will allow users to dynamically control the backup speed.
  6. By default, DBS performs database recovery at the maximum speed. We recommend that you use an idle database as the backup target.

Full backup


When performing full backups, DBS first splits each table into fragments and uses multiple threads to read data in parallel. The service will execute SQL statements as below. As data is stored on disks, data reading impacts database I/O performance. DBS will not lock the database during backup process. This reduces the impact on database performance.

Note: If the backup source database has tables without a primary key, or tables with a string type primary key or composite key, the SQL performance may be compromised.

  1. SELECT col1,col2,col3,coln
  2. FROM db1.tab1
  3. WHERE id >= xxx
  4. AND id < xxx
  5. ORDER BY id ASC


Instance type Parallel threads
All instance types 4~16

Incremental backup


When performing incremental backups, DBS reads database transaction logs in real time. Log reading speed is dynamically adjusted in response to log generation speed. Log reading has little impact on database CPU performance. As database logs are stored in the log cache, the impact of log reading on database I/O performance is minimal.


Instance type Parallel threads
All instance types 1~2