External Storages

Backup & Restore (BR), TiDB Lighting, and Dumpling support reading and writing data on the local filesystem and on Amazon S3. BR also supports reading and writing data on the Google Cloud Storage (GCS). These are distinguished by the URL scheme in the --storage parameter passed into BR, in the -d parameter passed into TiDB Lightning, and in the --output (-o) parameter passed into Dumpling.

Schemes

The following services are supported:

URL parameters

Cloud storages such as S3 and GCS sometimes require additional configuration for connection. You can specify parameters for such configuration. For example:

• Use Dumpling to export data to S3:

  ./dumpling -u root -h 127.0.0.1 -P 3306 -B mydb -F 256MiB \
      -o 's3://my-bucket/sql-backup?region=us-west-2'
  

• Use TiDB Lightning to import data from S3:

  ./tidb-lightning --tidb-port=4000 --pd-urls=127.0.0.1:2379 --backend=local --sorted-kv-dir=/tmp/sorted-kvs \
      -d 's3://my-bucket/sql-backup?region=us-west-2'
  

• Use TiDB Lightning to import data from S3 (using the path style in the request mode):

  ./tidb-lightning --tidb-port=4000 --pd-urls=127.0.0.1:2379 --backend=local --sorted-kv-dir=/tmp/sorted-kvs \
      -d 's3://my-bucket/sql-backup?force-path-style=true&endpoint=http://10.154.10.132:8088'
  

• Use BR to back up data to GCS:

  ./br backup full -u 127.0.0.1:2379 \
      -s 'gcs://bucket-name/prefix'
  

S3 URL parameters

1. $AWS_ACCESS_KEY_ID and $AWS_SECRET_ACCESS_KEY environment variables
2. $AWS_ACCESS_KEY and $AWS_SECRET_KEY environment variables
3. Shared credentials file on the tool node at the path specified by the $AWS_SHARED_CREDENTIALS_FILE environment variable
4. Shared credentials file on the tool node at ~/.aws/credentials
5. Current IAM role of the Amazon EC2 container
6. Current IAM role of the Amazon ECS task

GCS URL parameters

When credentials-file is not specified, the migration tool will try to infer the credentials from the environment, in the following order:

1. Content of the file on the tool node at the path specified by the $GOOGLE_APPLICATION_CREDENTIALS environment variable
2. Content of the file on the tool node at ~/.config/gcloud/application_default_credentials.json
3. When running in GCE or GAE, the credentials fetched from the metadata server.

Command-line parameters

In addition to the URL parameters, BR and Dumpling also support specifying these configurations using command-line parameters. For example:

./dumpling -u root -h 127.0.0.1 -P 3306 -B mydb -F 256MiB \
    -o 's3://my-bucket/sql-backup' \
    --s3.region 'us-west-2'

If you have specified URL parameters and command-line parameters at the same time, the URL parameters are overwritten by the command-line parameters.

S3 command-line parameters

GCS command-line parameters

BR sending credentials to TiKV

By default, when using S3 and GCS destinations, BR will send the credentials to every TiKV nodes to reduce setup complexity.

However, this is unsuitable on cloud environment, where every node has their own role and permission. In such cases, you need to disable credentials sending with --send-credentials-to-tikv=false (or the short form -c=0):

./br backup full -c=0 -u pd-service:2379 -s 's3://bucket-name/prefix'

When using SQL statements to back up and restore data, you can add the SEND_CREDENTIALS_TO_TIKV = FALSE option:

BACKUP DATABASE * TO 's3://bucket-name/prefix' SEND_CREDENTIALS_TO_TIKV = FALSE;

This option is not supported in TiDB Lightning and Dumpling, because the two applications are currently standalone.