Use Dumpling to Export Data

This document introduces the data export tool - Dumpling. Dumpling exports data stored in TiDB/MySQL as SQL or CSV data files and can be used to make a logical full backup or export. Dumpling also supports exporting data to Amazon S3.

You can get Dumpling using TiUP by running tiup install dumpling. Afterwards, you can use tiup dumpling ... to run Dumpling.

The Dumpling installation package is included in the TiDB Toolkit. To download the TiDB Toolkit, see Download TiDB Tools.

For detailed usage of Dumpling, use the --help option or refer to Option list of Dumpling.

When using Dumpling, you need to execute the export command on a running cluster.

TiDB also provides other tools that you can choose to use as needed.

• For backups of SST files (key-value pairs) or backups of incremental data that are not sensitive to latency, refer to BR.
• For real-time backups of incremental data, refer to TiCDC.
• All exported data can be imported back to TiDB using TiDB Lightning.

Dumpling has the following advantages:

• Support exporting data in multiple formats, including SQL and CSV.
• Support the table-filter feature, which makes it easier to filter data.
• Support exporting data to Amazon S3 cloud storage.
• More optimizations are made for TiDB:
  • Support configuring the memory limit of a single TiDB SQL statement.
  • If Dumpling can access the PD address and the INFORMATION_SCHEMA.CLUSTER_INFO table of the TiDB cluster, Dumpling supports automatically adjusting the GC safe point time to block GC for TiDB v4.0.0 and later versions.
  • Use TiDB's hidden column _tidb_rowid to optimize the performance of concurrent data export from a single table.
  • For TiDB, you can set the value of tidb_snapshot to specify the time point of the data backup. This ensures the consistency of the backup, instead of using FLUSH TABLES WITH READ LOCK to ensure the consistency.

Export data from TiDB or MySQL

Required privileges

• PROCESS: Required to query the cluster information to obtain the PD address and then control GC via the PD.
• SELECT: Required when exporting tables.
• RELOAD: Required when using consistency flush. Note that only TiDB supports this privilege. When the upstream is an RDS database or a managed service, you can ignore this privilege.
• LOCK TABLES: Required when using consistency lock. This privilege must be granted for all the databases and tables to be exported.
• REPLICATION CLIENT: Required when exporting metadata to record data snapshot. This privilege is optional and you can ignore it if you do not need to export metadata.

Export to SQL files

This document assumes that there is a TiDB instance on the 127.0.0.1:4000 host and that this TiDB instance has a root user without a password.

Dumpling exports data to SQL files by default. You can also export data to SQL files by adding the --filetype sql flag:

dumpling -u root -P 4000 -h 127.0.0.1 --filetype sql -t 8 -o /tmp/test -r 200000 -F 256MiB

In the command above:

• The -h, -P, and -u option respectively mean the address, the port, and the user. If a password is required for authentication, you can use -p $YOUR_SECRET_PASSWORD to pass the password to Dumpling.
• The -o (or --output) option specifies the export directory of the storage, which supports an absolute local file path or an external storage URI.
• The -t option specifies the number of threads for the export. Increasing the number of threads improves the concurrency of Dumpling and the export speed, and also increases the database's memory consumption. Therefore, it is not recommended to set the number too large. Usually, it's less than 64.
• The -r option enables the in-table concurrency to speed up the export. The default value is 0, which means disabled. A value greater than 0 means it is enabled, and the value is of INT type. When the source database is TiDB, a -r value greater than 0 indicates that the TiDB region information is used for splitting, and reduces the memory usage. The specific -r value does not affect the split algorithm. When the source database is MySQL and the primary key is of the INT type, specifying -r can also enable the in-table concurrency.
• The -F option is used to specify the maximum size of a single file (the unit here is MiB; inputs like 5GiB or 8KB are also acceptable). It is recommended to keep its value to 256 MiB or less if you plan to use TiDB Lightning to load this file into a TiDB instance.

URI formats of the storage services

This section describes the URI formats of the storage services, including Amazon S3, GCS, and Azure Blob Storage. The URI format is as follows:

[scheme]://[host]/[path]?[parameters]

For more information, see URI Formats of External Storage Services.

Export to CSV files

You can export data to CSV files by adding the --filetype csv argument.

When you export data to CSV files, you can use --sql <SQL> to filter the records with the SQL statements. For example, you can export all records that match id < 100 in test.sbtest1 using the following command:

./dumpling -u root -P 4000 -h 127.0.0.1 -o /tmp/test --filetype csv --sql 'select * from `test`.`sbtest1` where id < 100' -F 100MiB --output-filename-template 'test.sbtest1.{{.Index}}'

In the command above:

• The --sql option can be used only for exporting to CSV files. The command above executes the SELECT * FROM <table-name> WHERE id <100 statement on all tables to be exported. If a table does not have the specified field, the export fails.

• When you use the --sql option, Dumpling cannot obtain the exported table and schema information. You can specify the file name format of the CSV files using the --output-filename-template option, which facilitates the subsequent use of TiDB Lightning to import the data file. For example, --output-filename-template='test.sbtest1.{{.Index}}' specifies that the exported CSV files are named as test.sbtest1.000000000 or test.sbtest1.000000001.

• You can use options like --csv-separator and --csv-delimiter to configure the CSV file format. For details, refer to the Dumpling option list.

Compress the exported data files

You can use the --compress <format> option to compress the CSV and SQL data and table structure files exported by Dumpling. This parameter supports the following compression algorithms: gzip, snappy, and zstd. The compression is disabled by default.

• This option only compresses individual data and table structure files. It cannot compress the entire folder and generate a single compressed package.
• This option can save disk space, but it also slows down the export speed and increases CPU consumption. Use this option with caution in scenarios where the export speed is critical.
• For TiDB Lightning v6.5.0 and later versions, you can use compressed files exported by Dumpling as the data source without additional configuration.

Format of exported files

• metadata: The start time of the exported files and the position of the master binary log.

  cat metadata
  

  Started dump at: 2020-11-10 10:40:19
  SHOW MASTER STATUS:
          Log: tidb-binlog
          Pos: 420747102018863124
  Finished dump at: 2020-11-10 10:40:20
  

• {schema}-schema-create.sql: The SQL file used to create the schema

  cat test-schema-create.sql
  

  CREATE DATABASE `test` /*!40100 DEFAULT CHARACTER SET utf8mb4 */;
  

• {schema}.{table}-schema.sql: The SQL file used to create the table

  cat test.t1-schema.sql
  

  CREATE TABLE `t1` (
    `id` int(11) DEFAULT NULL
  ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;
  

• {schema}.{table}.{0001}.{sql|csv}: The date source file

  cat test.t1.0.sql
  

  /*!40101 SET NAMES binary*/;
  INSERT INTO `t1` VALUES
  (1);
  

• *-schema-view.sql, *-schema-trigger.sql, *-schema-post.sql: Other exported files

Export data to Amazon S3 cloud storage

Starting from v4.0.8, Dumpling supports exporting data to cloud storages. If you need to back up data to Amazon S3, you need to specify the Amazon S3 storage path in the -o parameter.

You need to create an Amazon S3 bucket in the specified region (see the Amazon documentation - How do I create an S3 Bucket). If you also need to create a folder in the bucket, see the Amazon documentation - Creating a folder.

Pass SecretKey and AccessKey of the account with the permission to access the Amazon S3 backend storage to the Dumpling node as environment variables.

export AWS_ACCESS_KEY_ID=${AccessKey}
export AWS_SECRET_ACCESS_KEY=${SecretKey}

Dumpling also supports reading credential files from ~/.aws/credentials. For more information about URI parameter descriptions, see URI Formats of External Storage Services.

./dumpling -u root -P 4000 -h 127.0.0.1 -r 200000 -o "s3://${Bucket}/${Folder}"

Filter the exported data

Use the --where option to filter data

By default, Dumpling exports all databases except system databases (including mysql, sys, INFORMATION_SCHEMA, PERFORMANCE_SCHEMA, METRICS_SCHEMA, and INSPECTION_SCHEMA). You can use --where <SQL where expression> to select the records to be exported.

./dumpling -u root -P 4000 -h 127.0.0.1 -o /tmp/test --where "id < 100"

The above command exports the data that matches id < 100 from each table. Note that you cannot use the --where parameter together with --sql.

Use the --filter option to filter data

Dumpling can filter specific databases or tables by specifying the table filter with the --filter option. The syntax of table filters is similar to that of .gitignore. For details, see Table Filter.

./dumpling -u root -P 4000 -h 127.0.0.1 -o /tmp/test -r 200000 --filter "employees.*" --filter "*.WorkOrder"

The above command exports all the tables in the employees database and the WorkOrder tables in all databases.

Use the -B or -T option to filter data

Dumpling can also export specific databases with the -B option or specific tables with the -T option.

Examples:

• -B employees exports the employees database.
• -T employees.WorkOrder exports the employees.WorkOrder table.

Improve export efficiency through concurrency

The exported file is stored in the ./export-<current local time> directory by default. Commonly used options are as follows:

• The -t option specifies the number of threads for the export. Increasing the number of threads improves the concurrency of Dumpling and the export speed, and also increases the database's memory consumption. Therefore, it is not recommended to set the number too large.
• The -r option enables the in-table concurrency to speed up the export. The default value is 0, which means disabled. A value greater than 0 means it is enabled, and the value is of INT type. When the source database is TiDB, a -r value greater than 0 indicates that the TiDB region information is used for splitting, and reduces the memory usage. The specific -r value does not affect the split algorithm. When the source database is MySQL and the primary key is of the INT type, specifying -r can also enable the in-table concurrency.
• The --compress <format> option specifies the compression format of the dump. It supports the following compression algorithms: gzip, snappy, and zstd. This option can speed up dumping of data if storage is the bottleneck or if storage capacity is a concern. The drawback is an increase in CPU usage. Each file is compressed individually.

With the above options specified, Dumpling can have a quicker speed of data export.

Adjust Dumpling's data consistency options

Dumpling uses the --consistency <consistency level> option to control the way in which data is exported for "consistency assurance". When using snapshot for consistency, you can use the --snapshot option to specify the timestamp to be backed up. You can also use the following levels of consistency:

• flush: Use FLUSH TABLES WITH READ LOCK to temporarily interrupt the DML and DDL operations of the replica database, to ensure the global consistency of the backup connection, and to record the binlog position (POS) information. The lock is released after all backup connections start transactions. It is recommended to perform full backups during off-peak hours or on the MySQL replica database.
• snapshot: Get a consistent snapshot of the specified timestamp and export it.
• lock: Add read locks on all tables to be exported.
• none: No guarantee for consistency.
• auto: Use flush for MySQL and snapshot for TiDB.

After everything is done, you can see the exported file in /tmp/test:

ls -lh /tmp/test | awk '{print $5 "\t" $9}'

140B  metadata
66B   test-schema-create.sql
300B  test.sbtest1-schema.sql
190K  test.sbtest1.0.sql
300B  test.sbtest2-schema.sql
190K  test.sbtest2.0.sql
300B  test.sbtest3-schema.sql
190K  test.sbtest3.0.sql

Export historical data snapshots of TiDB

Dumpling can export the data of a certain tidb_snapshot with the --snapshot option specified.

The --snapshot option can be set to a TSO (the Position field output by the SHOW MASTER STATUS command) or a valid time of the datetime data type (in the form of YYYY-MM-DD hh:mm:ss), for example:

./dumpling --snapshot 417773951312461825
./dumpling --snapshot "2020-07-02 17:12:45"

The TiDB historical data snapshots when the TSO is 417773951312461825 and the time is 2020-07-02 17:12:45 are exported.

Control the memory usage of exporting large tables

When Dumpling is exporting a large single table from TiDB, Out of Memory (OOM) might occur because the exported data size is too large, which causes connection abort and export failure. You can use the following parameters to reduce the memory usage of TiDB:

• Setting -r to split the data to be exported into chunks. This reduces the memory overhead of TiDB's data scan and enables concurrent table data dump to improve export efficiency. When the upstream database is TiDB v3.0 or later versions, a -r value greater than 0 indicates that the TiDB region information is used for splitting and the specific -r value does not affect the split algorithm.
• Reduce the value of --tidb-mem-quota-query to 8589934592 (8 GB) or lower. --tidb-mem-quota-query controls the memory usage of a single query statement in TiDB.
• Adjust the --params "tidb_distsql_scan_concurrency=5" parameter. tidb_distsql_scan_concurrency is a session variable which controls the concurrency of the scan operations in TiDB.

Manually set the TiDB GC time

When exporting data from TiDB (less than 1 TB), if the TiDB version is v4.0.0 or later and Dumpling can access the PD address and the INFORMATION_SCHEMA.CLUSTER_INFO table of the TiDB cluster, Dumpling automatically adjusts the GC safe point to block GC without affecting the original cluster.

However, in either of the following scenarios, Dumpling cannot automatically adjust the GC time:

• The data size is very large (more than 1 TB).
• Dumpling cannot connect directly to PD, for example, if the TiDB cluster is on TiDB Cloud or on Kubernetes that is separated from Dumpling.

In such scenarios, you must manually extend the GC time in advance to avoid export failure due to GC during the export process.

To manually adjust the GC time, use the following SQL statement:

SET GLOBAL tidb_gc_life_time = '720h';

After Dumpling exits, regardless of whether the export is successful or not, you must set the GC time back to its original value (the default value is 10m).

SET GLOBAL tidb_gc_life_time = '10m';

Option list of Dumpling