Migrate Data from MySQL to TiDB

Use the `mydumper` / `loader` tool to export and import all the data

You can use mydumper to export data from MySQL and loader to import the data into TiDB.

Note: Although TiDB also supports the official mysqldump tool from MySQL for data migration, it is not recommended to use it. Its performance is much lower than mydumper / loader and it takes much time to migrate large amounts of data. mydumper/loader is more powerful. For more information, see https://github.com/maxbube/mydumper.

Export data from MySQL

Use the mydumper tool to export data from MySQL by using the following command:

./bin/mydumper -h 127.0.0.1 -P 3306 -u root -t 16 -F 64 -B test -T t1,t2 --skip-tz-utc -o ./var/test

In this command,

-B test: means the data is exported from the test database.
-T t1,t2: means only the t1 and t2 tables are exported.
-t 16: means 16 threads are used to export the data.
-F 64: means a table is partitioned into chunks and one chunk is 64MB.
--skip-tz-utc: the purpose of adding this parameter is to ignore the inconsistency of time zone setting between MySQL and the data exporting machine and to disable automatic conversion.

Note: On the Cloud platforms which require the super privilege, such as on the Aliyun platform, add the --no-locks parameter to the command. If not, you might get the error message that you don’t have the privilege.

Import data to TiDB

Use loader to import the data from MySQL to TiDB. See Loader instructions for more information.

./bin/loader -h 127.0.0.1 -u root -P 4000 -t 32 -d ./var/test

After the data is imported, you can view the data in TiDB using the MySQL client:

mysql -h127.0.0.1 -P4000 -uroot

mysql> show tables;
+----------------+
| Tables_in_test |
+----------------+
| t1             |
| t2             |
+----------------+

mysql> select * from t1;
+----+------+
| id | age  |
+----+------+
|  1 |    1 |
|  2 |    2 |
|  3 |    3 |
+----+------+

mysql> select * from t2;
+----+------+
| id | name |
+----+------+
|  1 | a    |
|  2 | b    |
|  3 | c    |
+----+------+

Best practice

To migrate data quickly, especially for huge amount of data, you can refer to the following recommendations.

Keep the exported data file as small as possible and it is recommended keep it within 64M. You can use the -F parameter to set the value.
You can adjust the -t parameter of loader based on the number and the load of TiKV instances. For example, if there are three TiKV instances, -t can be set to 3 * (1 ~ n). If the load of TiKV is too high and the log backoffer.maxSleep 15000ms is exceeded is displayed many times, decrease the value of -t; otherwise, increase it.

A sample and the configuration

The total size of the exported files is 214G. A single table has 8 columns and 2 billion rows.
The cluster topology:
- 12 TiKV instances: 4 nodes, 3 TiKV instances per node
- 4 TiDB instances
- 3 PD instances
The configuration of each node:
- CPU: Intel Xeon E5-2670 v3 @ 2.30GHz
- 48 vCPU [2 x 12 physical cores]
- Memory: 128G
- Disk: sda [raid 10, 300G] sdb[RAID 5, 2T]
- Operating System: CentOS 7.3
The -F parameter of mydumper is set to 16 and the -t parameter of loader is set to 64.

Results: It takes 11 hours to import all the data, which is 19.4G/hour.

Use the `syncer` tool to import data incrementally (optional)

The previous section introduces how to import all the history data from MySQL to TiDB using mydumper/loader. But this is not applicable if the data in MySQL is updated after the migration and it is expected to import the updated data quickly.

Therefore, TiDB provides the syncer tool for an incremental data import from MySQL to TiDB.

See Download the TiDB enterprise toolset to download the syncer tool.

Download the TiDB enterprise toolset (Linux)

# Download the enterprise tool package.
wget http://download.pingcap.org/tidb-enterprise-tools-latest-linux-amd64.tar.gz
wget http://download.pingcap.org/tidb-enterprise-tools-latest-linux-amd64.sha256

# Check the file integrity. If the result is OK, the file is correct.
sha256sum -c tidb-enterprise-tools-latest-linux-amd64.sha256

# Extract the package.
tar -xzf tidb-enterprise-tools-latest-linux-amd64.tar.gz
cd tidb-enterprise-tools-latest-linux-amd64

Assuming the data from t1 and t2 is already imported to TiDB using mydumper/loader. Now we hope that any updates to these two tables are synchronised to TiDB in real time.

Obtain the position to synchronise

The data exported from MySQL contains a metadata file which includes the position information. Take the following metadata information as an example:

Started dump at: 2017-04-28 10:48:10
SHOW MASTER STATUS:
    Log: mysql-bin.000003
    Pos: 930143241
    GTID:

Finished dump at: 2017-04-28 10:48:11

The position information (Pos: 930143241) needs to be stored in the syncer.meta file for syncer to synchronize:

# cat syncer.meta
binlog-name = "mysql-bin.000003"
binlog-pos = 930143241

Note: The syncer.meta file only needs to be configured once when it is first used. The position will be automatically updated when binlog is synchronised.

Start `syncer`

The config.toml file for syncer:

log-level = "info"

server-id = 101

# The file path for meta:
meta = "./syncer.meta"
worker-count = 16
batch = 10

# The testing address for pprof. It can also be used by Prometheus to pull the syncer metrics.
status-addr = ":10081"

skip-sqls = ["ALTER USER", "CREATE USER"]

# Support whitelist filter. You can specify the database and table to be synchronised. For example:
# Synchronise all the tables of db1 and db2:
replicate-do-db = ["db1","db2"]

# Synchronise db1.table1.
[[replicate-do-table]]
db-name ="db1"
tbl-name = "table1"

# Synchronise db3.table2.
[[replicate-do-table]]
db-name ="db3"
tbl-name = "table2"

# Support regular expressions. Start with '~'  to use regular expressions.
# To synchronise all the databases that start with `test`:
replicate-do-db = ["~^test.*"]

# The sharding synchronising rules support wildcharacter.
# 1. The asterisk character (*, also called "star") matches zero or more characters,
#    for example, "doc*" matches "doc" and "document" but not "dodo";
#    asterisk character must be in the end of the wildcard word,
#    and there is only one asterisk in one wildcard word.
# 2. The question mark ? matches exactly one character.
#[[route-rules]]
#pattern-schema = "route_*"
#pattern-table = "abc_*"
#target-schema = "route"
#target-table = "abc"

#[[route-rules]]
#pattern-schema = "route_*"
#pattern-table = "xyz_*"
#target-schema = "route"
#target-table = "xyz"

[from]
host = "127.0.0.1"
user = "root"
password = ""
port = 3306

[to]
host = "127.0.0.1"
user = "root"
password = ""
port = 4000

Start syncer:

./bin/syncer -config config.toml
2016/10/27 15:22:01 binlogsyncer.go:226: [info] begin to sync binlog from position (mysql-bin.000003, 1280)
2016/10/27 15:22:01 binlogsyncer.go:130: [info] register slave for master server 127.0.0.1:3306
2016/10/27 15:22:01 binlogsyncer.go:552: [info] rotate to (mysql-bin.000003, 1280)
2016/10/27 15:22:01 syncer.go:549: [info] rotate binlog to (mysql-bin.000003, 1280)

Insert data into MySQL

INSERT INTO t1 VALUES (4, 4), (5, 5);

Log in TiDB and view the data

mysql -h127.0.0.1 -P4000 -uroot -p
mysql> select * from t1;
+----+------+
| id | age  |
+----+------+
|  1 |    1 |
|  2 |    2 |
|  3 |    3 |
|  4 |    4 |
|  5 |    5 |
+----+------+

syncer outputs the current synchronised data statistics every 30 seconds:

2017/06/08 01:18:51 syncer.go:934: [info] [syncer]total events = 15, total tps = 130, recent tps = 4,
master-binlog = (ON.000001, 11992), master-binlog-gtid=53ea0ed1-9bf8-11e6-8bea-64006a897c73:1-74,
syncer-binlog = (ON.000001, 2504), syncer-binlog-gtid = 53ea0ed1-9bf8-11e6-8bea-64006a897c73:1-17
2017/06/08 01:19:21 syncer.go:934: [info] [syncer]total events = 15, total tps = 191, recent tps = 2,
master-binlog = (ON.000001, 11992), master-binlog-gtid=53ea0ed1-9bf8-11e6-8bea-64006a897c73:1-74,
syncer-binlog = (ON.000001, 2504), syncer-binlog-gtid = 53ea0ed1-9bf8-11e6-8bea-64006a897c73:1-35

You can see that by using syncer, the updates in MySQL are automatically synchronised in TiDB.