Important

You are viewing the archived documentation of TiDB, which no longer receives updates. It is recommended that you use the latest LTS version of the TiDB database.

TiDB Binlog Cluster Deployment

This document describes two methods of deploying TiDB Binlog:

Deploy TiDB Binlog using TiDB Ansible
Deploy TiDB Binlog using a Binary package

It is recommended to deploy TiDB Binlog using TiDB Ansible. If you just want to do a simple testing, you can deploy TiDB Binlog using a Binary package.

Hardware requirements

Pump and Drainer are deployed and operate on 64-bit universal hardware server platforms with Intel x86-64 architecture.

In environments of development, testing and production, the requirements on server hardware are as follows:

Service	The Number of Servers	CPU	Disk	Memory
Pump	3	8 core+	SSD, 200 GB+	16G
Drainer	1	8 core+	SAS, 100 GB+ (If binlogs are output as local files, the disk size depends on how long these files are retained.)	16G

Deploy TiDB Binlog using TiDB Ansible

Step 1: Download TiDB Ansible

Use the TiDB user account to log in to the central control machine and go to the /home/tidb directory. The information about the branch of TiDB Ansible and the corresponding TiDB version is as follows. If you have questions regarding which version to use, email to info@pingcap.com for more information or file an issue.
tidb-ansible branch TiDB version Note
release-3.0 3.0 stable The latest 3.0 stable version. For use in production environments (recommended).
Use the following command to download the corresponding branch of TiDB Ansible from the TiDB Ansible project on GitHub. The default folder name is tidb-ansible.
- Download the 3.0 branch:
```
git clone -b release-3.0 https://github.com/pingcap/tidb-ansible.git
```

tidb-ansible branch	TiDB version	Note
release-3.0	3.0 stable	The latest 3.0 stable version. For use in production environments (recommended).

Step 2: Deploy Pump

Modify the tidb-ansible/inventory.ini file.
1. Set enable_binlog = True to start binlog of the TiDB cluster.
```
## binlog trigger
enable_binlog = True
```
2. Add the deployment machine IPs for pump_servers.
```
## Binlog Part
[pump_servers]
172.16.10.72
172.16.10.73
172.16.10.74
```
  Pump retains the data of the latest 7 days by default. You can modify the value of the gc variable in the tidb-ansible/conf/pump.yml file (or tidb-ansible/conf/pump-cluster.yml in TiDB 3.0.2 or earlier versions) and remove the related comments:
```
global:
  # an integer value to control the expiry date of the binlog data, which indicates for how long (in days) the binlog data would be stored
  # must be bigger than 0
  # gc: 7
```
  Make sure the space of the deployment directory is sufficient for storing Binlog. For more details, see Configure the deployment directory. You can also set a separate deployment directory for Pump.
```
## Binlog Part
[pump_servers]
pump1 ansible_host=172.16.10.72 deploy_dir=/data1/pump
pump2 ansible_host=172.16.10.73 deploy_dir=/data2/pump
pump3 ansible_host=172.16.10.74 deploy_dir=/data3/pump
```
Deploy and start the TiDB cluster containing Pump.
After configuring the inventory.ini file, you can choose one method from below to deploy the TiDB cluster.
Method #1: Add Pump on the existing TiDB cluster.
1. Deploy pump_servers and node_exporters.
```
ansible-playbook deploy.yml --tags=pump -l ${pump1_ip},${pump2_ip},[${alias1_name},${alias2_name}]
```
  Note
  Do not add a space after the commas in the above command. Otherwise, an error is reported.
2. Start pump_servers.
```
ansible-playbook start.yml --tags=pump
```
3. Update and restart tidb_servers.
```
ansible-playbook rolling_update.yml --tags=tidb
```
4. Update the monitoring data.
```
ansible-playbook rolling_update_monitor.yml --tags=prometheus
```
Method #2: Deploy a TiDB cluster containing Pump from scratch.
For how to use TiDB Ansible to deploy the TiDB cluster, see Deploy TiDB Using TiDB Ansible.

Check the Pump status.

Use binlogctl to check the Pump status. Change the pd-urls parameter to the PD address of the cluster. If State is online, Pump is started successfully.

cd /home/tidb/tidb-ansible &&
resources/bin/binlogctl -pd-urls=http://172.16.10.72:2379 -cmd pumps

INFO[0000] pump: {NodeID: ip-172-16-10-72:8250, Addr: 172.16.10.72:8250, State: online, MaxCommitTS: 403051525690884099, UpdateTime: 2018-12-25 14:23:37 +0800 CST}
INFO[0000] pump: {NodeID: ip-172-16-10-73:8250, Addr: 172.16.10.73:8250, State: online, MaxCommitTS: 403051525703991299, UpdateTime: 2018-12-25 14:23:36 +0800 CST}
INFO[0000] pump: {NodeID: ip-172-16-10-74:8250, Addr: 172.16.10.74:8250, State: online, MaxCommitTS: 403051525717360643, UpdateTime: 2018-12-25 14:23:35 +0800 CST}

Step 3: Deploy Drainer

Obtain the value of initial_commit_ts.
When Drainer starts for the first time, the timestamp information initial_commit_ts is required.
- If the replication is started from the latest time point, you just need to set initial_commit_ts to -1.
- If the downstream database is MySQL or TiDB, to ensure data integrity, you need to perform full data backup and recovery. In this case, the value of initial_commit_ts must be the timestamp information of the full backup.
  If you use mydumper to perform full data backup, you can get the timestamp by referring to the Pos field in the metadata file from the export directory. An example of the metadata file is as follows:
```
Started dump at: 2019-12-30 13:25:41
SHOW MASTER STATUS:
        Log: tidb-binlog
        Pos: 413580274257362947
        GTID:
Finished dump at: 2019-12-30 13:25:41
```
Modify the tidb-ansible/inventory.ini file.
Add the deployment machine IPs for drainer_servers. Set initial_commit_ts to the value you have obtained, which is only used for the initial start of Drainer.
- Assume that the downstream is MySQL with the alias drainer_mysql:
```
[drainer_servers]
drainer_mysql ansible_host=172.16.10.71 initial_commit_ts="402899541671542785"
```
- Assume that the downstream is file with the alias drainer_file:
```
[drainer_servers]
drainer_file ansible_host=172.16.10.71 initial_commit_ts="402899541671542785"
```

Modify the configuration file.

Assume that the downstream is MySQL:

cd /home/tidb/tidb-ansible/conf &&
cp drainer-cluster.toml drainer_mysql_drainer.toml &&
vi drainer_mysql_drainer.toml

Note

Name the configuration file as alias_drainer.toml. Otherwise, the customized configuration file cannot be found during the deployment process. Note that in v3.0.0 and v3.0.1, you should name the configuration file as alias_drainer-cluster.toml.

Set db-type to mysql and configure the downstream MySQL information:

# downstream storage, equal to --dest-db-type
# Valid values are "mysql", "file", "tidb", and "kafka".
db-type = "mysql"

# the downstream MySQL protocol database
[syncer.to]
host = "172.16.10.72"
user = "root"
password = "123456"
port = 3306
# Time and size limits for flash batch write

Assume that the downstream is incremental backup data:

cd /home/tidb/tidb-ansible/conf &&
cp drainer-cluster.toml drainer_file_drainer.toml &&
vi drainer_file_drainer.toml

Set db-type to file.

# downstream storage, equal to --dest-db-type
# Valid values are "mysql", "file", "tidb", and "kafka".
db-type = "file"

# Uncomment this if you want to use "file" as "db-type".
[syncer.to]
# default data directory: "{{ deploy_dir }}/data.drainer"
dir = "data.drainer"

Deploy Drainer.
```
ansible-playbook deploy_drainer.yml
```
Start Drainer.
```
ansible-playbook start_drainer.yml
```

Deploy TiDB Binlog using a Binary package

Download the official Binary package

Run the following commands to download the packages:

version="v3.0" for latest stable release of TiDB 3.0
wget https://download.pingcap.org/tidb-v3.0-linux-amd64.{tar.gz,sha256}

Check the file integrity. If the result is OK, the file is correct.

sha256sum -c tidb-v3.0-linux-amd64.sha256

For TiDB v2.1.0 GA or later versions, Pump and Drainer are already included in the TiDB download package. For other TiDB versions, you need to download Pump and Drainer separately using the following command:

version="latest" for nightly builds &&
wget https://download.pingcap.org/tidb-latest-linux-amd64.{tar.gz,sha256}

Check the file integrity. If the result is OK, the file is correct.

sha256sum -c tidb-latest-linux-amd64.sha256

The usage example

Assuming that you have three PD nodes, one TiDB node, two Pump nodes, and one Drainer node, the information of each node is as follows:

Node	IP
TiDB	192.168.0.10
PD1	192.168.0.16
PD2	192.168.0.15
PD3	192.168.0.14
Pump	192.168.0.11
Pump	192.168.0.12
Drainer	192.168.0.13

The following part shows how to use Pump and Drainer based on the nodes above.

Deploy Pump using the binary.

To view the command line parameters of Pump, execute ./bin/pump -help:

Usage of Pump:
-L string
    the output information level of logs: debug, info, warn, error, fatal ("info" by default)
-V
    the print version information
-addr string
    the RPC address through which Pump provides the service (-addr="192.168.0.11:8250")
-advertise-addr string
    the RPC address through which Pump provides the external service (-advertise-addr="192.168.0.11:8250")
-config string
    the path of the configuration file. If you specify the configuration file, Pump reads the configuration in the configuration file first. If the corresponding configuration also exits in the command line parameters, Pump uses the configuration of the command line parameters to cover that of the configuration file.
-data-dir string
    the path where the Pump data is stored
-gc int
    the number of days to retain the data in Pump ("7" by default)
-heartbeat-interval int
    the interval of the heartbeats Pump sends to PD (in seconds)
-log-file string
    the file path of logs
-log-rotate string
    the switch frequency of logs (hour/day)
-metrics-addr string
    the Prometheus Pushgateway address. If not set, it is forbidden to report the monitoring metrics.
-metrics-interval int
    the report frequency of the monitoring metrics ("15" by default, in seconds)
-node-id string
    the unique ID of a Pump node. If you do not specify this ID, the system automatically generates an ID based on the host name and listening port.
-pd-urls string
    the address of the PD cluster nodes (-pd-urls="http://192.168.0.16:2379,http://192.168.0.15:2379,http://192.168.0.14:2379")
-fake-binlog-interval int
    the frequency at which a Pump node generates fake binlog ("3" by default, in seconds)

Taking deploying Pump on "192.168.0.11" as an example, the Pump configuration file is as follows:

# Pump Configuration

# the bound address of Pump
addr = "192.168.0.11:8250"

# the address through which Pump provides the service
advertise-addr = "192.168.0.11:8250"

# the number of days to retain the data in Pump ("7" by default)
gc = 7

# the directory where the Pump data is stored
data-dir = "data.pump"

# the interval of the heartbeats Pump sends to PD (in seconds)
heartbeat-interval = 2

# the address of the PD cluster nodes (each separated by a comma with no whitespace)
pd-urls = "http://192.168.0.16:2379,http://192.168.0.15:2379,http://192.168.0.14:2379"

# [security]
# This section is generally commented out if no special security settings are required.
# The file path containing a list of trusted SSL CAs connected to the cluster.
# ssl-ca = "/path/to/ca.pem"
# The path to the X509 certificate in PEM format that is connected to the cluster.
# ssl-cert = "/path/to/drainer.pem"
# The path to the X509 key in PEM format that is connected to the cluster.
# ssl-key = "/path/to/drainer-key.pem"

# [storage]
# Set to true (by default) to guarantee reliability by ensuring binlog data is flushed to the disk.
# sync-log = true

# When the available disk capacity is less than the set value, Pump stops writing data.
# 42 MB -> 42000000, 42 mib -> 44040192
# default: 10 gib
# stop-write-at-available-space = "10 gib"
# The LSM DB settings embedded in Pump. Unless you know this part well, it is usually commented out.
# [storage.kv]
# block-cache-capacity = 8388608
# block-restart-interval = 16
# block-size = 4096
# compaction-L0-trigger = 8
# compaction-table-size = 67108864
# compaction-total-size = 536870912
# compaction-total-size-multiplier = 8.0
# write-buffer = 67108864
# write-L0-pause-trigger = 24
# write-L0-slowdown-trigger = 17

The example of starting Pump:
```
./bin/pump -config pump.toml
```
If the command line parameters is the same with the configuration file parameters, the values of command line parameters are used.

Deploy Drainer using binary.

To view the command line parameters of Drainer, execute ./bin/drainer -help:

Usage of Drainer:
-L string
    the output information level of logs: debug, info, warn, error, fatal ("info" by default)
-V
    the print version information
-addr string
    the address through which Drainer provides the service (-addr="192.168.0.13:8249")
-c int
    the number of the concurrency of the downstream for replication. The bigger the value, the better throughput performance of the concurrency ("1" by default).
-cache-binlog-count int
    the limit on the number of binlog items in the cache ("8" by default)
    If a large single binlog item in the upstream causes OOM in Drainer, try to lower the value of this parameter to reduce memory usage.
-config string
    the directory of the configuration file. Drainer reads the configuration file first.
    If the corresponding configuration exists in the command line parameters, Drainer uses the configuration of the command line parameters to cover that of the configuration file.
-data-dir string
    the directory where the Drainer data is stored ("data.drainer" by default)
-dest-db-type string
    the downstream service type of Drainer
    The value can be "mysql", "tidb", "kafka", and "file". ("mysql" by default)
-detect-interval int
    the interval of checking the online Pump in PD ("10" by default, in seconds)
-disable-detect
    whether to disable the conflict monitoring
-disable-dispatch
    whether to disable the SQL feature of splitting a single binlog file. If it is set to "true", each binlog file is restored to a single transaction for replication based on the order of binlogs.
    It is set to "False", when the downstream is MySQL.
-ignore-schemas string
    the db filter list ("INFORMATION_SCHEMA,PERFORMANCE_SCHEMA,mysql,test" by default)
    It does not support the Rename DDL operation on tables of `ignore schemas`.
-initial-commit-ts
    If Drainer does not have the related breakpoint information, you can configure the related breakpoint information using this parameter. ("-1" by default)
    If the value of this parameter is `-1`, Drainer automatically obtains the latest timestamp from PD.
-log-file string
    the path of the log file
-log-rotate string
    the switch frequency of log files, hour/day
-metrics-addr string
    the Prometheus Pushgateway address
    It it is not set, the monitoring metrics are not reported.
-metrics-interval int
    the report frequency of the monitoring metrics ("15" by default, in seconds)
-node-id string
    the unique ID of a Drainer node. If you do not specify this ID, the system automatically generates an ID based on the host name and listening port.
-pd-urls string
    the address of the PD cluster nodes (-pd-urls="http://192.168.0.16:2379,http://192.168.0.15:2379,http://192.168.0.14:2379")
-safe-mode
    Whether to enable safe mode so that data can be written into the downstream MySQL/TiDB repeatedly.
    This mode replaces the `INSERT` statement with the `REPLACE` statement and splits the `UPDATE` statement into `DELETE` plus `REPLACE`.
-txn-batch int
    the number of SQL statements of a transaction which are output to the downstream database ("1" by default)

Taking deploying Drainer on "192.168.0.13" as an example, the Drainer configuration file is as follows:

# Drainer Configuration.

# the address through which Drainer provides the service ("192.168.0.13:8249")
addr = "192.168.0.13:8249"

# the address through which Drainer provides the external service
advertise-addr = "192.168.0.13:8249"

# the interval of checking the online Pump in PD ("10" by default, in seconds)
detect-interval = 10

# the directory where the Drainer data is stored "data.drainer" by default)
data-dir = "data.drainer"

# the address of the PD cluster nodes (each separated by a comma with no whitespace)
pd-urls = "http://192.168.0.16:2379,http://192.168.0.15:2379,http://192.168.0.14:2379"

# the directory of the log file
log-file = "drainer.log"

# Drainer compresses the data when it gets the binlog from Pump. The value can be "gzip". If it is not configured, it will not be compressed
# compressor = "gzip"

# [security]
# This section is generally commented out if no special security settings are required.
# The file path containing a list of trusted SSL CAs connected to the cluster.
# ssl-ca = "/path/to/ca.pem"
# The path to the X509 certificate in PEM format that is connected to the cluster.
# ssl-cert = "/path/to/pump.pem"
# The path to the X509 key in PEM format that is connected to the cluster.
# ssl-key = "/path/to/pump-key.pem"

# Syncer Configuration
[syncer]
# If the item is set, the sql-mode will be used to parse the DDL statement.
# If the downstream database is MySQL or TiDB, then the downstream sql-mode
# is also set to this value.
# sql-mode = "STRICT_TRANS_TABLES,NO_ENGINE_SUBSTITUTION"

# the number of SQL statements of a transaction that are output to the downstream database ("20" by default)
txn-batch = 20

# the number of the concurrency of the downstream for replication. The bigger the value,
# the better throughput performance of the concurrency ("16" by default)
worker-count = 16

# whether to disable the SQL feature of splitting a single binlog file. If it is set to "true",
# each binlog file is restored to a single transaction for replication based on the order of binlogs.
# If the downstream service is MySQL, set it to "False".
disable-dispatch = false

# In safe mode, data can be written into the downstream MySQL/TiDB repeatedly.
# This mode replaces the `INSERT` statement with the `REPLACE` statement and replaces the `UPDATE` statement with `DELETE` plus `REPLACE` statements.
safe-mode = false

# the downstream service type of Drainer ("mysql" by default)
# Valid value: "mysql", "file", "tidb", and "kafka".
db-type = "mysql"

# If `commit ts` of the transaction is in the list, the transaction is filtered and not replicated to the downstream.
ignore-txn-commit-ts = []

# the db filter list ("INFORMATION_SCHEMA,PERFORMANCE_SCHEMA,mysql,test" by default)
# Does not support the Rename DDL operation on tables of `ignore schemas`.
ignore-schemas = "INFORMATION_SCHEMA,PERFORMANCE_SCHEMA,mysql"

# `replicate-do-db` has priority over `replicate-do-table`. When they have the same `db` name,
# regular expressions are supported for configuration.
# The regular expression should start with "~".

# replicate-do-db = ["~^b.*","s1"]

# [syncer.relay]
# It saves the directory of the relay log. The relay log is not enabled if the value is empty.
# The configuration only comes to effect if the downstream is TiDB or MySQL.
# log-dir = ""
# the maximum size of each file
# max-file-size = 10485760

# [[syncer.replicate-do-table]]
# db-name ="test"
# tbl-name = "log"

# [[syncer.replicate-do-table]]
# db-name ="test"
# tbl-name = "~^a.*"

# Ignore the replication of some tables
# [[syncer.ignore-table]]
# db-name = "test"
# tbl-name = "log"

# the server parameters of the downstream database when `db-type` is set to "mysql"
[syncer.to]
host = "192.168.0.13"
user = "root"
password = ""
# `encrypted_password` is encrypted using `./binlogctl -cmd encrypt -text string`.
# When `encrypted_password` is not empty, the `password` above will be ignored.
encrypted_password = ""
port = 3306

[syncer.to.checkpoint]
# When the checkpoint type is "mysql" or "tidb", this option can be enabled to change the database that saves the checkpoint
# schema = "tidb_binlog"
# Currently only the "mysql" and "tidb" checkpoint types are supported
# You can remove the comment tag to control where to save the checkpoint
# The default method of saving the checkpoint for the downstream db-type:
# mysql/tidb -> in the downstream MySQL or TiDB database
# file/kafka -> file in `data-dir`
# type = "mysql"
# host = "127.0.0.1"
# user = "root"
# password = ""
# `encrypted_password` is encrypted using `./binlogctl -cmd encrypt -text string`.
# When `encrypted_password` is not empty, the `password` above will be ignored.
# encrypted_password = ""
# port = 3306

# the directory where the binlog file is stored when `db-type` is set to `file`
# [syncer.to]
# dir = "data.drainer"

# the Kafka configuration when `db-type` is set to "kafka"
# [syncer.to]
# only one of kafka-addrs and zookeeper-addrs is needed. If both are present, the program gives priority to the kafka address in zookeeper.
# zookeeper-addrs = "127.0.0.1:2181"
# kafka-addrs = "127.0.0.1:9092"
# kafka-version = "0.8.2.0"
# kafka-max-messages = 1024

# the topic name of the Kafka cluster that saves the binlog data. The default value is <cluster-id>_obinlog
# To run multiple Drainers to replicate data to the same Kafka cluster, you need to set different `topic-name`s for each Drainer.
# topic-name = ""

Starting Drainer:
Note
If the downstream is MySQL/TiDB, to guarantee the data integrity, you need to obtain the initial-commit-ts value and make a full backup of the data and restore the data before the initial start of Drainer. For details, see Deploy Drainer.
When Drainer is started for the first time, use the initial-commit-ts parameter.
```
./bin/drainer -config drainer.toml -initial-commit-ts {initial-commit-ts}
```
If the command line parameter and the configuration file parameter are the same, the parameter value in the command line is used.

Starting TiDB server:
- After starting Pump and Drainer, start TiDB server with binlog enabled by adding this section to your config file for TiDB server:
```
[binlog]
enable=true
```
- TiDB server will obtain the addresses of registered Pumps from PD and will stream data to all of them. If there are no registered Pump instances, TiDB server will refuse to start or will block starting until a Pump instance comes online.

Note

When TiDB is running, you need to guarantee that at least one Pump is running normally.
To enable the TiDB Binlog service in TiDB server, use the -enable-binlog startup parameter in TiDB, or add enable=true to the [binlog] section of the TiDB server configuration file.
Make sure that the TiDB Binlog service is enabled in all TiDB instances in a same cluster, otherwise upstream and downstream data inconsistency might occur during data replication. If you want to temporarily run a TiDB instance where the TiDB Binlog service is not enabled, set run_ddl=false in the TiDB configuration file.
Drainer does not support the rename DDL operation on the table of ignore schemas (the schemas in the filter list).
If you want to start Drainer in an existing TiDB cluster, generally you need to make a full backup of the cluster data, obtain snapshot timestamp, import the data to the target database, and then start Drainer to replicate the incremental data from corresponding snapshot timestamp.
When the downstream database is TiDB or MySQL, ensure that the sql_mode in the upstream and downstream databases are consistent. In other words, the sql_mode should be the same when each SQL statement is executed in the upstream and replicated to the downstream. You can execute the select @@sql_mode; statement in the upstream and downstream respectively to compare sql_mode.
When a DDL statement is supported in the upstream but incompatible with the downstream, Drainer fails to replicate data. An example is to replicate the CREATE TABLE t1(a INT) ROW_FORMAT=FIXED; statement when the downstream database MySQL uses the InnoDB engine. In this case, you can configure skipping transactions in Drainer, and manually execute compatible statements in the downstream database.