TiDB Binlog Cluster Operations

This document introduces the following TiDB Binlog cluster operations:

  • The state of a Pump and Drainer nodes
  • Starting or exiting a Pump or Drainer process
  • Managing the TiDB Binlog cluster by using the binlogctl tool or by directly performing SQL operations in TiDB

Pump or Drainer state

Pump or Drainer state description:

  • online: running normally
  • pausing: in the pausing process
  • paused: has been stopped
  • closing: in the offline process
  • offline: has been offline

Starting and exiting a Pump or Drainer process

Pump

  • Starting: When started, the Pump node notifies all Drainer nodes in the online state. If the notification is successful, the Pump node sets its state to online. Otherwise, the Pump node reports an error, sets its state to paused and exits the process.
  • Exiting: The Pump node enters the paused or offline state before the process is exited normally; if the process is exited abnormally (caused by the kill -9 command, process panic, crash), the node is still in the online state.
    • Pause: You can pause a Pump process by using the kill command (not kill -9), pressing Ctrl+C or using the pause-pump command in the binlogctl tool. After receiving the pause instruction, the Pump node sets its state to pausing, stops receiving binlog write requests and stops providing binlog data to Drainer nodes. After all threads are safely exited, the Pump node updates its state to paused and exits the process.
    • Offline: You can close a Pump process only by using the offline-pump command in the binlogctl tool. After receiving the offline instruction, the Pump node sets its state to closing and stops receiving the binlog write requests. The Pump node continues providing binlog to Drainer nodes until all binlog data is consumed by Drainer nodes. Then, the Pump node sets its state to offline and exits the process.

Drainer

  • Starting: When started, the Drainer node sets its state to online and tries to pull binlogs from all Pump nodes which are not in the offline state. If it fails to get the binlogs, it keeps trying.
  • Exiting: The Drainer node enters the paused or offline state before the process is exited normally; if the process is exited abnormally (caused by kill -9, process panic, crash), the Drainer node is still in the online state.
    • Pause: You can pause a Drainer process by using the kill command (not kill -9), pressing Ctrl+C or using the pause-drainer command in the binlogctl tool. After receiving the pause instruction, the Drainer node sets its state to pausing and stops pulling binlogs from Pump nodes. After all threads are safely exited, the Drainer node sets its state to paused and exits the process.
    • Offline: You can close a Drainer process only by using the offline-drainer command in the binlogctl tool. After receiving the offline instruction, the Drainer node sets its state to closing and stops pulling binlogs from Pump nodes. After all threads are safely exited, the Drainer node updates its state to offline and exits the process.

For how to pause, close, check, and modify the state of Drainer, see the binlogctl guide.

Use binlogctl to manage Pump/Drainer

binlogctl is an operations tool for TiDB Binlog with the following features:

  • Checking the state of Pump or Drainer
  • Pausing or closing Pump or Drainer
  • Handling the abnormal state of Pump or Drainer

For detailed usage of binlogctl, refer to binlogctl overview.

Use SQL statements to manage Pump or Drainer

To view or modify binlog related states, execute corresponding SQL statements in TiDB.

  • Check whether binlog is enabled:

    show variables like "log_bin";
    +---------------+-------+ | Variable_name | Value | +---------------+-------+ | log_bin | 0 | +---------------+-------+

    When the Value is 0, binlog is enabled. When the Value is 1, binlog is disabled.

  • Check the status of all the Pump or Drainer nodes:

    show pump status;
    +--------|----------------|--------|--------------------|---------------------| | NodeID | Address | State | Max_Commit_Ts | Update_Time | +--------|----------------|--------|--------------------|---------------------| | pump1 | 127.0.0.1:8250 | Online | 408553768673342237 | 2019-05-01 00:00:01 | +--------|----------------|--------|--------------------|---------------------| | pump2 | 127.0.0.2:8250 | Online | 408553768673342335 | 2019-05-01 00:00:02 | +--------|----------------|--------|--------------------|---------------------|
    show drainer status;
    +----------|----------------|--------|--------------------|---------------------| | NodeID | Address | State | Max_Commit_Ts | Update_Time | +----------|----------------|--------|--------------------|---------------------| | drainer1 | 127.0.0.3:8249 | Online | 408553768673342532 | 2019-05-01 00:00:03 | +----------|----------------|--------|--------------------|---------------------| | drainer2 | 127.0.0.4:8249 | Online | 408553768673345531 | 2019-05-01 00:00:04 | +----------|----------------|--------|--------------------|---------------------|
  • Modify the state of a Pump or Drainer node in abnormal situations

    change pump to node_state ='paused' for node_id 'pump1';
    Query OK, 0 rows affected (0.01 sec)
    change drainer to node_state ='paused' for node_id 'drainer1';
    Query OK, 0 rows affected (0.01 sec)

    Executing the above SQL statements works the same as the update-pump or update-drainer commands in binlogctl. Use the above SQL statements only when the Pump or Drainer node is in abnormal situations.