Support traffic replay

Feature Request

Describe your feature request related problem

There are some cases when users want to capture the traffic on the production cluster and replay the traffic on a testing cluster:

• A new TiDB version may have compatibility breakers, such as the statements failing, running slower, or resulting in different query results.
• When the cluster runs unexpectedly, users want to capture the traffic so that they can investigate it later by replaying the traffic.
• Test the maximum throughput of a scaled-up or scaled-down cluster using the real workload instead of Sysbench or TPCC.

Some traffic replay tools are widely used, including tcpcopy, mysql-replay, and query-playback. Tcpcopy and mysql-replay capture data like tcpdump, while query-playback is based on slow logs. Although some of them are built for MySQL, deploying them on the proxy instance also works.

However, they have some limitations:

• Users need to learn the deployment and usage of these tools.
• Tcpcopy only captures new connections, which is unfriendly for persistent connections. Mysql-replay can capture existing connections but it loses session states such as prepared statements and session variables, which may make replay fail.
• The tools replay the traffic with one username and one current schema, which requires modification to the testing cluster.
• Tcpcopy and mysql-replay don't support TLS because they can't decode the encrypted data.
• Users need to verify the results and performance manually.

Describe the feature you'd like

Capturing traffic on one TiDB cluster and replaying the traffic on a new TiDB cluster to verify the SQL compatibility and performance of the new TiDB cluster.

Tasks

Design

• #643

Traffic Capture

P1:

• #645
• #647
• #674
• #676
• #672
• #681

P2:

• #726
• #717
• #719
• #729
• #722

P3:

• Support more command types

Traffic Replay

P1:

• #652
• #654
• #657
• #659
• #678
• #687
• #700

P2:

• #711

P3:

• Streaming reading and sending large commands
• Track and limit the memory of pending commands
• Support more auth plugins
• Support replaying as fast as possible
• #866
• Support reading from MySQL slow logs
• Support reading from tcpdump files
• Support reading from TiDB statements summary tables
• Support reading from MySQL statements summary tables

Comparison and Report

P1:

• #661
• #683

User Interface

P1:

• #665
• #667
• #669
• #685

P2:

• #723
• #735
• #737
• #740
• #744
• #742
• #748

P3:

• Show the estimated disk usage of the traffic files
• Run as a replayer client