Understanding MySQL Slave Net Timeout, Master Heartbeat Period, and Their Impact on Relay Log Accumulation

This article presents a real‑world MySQL replication case in which a large number of tiny relay‑log files (about 2,600) were generated due to mismatched timeout settings between the slave and the master.

Case description : The slave’s slave_net_timeout was set to 10 seconds while the master’s MASTER_HEARTBEAT_PERIOD defaulted to 30 seconds. When the master sent no events, the slave’s I/O thread timed out, disconnected, and repeatedly re‑connected, creating new relay logs that could not be cleaned up.

Parameter analysis : slave_net_timeout defines the idle timeout for the slave I/O thread; since MySQL 5.7.7 it defaults to 60 s (previously 3600 s). If MASTER_HEARTBEAT_PERIOD is not explicitly set, it is derived as slave_net_timeout / 2 . The heartbeat ensures a periodic event is sent from master to slave to keep the connection alive.

Root cause : Three conditions trigger the issue: (1) MASTER_HEARTBEAT_PERIOD > slave_net_timeout ; (2) the master is idle, producing no events; (3) existing replication lag. Under these conditions the master kills the “zombie” dump thread, the slave reconnects, and a new relay log is created each time.

Solution : Set slave_net_timeout to at least twice the value of MASTER_HEARTBEAT_PERIOD (or adjust the heartbeat period accordingly) and restart the replication threads. Example commands:

set global slave_net_timeout=10;

stop slave sql_thread;

stop slave;

start slave io_thread;

After adjusting the parameters, the relay‑log accumulation stops.

Implementation details : The article walks through the source‑code paths where slave_net_timeout is applied (via mysql_options and mysql_real_connect ), how MASTER_HEARTBEAT_PERIOD is set during CHANGE MASTER , and how the dump thread uses the heartbeat to send periodic events. Key snippets include:

case MYSQL_OPT_CONNECT_TIMEOUT:
    mysql->options.connect_timeout = *(uint*)arg;
    break;

char query_format[] = "SET @master_heartbeat_period= %s";
char query[sizeof(query_format) - 2 + sizeof(llbuf)];

if (mi->heartbeat_period != 0.0) {
    set_timespec_nsec(&ts, mi->heartbeat_period);
    ret = mysql_bin_log.wait_for_update_bin_log(m_thd, &ts);
    if (ret == ETIMEDOUT) send_heartbeat_event(log_pos);
}

The article also provides a simulation script that reproduces the problem by stopping the slave SQL thread, setting a low slave_net_timeout , and observing the generation of relay logs every ~10 seconds.

Finally, a flowchart from the author’s book “Deep Understanding of MySQL Master‑Slave Principles” is shown to illustrate the dump‑thread lifecycle.