Throttler: How to compute replication lag?

[newborn22]

replicationLagQuery = select unix_timestamp(now(6))-max(ts/1000000000) as replication_lag from mysql.heartbeat

How can this SQL accurately calculate replication lag?

[newborn22]

How does ReplTracker works? ReplTracker is a component that manages heartbeat table;

[newborn22]

The function enableWrites will start a ticker which writes timestamps to hearttable table at regular intervals.

The replication lag refers to the time taken by the replic node to synchronise the binlog from the primary node.

Heartbeats inserted by the throttler to the primary node are synchronised to the replic nodes. When the throttler receives a request from the application, it requests data from the heartbeat table at a fixed interval. Then the heartbeat is subtracted from the select time to get the replication lag, the specific sql is select unix_timestamp(now(6))-max(ts/1000000000) as replication_lag from mysql.heartbeat

While the method above can not accurately represent the replication lag between the primary and replic nodes, since the throttler just simply compares it to the predetermined threshold, it is still able to report when there is excessive traffic.

Note that when an App makes a request to the throttler, the throttler queries the heartbeat faster and reduces the frequency of querying the heartbeat when there is no application request, i.e., it enters sleep mode.

However the frequency of inserting heartbeats into the table by the throttler does not decrease. The insert sql is like: "INSERT INTO %s.heartbeat (ts, tabletUid, keyspaceShard) VALUES (%a, %a, %a) ON DUPLICATE KEY UPDATE ts=VALUES(ts), tabletUid=VALUES(tabletUid)"