Impact of Adjusting System Time on MySQL Runtime Behavior

When the system time is altered during MySQL execution, various internal timers and wait‑time calculations can be affected.

Experiment 1: After creating a test database, a session runs SLEEP . While the session is sleeping, the server clock is moved forward by 10 seconds. The SLEEP call returns immediately after only about 0.82 seconds, showing that the sleep duration is based on the system clock.

Experiment 2: One session locks a table. In a second session, a timestamp is printed, lock_wait_timeout is adjusted, and the locked table is accessed. The system clock is then moved backward by 10 seconds. When the request finally times out, the measured lock duration is 17 seconds, indicating that MySQL’s MDL wait‑time calculation is directly influenced by the system clock.

Tip: The method used to obtain the system time in the article reads from /proc/timer_list , which reflects a monotonic clock that is not affected by manual time changes. Using a monotonic clock (e.g., via clock_gettime(CLOCK_MONOTONIC) ) provides an objective measurement of actual wait time.

Beyond MDL, adjusting the system time can also impact semi‑synchronous replication wait times, delayed replication lag calculations, and other time‑sensitive operations. The recommended practice is to avoid changing the system clock while MySQL is running; if a change is necessary, restart MySQL promptly.

Thought question: Design an experiment to measure InnoDB lock wait times and compare the results with those observed for MDL, exploring why the differences occur.