There are two things that affect slot adjustment:

• The interval of CPULoad ticks: This can be either 1ms or 250ms. We used to log every change in this interval, but no longer do that since it was spamming the logs. An alternative would be to increment two cumulative duration metrics representing the cumulative duration spent under each regime.
• Interaction between increments and decrements: we have observed slow changes in the total slots in the past e.g. https://cockroachlabs.slack.com/archives/C01SRKWGHG8/p1667511459730139?thread_ts=1667418886.450229&cid=C01SRKWGHG8, and been unsure whether it was because we were both increasing and decreasing, or because the increase was rare. Having two counters counting the number of increments and decrements would tell us what happened.

Jira issue: CRDB-21914