Cron every 30 seconds: */30 * * * * * and the two line trick

Half a minute is the shortest interval most people actually need, and it is the one the sleep trick handles cleanly. In a six field scheduler you write */30 in the leading seconds field and get second 0 and second 30. In a system crontab you write the same job twice, once as it is and once behind a sleep 30, and the two lines interleave to give the same two runs a minute.

The interleaving is only as good as the runtime of the job. If the first copy takes longer than thirty seconds, the sleeping copy starts while it is still going, and the two are now overlapping rather than alternating. A lock does not help here, because skipping the second run turns a thirty second cadence into a sixty second one without saying so.

Both forms share a subtler issue: the second run of each minute belongs to the same cron tick as the first. If the daemon is stopped between them, you lose the second run and not the first, so a restart during that window produces one missing sample rather than a clean gap. Anything comparing consecutive samples should read timestamps rather than assume spacing.

As with any sub minute schedule, check whether the thing being polled supports being asked less often. Thirty second polling of an endpoint that only updates every five minutes is 2,880 requests a day to learn something 288 requests would have told you.