Typically you would want to record the events leading up to the upset, because those are usually more informative than the (successful or failed) upset recovery itself.
Many security systems (that are not configured to record continuously) keep a ~3-10 second buffer of footage in memory, so that when an event/trigger occurs they can store footage from before the event, as well as however long they're configured to record after the event.
Making that buffer larger means more memory and a larger chance of the buffer not being successfully transferred to permanent storage in case of e.g. a power cut or impact.