A few scenarios are explored in this case study. Generally you would need to recreate a mirror in the event of:
- Catastrophic failure of the physical hardware on which it resides
- A break in connectivity long enough for it to fall too far behind the primary
Of course, the same planning applies in the case of needing to recreate the primary from the mirror! It is a matter of your own topology whether your strategy for recovery is to go to the mirror, or to tape.
Often when making major changes, at my site we will pause database mirrors (or the Oracle equivalent, defer log shipping) until we are happy, then resume it. This gives us a measure of protection against things going hatstand for any reason, including operator error, problems wth new code, etc.