For example, a book may tell you never to use aggregate functions (such as AVG) when
transaction response time is critical. The underlying reason is that such functions must scan
substantial amounts of data and therefore may block other queries. So the rule is generally true,
but it may not hold if the average applies to a few tuples that have been selected by an index.
The point of the example is that the tuner must understand the reason for the rule, namely,
long transactions that access large portions of shared data may delay concurrent online
transactions. The well-informed tuner will then take this rule for what it is: an example of a
principle (don't scan large amounts of data in a highly concurrent environment) rather than a
principle itself.