jstests/sharding are expensive, and that's not immediately clear to most developers.

These tests are similar to noPassthrough tests in that each test case brings up and tears down its own sharded cluster instance. However, unlike noPassthrough tests, which are run just once per variant, jstests/sharding tests are run about a dozen times per variant, causing a dozen sharded cluster setups and teardowns. This is very expensive. Most of the time, the developers mean to run it just once, and are unaware of the implications of placing the test under jstests/sharding.

One way to raise this awareness could be to add a lint rule that points to a wiki page to help decision-making, with the expectation that most tests should ideally be noPassthrough.