Currently, the various plan stability tests are stand-alone, that is, they load or generate their dataset at the start of the test and put it in a database named after the test.

With multiple tests sharing the same dataset, this results in a duplication of effort. In particular, loading a scale 1 TPC-H multiple times would take too much time.

Instead, as hagar.mohamed@mongodb.com suggested, we can load the datasets once and make them available for multiple tests.

In practice, this requires coding a fixture or an eval section in the passthrough config file, as well as modifying the test slightly to use the common database.

We may also have a linter that checks that a test has not left behind any database objects – if this linter fires, it needs to be adjusted to permit the datasets to remain loaded between tests.