Cleanup

One test’s rows must not leak into the next. Call CleanAsync() at the start of a test:

await harness.CleanAsync();

In a DbTestBase class, harness is Harness.

What it does

CleanAsync() empties every user table, whatever the foreign keys between them, and puts every identity column back to its seed. The next row gets the same id on every engine.

The mechanism differs by engine:

SQL Server

A table that no foreign key references is truncated. That empties it and resets its identity to exactly the declared seed, whatever the column type.

SQL Server refuses to truncate a table a foreign key references, even a disabled one. Those tables are emptied with DELETE while every constraint is disabled, and the constraints are re-enabled WITH CHECK so the optimizer still trusts them. Their identity is then reseeded.

Caution

If a referenced table has an identity column whose seed minus increment is outside the column type’s range (for example a tinyint IDENTITY(0,1) that has held rows), no reseed can make the next id the seed. The clean then fails with a message naming the table.

PostgreSQL

One statement: TRUNCATE TABLE ... RESTART IDENTITY CASCADE over every base table outside pg_catalog and information_schema. It empties every table and resets every owned sequence to its start value.

MySQL

Foreign key checks are switched off, every table is truncated, and the checks are switched back on, all in one command.

MySQL keeps no seed apart from the counter: TRUNCATE resets AUTO_INCREMENT to 1, so a table created with AUTO_INCREMENT = n restarts at 1.

What it does not do

• It does not touch the shape of objects. That is the shape guard’s job.
• It also empties reference data that a seed script inserted. Seed it again after cleaning. With a SQL Server database project, see the dacpac guide for how the sample does it.