Designing database reliant tests for parallel execution

Question

Looking for some best practice here:

I'm looking to future proof the selenium tests that I'm building. They'll involve shared elements (products/locations/users), and it seems like a bad pattern to be building unique, test specific data for each test in isolation.

So how do I avoid my setup/teardown potentially wrecking other tests?

Answer 1

Bootstrap a clean environment for each test to test in isolation. This might look like a lot of overhead, but is very needed if you want to parallel run your tests and not get flickering and possible untrusted tests.

Personally I create a session GUID and then generate test data and store in it in a database with the GUID. Boot an environment and let it point to the dataset. Automating this should already be part of the automated build-pipeline, if not this is a good reason to start working on it :)

Also read my answers to https://sqa.stackexchange.com/a/12225/3201 and https://sqa.stackexchange.com/a/13660/3201 which are about managing testdata and environments.

Answer 2

If database creation is slow compared to TRUNCATE, create a pool of databases

Then:

• when each test finishes, it can truncate the DB, and marks that DB as unused in a synchronized in-memory set data structure
• if a test starts and there is no free DB, it just creates the DB and adds it to the pool

Then you end up with databases named like:

• testdb_0
• testdb_1
• testdb_2

Consider using in-memory database

It might be worth checking if using e.g. SQLite in-memory can considerably speed up the DB creation process. If it does, you can just use one DB per test as mentioned at https://sqa.stackexchange.com/a/16855/49066