FIELD NOTES /
How do you design migration tests on a live environment?
In an environment where timeseries data is the one mostly being consumed and being build upon, how do you design the tests necessary for the migration of such a system? In general, that would not be such a special problem if the target test is not to have the “todays” data as the main reference for a successful migration. So the problem is, that the target is moving every day. If I had understood the problem and the approach to the migration initially, I would not have accepted such a premise. The source system does not have a way to fake time, but in the target system we have a way of faking the time. But that is not enough. The timeseries are being backfilled on the source system. That means that data that has been processed as data in the past can change - often. There are many problems, when trying to test the migration but here are a few we are going to focus on:
High level overview
