Core FR: Check that only an 'update' occurred in a test

[danawillow]

In hashicorp/terraform#15101 and hashicorp/terraform#15102, it was discovered that tests that we thought were testing update were actually just doing a destroy and recreate. It would be great to have another option in TestStep that would confirm that no resources were destroyed or created in a step, only updated. In this particular scenario, if the test had been written correctly, it would have caught the bug leading to hashicorp/terraform#15056 potentially before it was caught by a user.

[nat-henderson]

+1!

The workaround for this, right now, is to manually examine the debug output for every single test every time we write any code which touches Update. That's time-consuming, and it'd be awesome to have a ValidatePlanFunc or something like that. Accidentally deleting / recreating someone's infrastructure when they just wanted it updated can be unrecoverably destructive, and it'd be rad to have unit tests that prove we aren't going to do that.

[apparentlymart]

Another possible workaround here would be to save the value of some id attribute that changes when the object is replaced in the prior step, and then in the subsequent check step verify that the two ids match. This will, of course, work only if there is some id attribute on the object that remains static on update and changes on replace.

I agree that it'd be more ergonomic to either assert against the planned changes or to declare in the test step that it should contain no replace actions.

I'm sorry for the long silence here. I've updated the labels on this issue to put it on the radar of the team that is responsible for the provider SDK (helper/schema, etc), but they are currently focused elsewhere and so will likely return to this at a later date, in conjunction with other related improvements.

[nat-henderson]

Unfortunately that's not usually the case in GCP, but I can see how it might be elsewhere, which would definitely make this a lower-priority improvement. Thanks for the update!

[pdecat]

Hi, I've submitted hashicorp/terraform-plugin-sdk#329 which should address this issue (and hashicorp/terraform-plugin-sdk#222 which looks very similar).