fix: resolve intermittent Tokenserver test failure #1171
Conversation
|
This is still not working...converting to a draft |
|
Whew. The problem was that users were sometimes being created with the same timestamp, which caused a problem here: since the two most recently created users share the same timestamp, neither of those users were being replaced, resulting in two "active" user records for a given email address. |
|
oh, huh. |
| @@ -162,20 +162,32 @@ def test_cleanup_of_old_records(self): | |||
| # Do a sleep halfway through so we can test use of grace period. | |||
| email1 = '[email protected]' | |||
| user1 = self.database.allocate_user(email1) | |||
| # We have to sleep between every user create/update operation: if two | |||
There was a problem hiding this comment.
I do wonder if this might eventually bite us (e.g. if the client is also in rust and is on a particularly close node, they may respond faster than one ms.) We might want to add an issue to check for active user records before writing a second and returning a 409 CONFLICT if one is detected.
There was a problem hiding this comment.
I'm not sure that the risk of this happening on the new Tokenserver is any greater than the risk of it happening on the old Tokenserver 🤔 Requests on the new Tokenserver will still take well over 1 ms to complete, so I think the only way this could happen would be if a client decided to bombard us with requests, and I think the old Tokenserver is just as vulnerable to that
Description
Add
time.sleeps to fix an intermittent timestamp issue in a unit testTesting
The failing test should pass on CI
Issue(s)
Closes #1170