bug: CheckInactiveUsers can race with user login and erroneously mark users dormant #21303
Description
Is there an existing issue for this?
- I have searched the existing issues
Current Behavior
While scale testing, a newly created user fails to log in.
Relevant Log Output
Error: create user: login as new user: POST https://primary.charlie.scaletest.dev/api/v2/users/login: unexpected status code 401: Your account is dormant. Contact an admin to reactivate your account.: Try logging in using 'coder login'.Expected Behavior
Newly created users can't be marked dormant by the job until 90 days after account creation.
Steps to Reproduce
Pretty sure this is a narrow race while the user is logging in, because we normally don't block login for dormant users.
user, err = ActivateDormantUser(api.Logger, &api.Auditor, api.Database)(ctx, user)
if err != nil {
httpapi.Write(ctx, rw, http.StatusInternalServerError, codersdk.Response{
Message: "Internal error.",
Detail: err.Error(),
})
return user, rbac.Subject{}, false
}
subject, userStatus, err := httpmw.UserRBACSubject(ctx, api.Database, user.ID, rbac.ScopeAll)
if err != nil {
logger.Error(ctx, "unable to fetch authorization user roles", slog.Error(err))
httpapi.Write(ctx, rw, http.StatusInternalServerError, codersdk.Response{
Message: "Internal error.",
})
return user, rbac.Subject{}, false
}
// If the user logged into a suspended account, reject the login request.
if userStatus != database.UserStatusActive {
httpapi.Write(ctx, rw, http.StatusUnauthorized, codersdk.Response{
Message: fmt.Sprintf("Your account is %s. Contact an admin to reactivate your account.", userStatus),
})
return user, rbac.Subject{}, false
}
First we activate the user, then re-query their status, and finally check whether it is now active. But, the dormant job could come in and mark the user dormant between these queries. It only checks last_seen_at, which isn't updated when the user is marked active.
For correct behavior, we should set last_seen_at in the query that activates the dormant user. This way the dormant job cannot mark us dormant again for 90 days.
Environment
- Host OS:
- Coder version:
Additional Context
No response