So my read of this is that the old slow version of the code was less likely to hit a race with delete but now that it is optimised it goes quiclkly enought to race with reservation deleted. I think we need to add a retry_on_deadlock decorator to the expiration
So my read of this is that the old slow version of the code was less likely to hit a race with delete but now that it is optimised it goes quiclkly enought to race with reservation deleted. I think we need to add a retry_on_deadlock decorator to the expiration