[pacaro]

How do you handle one way gates? Clearly at each deploy there are two different versions running concurrently, and as you make changes you do so knowing that, but as the system evolves their are points in (code) time that you can't go back to. Is this not a concern because you would never roll back that far?

[spladug]

Yeah, rollbacks are more of adding a revert to the top of the pile so we just make sure we roll back things that can be rolled back. This is important to think about when planning deploys.

[gusfoo]

> Yeah, rollbacks are more of adding a revert to the top of the pile

Neat. It reminds me of the method used in gaming nowadays, which is to just write a "savepoint" marker in to the message stream instead of pausing the entire game to save state.