Does iterating on features that require replays usually involve downtime?

[davidharting]

There may be some bad assumptions built into this question! Please point that out if I'm being wrong-headed here.

I've read through the docs and listened to Daniel on podcasts, and drawn two conclusions that seem to contradict a little

1. Make models that are derived from handle() methods minimal. Just what the UI needs.
2. Replaying events will likely involve downtime (truncate handled()d model tables, and replay. Downtime while waiting for that).

For an event-sourced app with active users and active development, does this cause tension?

If replaying events were free, you would want to do so as part of every deployment, just like migrations(right?). But they may involve downtime, so do you need to plan for when you replay events and fully release new features that involve updated projected models?

Curious what folks with production experience on Verbs have to say about this! Is this tension real?