V2: getting infinite loop redirects to itself

[matt (Matt Holt)]

Yes, this is not an ideal situation, but as of right now, the two known viable defaults are:

• Data loss (without --resume)
• Scratching heads (with --resume unnecessarily)

Obviously, data loss is unacceptable, as we must guarantee a durable system.

I’m open to ideas for an alternative default that strikes a happy balance between the two and has good guarantees!

@mattiasgeniar

For Caddyfile usage, that perhaps hasn’t seen an API call just yet (a counter at 0?), wouldn’t it make more sense to load the Caddyfile config when it starts?

Hmm, I don’t think that guarantees durability in the situation where an initial config file is loaded, then the config file undergoes changes but is not intended to be reloaded yet, then the machine loses power and is brought back up. Because the counter is at 0, Caddy would unintentionally load a broken/unintended configuration instead of resuming its previous, intended one.