Re: [Acme] idempotent requests (various issues with the spec)

[Stefan Bühler <[e-mail]>]

On Sat, 13 Jun 2015 19:04:02 -0400
[e-mail] wrote:

> Updates should be idempotent????
> 
> They are not by definition. 
> 
> In general, just use http features as little as possible. Caching is
> not at all helpful. Uri encoding is inferior to JSON structures.

Well, given there is a replay token, every action is actually
idempotent - repeating the request will not change the state further in
any way (after a request actually reached the server).

My argument is more about the underlying HTTP protocol. When you are in
the middle of a keep-alive connection, and send a POST, but the server
decided in just this moment to close the connection (yes, all fixed
with HTTP/2, at least they got this one right), the client does not know
whether the server actually received and processed the request.

With an idempotent request the client can just repeat it on a new
connection without asking for user feedback.

I think some proxies (and perhaps client implementations too) will
actually open a new connection for each POST request (although I just
found mails claiming squid does it for POST/PUT/DELETE).

regards,
Stefan