Re: FW: Libsoup patches // Performance metrics

[Pieter Ennes <pieter watchmouse com>]

Hi Sergio,

> En 09/04/13 13:37, Ennes, Pieter escribiu:
> > Hi,
> >
> > Linked to below are two old patches to libsoup-2.30 that I'd like to discuss for enhancement:
> >
> > 1) https://bitbucket.org/watchmouse/libsoup/commits/0e514092bb9733f4c548aa65ec13cca24f240b98
> >
> > A SoupMessage 'got-raw-chunk' event, analogous to the existing 'got-chunk', but emitted just before the 
> > chunk body is decoded. It allows the client to inspect the content as it goes over the wire, which we use 
> > to get an accurate measure of the number of bytes received, but I can imagine it could serve different 
> > purposes. Or I can imagine there are easier ways to obtain the same information.
>
> The current trend in libsoup is to use the stream infrastructure instead
> of the SoupMessage signals. One example of using the network raw data is
> the SoupCache that caches it on disk.

Sounds promising; will research how it's done there, thanks.


> > 2) https://bitbucket.org/watchmouse/libsoup/commits/0a1b41879e39e216b61cadc56c320661e3364566
> >
> > A SoupSession 'connection-starting' event, analogous to 'connection-created', allowing the client to get 
> > a sense of how much time was spent setting up the connection.
> >
> > Thirdly, I'm interested (but no patch yet) in being able to accurately measure the DNS lookup time per 
> > connection created. However, we could use some guidance for the best implementation considering the async 
> > case and and the recent DNS prefetching work.
>
> I think what you need is to connect to the recently added SoupMessage's
> "network-event" signal. It will allow you to properly measure things
> like DNS resolution, TLS handshaking or proxy negotiation times.

Great! That's indeed exactly what we need going forward.

It's good to see upstream mature in a way that obsoletes our hacks.

Thanks,

- Pieter