[tahoe-dev] privacy bug with google chart api

[Brian Warner]

On 7/24/10 1:04 PM, Greg Troxel wrote:
> 
> I was very surprised to see requests to
> 
>   http://chart.apis.google.com/chart
> 
> From the web UI. It seems that while one can argue that sending the
> hash of a serverid is not tha big a deal, it still reveals the IP
> address of a tahoe client to google.

Hm, yeah, I see what you mean. The chart API is only used on the "recent
uploads and downloads" portion of the human-facing webapi, specifically
the "map-update" phase of a mutable-file publish or retrieve. It reveals
three things to google:

 * a truncated version of the serverids that were used for a given
   mapupdate (as labels for certain graph rows)
 * the number of mapupdate operations this node has performed since boot
   (in the Referrer URL)
 * the internet-visible IP address of the web browser you're using to
   view that page (possibly the exterior address of a NAT box)

In return, we get an easier-to-view timeline of how servers responded
during the map-update operation, which only took me a day to implement.

I looked around briefly for a python-side way to build those sorts of
graphs (PIL comes to mind, but pure-python would be easier, as our
dependency set is already getting out of hand), but it looked like it'd
be harder to get running then the google chart api.

> By default, I think the web UI should not generate any off-machine
> traffic (other than what's necessary to do the tahoe ops requested).

Yeah, I'd be happy to see that graph generated by something else. I'd
also like to see graphs like that on all of our upload/download-status
pages, not just mutable mapupdate. In particular, I've been putting off
some #798 new-downloader work (the "OVERDUE" state machine, and the
slow/fast-server selection code) until I've got better visualization
tools to guide the changes.

I've been experimenting with a javascript library called SIMILE[1] which
provides a zoomable scrollable timeline widget, but I had some problems
making it work correctly. A library like that would require Tahoe web
clients to enable JS (which I've been avoiding for years now), but it
wouldn't involve any external servers.

cheers,
 -Brian

[1]: http://www.simile-widgets.org/timeline/