Re: [apps-discuss] tail -f over HTTP -- what am I missing?

[Nico Williams <[e-mail]>]

On Wed, Apr 30, 2014 at 9:55 AM, Larry Masinter <[e-mail]> wrote:
>>> What were Range requests built for?
>> Mostly, interacting with PDFs, AIUI.
>
> http://tools.ietf.org/html/draft-ietf-http-range-retrieval-00
> notes that restarting partial downloads was also an original motivation.

Append-only logs seem like a good fit for this: downloads are always
partial!  :) (until the log is closed anyways)

> For some reason,  I had expected the web to migrate to using
> compound documents
> http://en.wikipedia.org/wiki/Compound_document
>
> but to avoid having to download the entire package before you
> could start  rendering.

Log files aren't compound documents.  I keep hearing that I should
treat log files as collections of resources, as if having millions of
100-200 byte resources scales, as if having a file per log entry would
be a good idea.  There's a reason we have log files and not log
directories.  Nor is it feasible to assign URIs to individual log
entries in a log file -- or if it is we'd have to put offsets into
those URIs and end up doing range requests in a different way.

> I think MPEG-Dash designers considered using range retrieval
> but decided it was too unreliably available (many proxies
> not doing range retrieval right).

Thanks, that's useful information.

Nico
--