Re: do we still want sendfile enabled with our default conf files?

["William A. Rowe, Jr." <wr...@rowe-clan.net>]

At 09:57 AM 3/17/2005, Jeff Trawick wrote:

>so I don't misunderstand, what do you consider "buggy platform" that
>APR can identify?

I don't believe it can identify every scenario you mentioned.
For what it's worth, any Win32 box that is borked for sendfile
is also likely borked for acceptex.

What I believe we can do (on platforms we can delve this) is
determine if the i_dev of the file served is sitting on a non-local
resource, and cripple sendfile automatically.

It would require some new apr_ intellegence of course.

Re: do we still want sendfile enabled with our default conf files?

[Jeff Trawick <tr...@gmail.com>]

On Thu, 17 Mar 2005 13:33:20 -0600, William A. Rowe, Jr.
<wr...@rowe-clan.net> wrote:
> At 09:57 AM 3/17/2005, Jeff Trawick wrote:
> 
> >so I don't misunderstand, what do you consider "buggy platform" that
> >APR can identify?
> 
> I don't believe it can identify every scenario you mentioned.
> For what it's worth, any Win32 box that is borked for sendfile
> is also likely borked for acceptex.
> 
> What I believe we can do (on platforms we can delve this) is
> determine if the i_dev of the file served is sitting on a non-local
> resource, and cripple sendfile automatically.
> 
> It would require some new apr_ intellegence of course.

So far that's 1 of an unknown number of cases we can possibly
identify.  We're going to need

EnableSendfile maybe

which means that APR will try to find the situations where sendfile
might not work (e.g., IPv6 or serving from non-local filesystem or LSP
installed or ???? or ????), and not use sendfile in those cases.

If you really know that sendfile can work on your particular box, even
if it is {non-local filesystem, IPv6, whatever}, code

EnableSendfile yes

If APR can't figure out properly (i.e., you still have problems), code

EnableSendfile no

Re: do we still want sendfile enabled with our default conf files?

[Brian Akins <ba...@web.turner.com>]

William A. Rowe, Jr. wrote:
>
> What I believe we can do (on platforms we can delve this) is
> determine if the i_dev of the file served is sitting on a non-local
> resource, and cripple sendfile automatically.


But, what if I know that my platform can handle this?  For example, I 
use sendfile out of tmpfs for some stuff.  This only works on later 2.6 
Linux kernels (and works very well), but the file is not "local."

+1 and let smart people enable it.


-- 
Brian Akins
Lead Systems Engineer
CNN Internet Technologies