Re: limit on file descriptors (fwd)

[Randy Terbush <ra...@zyzzyva.com>]

> Solaris' stdio library is broken, and uses a byte to represent fd values.
> This appears not to have been fixed in Solaris 2.5.

I think we should consider _not_ holding the logfile descriptor open.
This would at least give us some growing room until we can sort out
a better solution. (if there is one)

Re: limit on file descriptors (fwd)

[Brian Behlendorf <br...@organic.com>]

On Wed, 5 Jun 1996, Randy Terbush wrote:
> > Solaris' stdio library is broken, and uses a byte to represent fd values.
> > This appears not to have been fixed in Solaris 2.5.
> 
> I think we should consider _not_ holding the logfile descriptor open.
> This would at least give us some growing room until we can sort out
> a better solution. (if there is one)

New directive, "TransferLogTransient", which explicitly opens and closes 
the log with each write.  That way sites can have persistant logs for 
heavy clients and save the higher overhead for less-frequently-used 
sites.  "ErrorLogTransient" is probably a good idea too.  I suppose the 
transient logs would then have to be owned by the same user as the child 
processes, a dangerous situation if there's a hole in the CGI scripts, 
but that's a hole for other reasons anyways.

	Brian

--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
brian@organic.com  |  We're hiring!  http://www.organic.com/Home/Info/Jobs/

Re: limit on file descriptors (fwd)

[David Robinson <dr...@esi.co.uk>]

On Wed, 5 Jun 1996, Randy Terbush wrote:

> > Solaris' stdio library is broken, and uses a byte to represent fd values.
> > This appears not to have been fixed in Solaris 2.5.
> 
> I think we should consider _not_ holding the logfile descriptor open.
> This would at least give us some growing room until we can sort out
> a better solution. (if there is one)

Sigh. The solution is well known; have a separate logging process.
(i.e. re-invent the syslog wheel.)

More secure and more efficient.

 David.