fd leaks plugged

[Rob Hartill <ha...@ooo.lanl.gov>]

Using the new per-request data structure, the file descriptor
leaks associated with aborted CGI requests didn't stand a chance.

I just pounded a 5 child server with constantly aborted CGI and
it stayed stable.

I plan to move a lot more globals into the data structure soon.

Is anyone working on patches for 0.7.1 ?
If not, I can power ahead without writing patch files and just
upload versions which get closer and closer to being completely
reentrant.
If you think this is getting undemocratic, say something before
I go to far.

One thing I need, which I haven't investigated so far, is a way
to pass the data structure pointer to a signal handler. easy ?

If I don't hit any problems, I might try to get this installed at
Cardiff by the end of the week.

--
Rob Hartill                           
http://nqcd.lanl.gov/~hartill/

Re: fd leaks plugged

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

On Tue, 30 May 1995, Rob Hartill wrote:
> Is anyone working on patches for 0.7.1 ?
> If not, I can power ahead without writing patch files and just
> upload versions which get closer and closer to being completely
> reentrant.
> If you think this is getting undemocratic, say something before
> I go to far.

I believe this stage is probably best accomplished by a small group 
focused on the issue, so I have no problem with this.  I'll try and keep 
up with testing it....

	Brian

--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
brian@organic.com  brian@hyperreal.com  http://www.[hyperreal,organic].com/

Aborting a connection

[Brian Tao <ta...@gate.sinica.edu.tw>]

On Tue, 30 May 1995, Rob Hartill wrote:
> 
> Using the new per-request data structure, the file descriptor
> leaks associated with aborted CGI requests didn't stand a chance.
> 
> I just pounded a 5 child server with constantly aborted CGI and
> it stayed stable.

    Quick technical question... from a client programming point of
view, how does one "abort a connection"?  Just close the socket on the
client's end and let the server figure it out?  Or is there a "nicer"
way of telling the server to stop sending data?  I'm thinking of
factoring in early termination and client drops into my HTTP benchmark
program, and closing the socket at random times seems to be the
easiest (and only?) way of doing it.
-- 
Brian ("Though this be madness, yet there is method in't") Tao
taob@gate.sinica.edu.tw <-- work ........ play --> taob@io.org