Re: [discuss] Authoxy reports Interrupted system call when accessing pages in IE

[bruce]

Thanks I will have a look asap but our servers will be down tomorrow until 3:30pm.
The local power supply authority is changing power cables tommorow and our UPS's only last 45 minutes. So we will shut down at 8:30am and back up at 3:30pm.
I shall try after that!!
Thanks for the CVS will try and get that tonight!!

As for multiple daemons, you imply that they spawn at the time
of incoming messages. If that is the case, then they will probably be too slow.
For a comparison, Apache spawns n daemons that wait for incoming messages
and when they are used and die, a new process is spawned/restarted immediately
and these daemons simply wait for their next message. In other words the spawning/slow process start has already been done. The number of daemons is specified in the httpd.conf file and the n daemons are spawned at startup.
In the case of authoxy, if you suddenly receive say 8 port 80/8080 requests
then all of a sudden the poor Unix system has to rapidly create 8 new daemons
with resources and attach the sockets etc... Would take ages and some requests would probably timeout. Live Apache daemons on the other hand have a cycle time of say 1/100
of a sec and if say 16 daemons are running at any one time they can handle say
1600 hits per second which is more like it. The hard work has already been done.

This was why ages ago I asked if we could have an indicator of how many
daemons were active etc in say the menu bar or somewhere else. Its really
a fascinating process and will allow control of how the authoxy server
will operate. 

Another one is the name server cacheing daemon although I've never really
got into that one.

Cheers,
Bruce.

PS Now to sort out CVS which I have never really got into I have to admit... Opps sorry


>Hi Bruce,
>
>On 09/06/2004, at 11:20 AM, bruce wrote:
>> Hope this helps perhaps..
>
>Yeah, it does. There are some good points in there - thanks very much!

>FYI, Authoxy is designed to spawn as many deamons as needed to handle

>the incoming requests. I haven't seen any evidence of them clashing 
>until now. In fact, they are designed to play nicely with each other.

>Nonetheless, your points about daemons not dying right away may indeed

>be relevant.
>
>> PS Love to see your source code - promise not to distribute!!
>
>Distribute all you want, it's open! I took a big step early in May and

>GPL'ed the code. Please go take a look: 
>
>
>Heath
>-- 
>  
>|   Heath Raftery<[EMAIL PROTECTED]> |
>|   HRSoftWorks  |
>||
>|   *There's nothing like a depressant to cheer you up*  |
>|   - Heard at Moe's Tavern  |
>| _\|/_  |
>|m(. .)m_|

[discuss] Authoxy reports Interrupted system call when accessing pages in IE

[Heath Raftery]

Hi gang,
Hoping someone has seen this before, because I'm in the dark a bit 
here. I have a user who is trying to get her G5 w/ OS X 10.3.4 working 
past an ISA proxy server w/ Small Business Server. So she's trying 
Authoxy's NTLM support.

Apparently it worked once upon a time, but now everytime she goes to 
view a web page in IE, Authoxy reports:

Jun  8 15:07:19 : Fatal Error: unable to connect to talker socket. 
Errno:
Interrupted system call
Jun  8 15:07:19 : Couldn't open connection to proxy server. Errno:
Interrupted system call
every second or so. She can surf around for a bit but gets those 
messages on every page she hits. Then it all falls apart when she 
attempts to access a secure site. IE shows an incomplete error msg "The 
attempt to load 'Accessing URL: ..." and then Authoxy spits this out:

Jun  8 15:09:12 : Fatal Error: unable to create shared memory
Since I developed Authoxy I have a little bit of inside knowledge about 
these error messages, but not enough to see exactly what the problem 
is. In fact, I've never actually seen the messages appear in practice.

The talker socket is the connection to the proxy. The failure is on the 
system call to connect() but the man page doesn't show any information 
on that particular Errno. A bit of web searching shows that 
"Interrupted system call" (also known as EINTR) occurs when a blocking 
call is interrupted by another signal (like a break or quit) but it 
shouldn't actually appear on OS X because FreeBSD is designed to 
automatically restart the call! Odd stuff - I'm leaning towards the 
possibility that perhaps IE itself isn't happy with the connection 
attempt (it times out perhaps) and cancels the call.

And the second error is unusual too. It occurs when the system call 
shmget fails in the NTLM code. Reasons are that the key, which is based 
on the PID of authoxyd, is not unique or because there are no resources 
left to create more shared memory. Anyone else seeing the error?

Any chance anyone has seen similar behaviour from Authoxy, and might 
know of things to try? It is very hard to troubleshoot these things 
without access to the site...

Regards,
Heath
--
 
|   Heath Raftery<[EMAIL PROTECTED]> |
|   HRSoftWorks  |
||
|   *The search for a new personality is futile; what is fruitful is |
| the interest the old personality can take in new activities*   |
| _\|/_  |
|m(. .)m_|