fcgi's life and interrupted system call errors

Brad Templeton (brad@clari.net)
Thu, 11 Sep 1997 16:50:57 -0700

Message-Id: <19970911165057.36297@clari.net>
Date: Thu, 11 Sep 1997 16:50:57 -0700
From: Brad Templeton <brad@clari.net>
To: fcgi-developers@openmarket.com
Subject: fcgi's life and interrupted system call errors

This list seems close to dead.  Is Fastcgi still viable?   I'm using
it for a moderately active CGI, and in spite of the fact that I am
running Apache 1.2 and fcgiapp.c vers 1.22 (from 1.5 dev kit) with the
supposed patch to fix the problem, about once a day it all shuts down,
and starts issuing errors of the sort you've seen before.


[Thu Sep 11 04:35:15 1997] access to /usr/local/etc/httpd/htdocs/npn.fcgi failed
 for 148.147.1.15, reason: mod_fastcgi: OS error 'Interrupted system call' while
 communicating with app
[Thu Sep 11 04:35:18 1997] read script input or send script output timed out for
 140.165.2.77
[Thu Sep 11 04:35:18 1997] access to /usr/local/etc/httpd/htdocs/npn.fcgi failed
 for 140.165.2.77, reason: mod_fastcgi: OS error 'Interrupted system call' while
 communicating with app
[Thu Sep 11 04:35:31 1997] read script input or send script output timed out for
 140.157.64.181
[Thu Sep 11 04:35:31 1997] access to /usr/local/etc/httpd/htdocs/npn.fcgi failed
 for 140.157.64.181, reason: mod_fastcgi: Could not connect to application, OS e
rror 'Interrupted system call'
[Thu Sep 11 04:35:55 1997] read script input or send script output timed out for
 140.165.2.77

(and so on.)  The entire fastcgi fails to operate until the server is kicked.
For now I have a cron job that looks in the log, and if it sees this
error, sends a kill -1 to the server to restart it.

I know people got this problem last year and the code fragment

	do {
             reqDataPtr->socket = accept(FCGI_LISTENSOCK_FILENO,
                                              (struct sockaddr *) &sa.un,
                                              &clilen);
              } while ((reqDataPtr->socket<0) && (errno==EINTR))

was put in, but I still get the problem.  Anybody else have similar
experience?  Running apache 1.2, not able to tell which version of
mod_fastcgi, but I think it's the most recent.  10 processes and listen
queue of 40.