multiple cups-polld running simultaneously

We were having the same problems here and it turned out that when cupsd was spawning off the cups-polld process, it was still running as root, so although cupsd was running as cupsys, cups-polld was running as root. When the init script killed cupsd (either though stop or restart), the cups-polld process was not killed. After many restarts, there could be several of these processes churning along in the background. The solution is (hopefully) simple. I've patched my init script to run the cupsd daemon initially as the cupsys user. I didn't experience any other issues related to this on our setups, but I'll continue testing. Our environment is Ubuntu Edgy with cupsys 1.2.4-2ubuntu3.

--- cupsys.orig 2006-10-09 12:54:23.000000000 -0400
+++ cupsys 2006-12-05 15:43:22.000000000 -0500
@@ -3,6 +3,7 @@
 PATH=/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin
 DAEMON=/usr/sbin/cupsd
 NAME=cupsd
+CHUID=cupsys
 PIDFILE=/var/run/cups/$NAME.pid
 DESC="Common Unix Printing System"

@@ -46,7 +47,7 @@
          modprobe -q ppdev || true # for ISO-1284 device name detection
        fi

- start-stop-daemon --start --quiet --oknodo --pidfile "$PIDFILE" --exec $DAEMON
+ start-stop-daemon --start --quiet --oknodo --chuid ${CHUID} --pidfile "$PIDFILE" --exec $DAEMON
        log_end_msg $?
        ;;
   stop)
@@ -57,7 +58,7 @@
   restart|force-reload)
        log_begin_msg "Restarting $DESC: $NAME"
        if start-stop-daemon --stop --quiet --retry 5 --oknodo --pidfile $PIDFILE --name $NAME; then
- start-stop-daemon --start --quiet --pidfile "$PIDFILE" --exec $DAEMON
+ start-stop-daemon --start --quiet --chuid ${CHUID} --pidfile "$PIDFILE" --exec $DAEMON
        fi
        log_end_msg $?
        ;;

Hi,

Thanks for your patch, I've tested it, but I get this messages in my logs (/var/log/cups/error_log)

E [06/Dec/2006:15:06:31 +0100] Unable to bind socket for address :::631 - Permission denied.
E [06/Dec/2006:15:06:31 +0100] Unable to bind socket for address 0.0.0.0:631 - Permission denied.
E [06/Dec/2006:15:06:31 +0100] cupsdStartBrowsing: Unable to bind broadcast socket - Permission denied.

I'll try to correct it in a few hours, when people will leave our office... I'll give you more informations later... :)

Regards
Mathieu

I can't get it working :'(

If I use your patch, I can only print from the server. The printers are not polled to the clients.

Moreover, cups takes 100% of CPU when running like this.

with your patch, [cups-polld] is effectively running as cupsys
ps axu | grep cup
mathieu 21783 0.0 0.0 3912 768 pts/1 S+ 17:08 0:00 tail -f /var/log/cups/error_log
cupsys 23636 89.8 0.0 40208 4068 ? Rs 17:23 3:40 /usr/sbin/cupsd
cupsys 23637 0.0 0.0 18304 1600 ? S 17:23 0:00 cups-polld jacinthe.seyssins.gripp.fr 631 30 631

and without it, it runs as root :(
ps axu | grep cup
mathieu 21783 0.0 0.0 3912 768 pts/1 S+ 17:08 0:00 tail -f /var/log/cups/error_log
cupsys 24634 7.0 0.0 39936 3664 ? Ss 17:27 0:00 /usr/sbin/cupsd
root 24635 0.0 0.0 18300 1592 ? S 17:27 0:00 cups-polld jacinthe.seyssins.gripp.fr 631 30 631

So, i'm still using my crontab's workaround :
0 0 * * * sudo /etc/init.d/cupsys stop && sudo pkill cups-polld && sudo /etc/init.d/cupsys start

We're using Dapper and cupsys 1.2.2-0ubuntu0.6.06

After running with my patch for a while, I saw the same behavior - cupsd running at 100% CPU. In our environment, we're running only clients, so the inability to bind port 631 wasn't immediately a problem. So, it looks like my "quick fix" is a bust.

My next thought is that this might actually be a CUPS problem. I looked through the source code a bit and in scheduler/dirsvc.c, the poller is spawned with:
    if (cupsdStartProcess(polld, argv, envp, -1, -1, statusfds[1], -1,
                          0, &(pollp->pid)) < 0)

And the actual cupsdStartProcess function in scheduler/process.c takes these arguments:
cupsdStartProcess(
    const char *command, /* I - Full path to command */
    char *argv[], /* I - Command-line arguments */
    char *envp[], /* I - Environment */
    int infd, /* I - Standard input file descriptor */
    int outfd, /* I - Standard output file descriptor */
    int errfd, /* I - Standard error file descriptor */
    int backfd, /* I - Backchannel file descriptor */
    int root, /* I - Run as root? */
    int *pid) /* O - Process ID */

This would seem to indicate that the polld process shouldn't be spawned as root (argument 8 = 0 ), but it is. I'm not much of a C programmer, so I could be completely off-base here, but this seems to be at least close to the problem. I may post this to the CUPS list and see if anyone knows anything about it.

See bug #30965 for more infor about the 100% usage, and some notes about the lingering cups-polld processes.

I get 100% CPU usage without patching anything, I only have to enable BrowsePoll.

Maybe the reason for the 100% CPU usage is that it can't connect properly to the spawned cups-polld, because of socket permissions (just guessing here). This could explain why it's spinning in select().