4.17. Load Balance Queues and Printer Pools

A Load Balance Queue provides a way to use multiple printers for a single print queue. All jobs are normally sent to the main or load balance queue which then dispatches the jobs to server queues or printers that do the actual printing as they become available. You can also send jobs to the individual server printers if they have special processing or setups required for a particular job. Because all of the server printers are shared by the load balance queue, they are said to be in a printer pool.

Edit the printcap file so it have the contents indicated below, create the /tmp/lp2 and /tmp/lp3 files with 0777 permissions, use checkpc -f to check the printcap, and then use lpc reread to restart the lpd server.

    # printcap
    lp:force_localhost
    lp:server
      :sd=/var/spool/lpd/%P
      :sv=lp2,lp3
    lp2:force_localhost
    lp2:server
      :ss=lp
      :sd=/var/spool/lpd/%P
      :lp=/tmp/lp2
    lp3:force_localhost
    lp3:server
      :ss=lp
      :sd=/var/spool/lpd/%P
      :lp=/tmp/lp2


The :sv=... option flags the queue as a load balance queue and lists the queues that are used for load balancing. The :ss=... option flags the queue as a server for a load balance queue and specifies the name of the load balance queue. When a job is sent to the load balance queue the lpd server checks to see which server queues are available and then the first one to become available.

Execute the following commands to print the /tmp/hi file and observe the results:

    h4: {274} % lpq
    Printer: lp@h4  (subservers lp2, lp3)
     Queue: no printable jobs in queue
     Status: job 'papowell@h4+42' removed at 07:29:57.924
    Server Printer: lp2@h4  (serving lp)
     Queue: no printable jobs in queue
    Server Printer: lp3@h4  (serving lp)
     Queue: no printable jobs in queue
    h4: {275} % lpr /tmp/hi
    h4: {276} % lpq
    Printer: lp@h4  (subservers lp2, lp3)
     Queue: 1 printable job
     Server: pid 4063 active
     Status: waiting for subserver to finish at 07:31:08.074
     Rank   Owner/ID           Class Job Files        Size Time
    1      papowell@h4+62        A    62 /tmp/hi         3 07:31:07
    Server Printer: lp2@h4  (serving lp)
     Queue: no printable jobs in queue
    Server Printer: lp3@h4  (serving lp)
     Queue: no printable jobs in queue
    h4: {277} % lpq
    Printer: lp@h4  (subservers lp2, lp3)
     Queue: no printable jobs in queue
     Status: no more jobs to process in load balance queue at 07:31:12.317
    Server Printer: lp2@h4  (serving lp)
     Queue: no printable jobs in queue
    Server Printer: lp3@h4  (serving lp)
     Queue: no printable jobs in queue
     Status: job 'papowell@h4+62' removed at 07:31:10.311


The first lpq command shows how the status is displayed for a load balance queue - the queue and its server queues are shown as well. Next, we use lpr to print a job (job id papowell@h4+62). We then use a couple of lpq commands to see how the job is first sent to the lp queue, which then forwards it to the lp3 queue, which then processes it and removes it. (For purposes of demonstration we have artificially slowed down the operation of the load balance queue so that the jobs will remain in the queue for sufficient time for us to display their status.) We can send another job to the load balance queue:

    h4: {278} % lpr /tmp/hi
    h4: {279} % lpq
    Printer: lp@h4  (subservers lp2, lp3)
     Queue: no printable jobs in queue
     Status: no more jobs to process in load balance queue at 07:37:17.953
    Server Printer: lp2@h4  (serving lp)
     Queue: no printable jobs in queue
     Status: job 'papowell@h4+89' removed at 07:37:15.936
    Server Printer: lp3@h4  (serving lp)
     Queue: no printable jobs in queue
     Status: job 'papowell@h4+81' removed at 07:36:40.116


This time we see that the job was put in lp2. The normal load balance queue operation is to use the server queues in round robin order.

While this simple configuration is suitable for a large number of configurations, there are situations where server queue must be chosen dynamically. For example, if the server queues are actually transferring jobs to remote clients then as soon as the job is sent to the remote client the queue appears empty and available for use. To correctly check to see if the queue is available, the status of the remote queue or destination of the server queue must be checked.

To handle this situation, a :chooser program or filter can be used. When the load balance queue is trying to decide where to send a job, it first checks the server queues to see if they are enabled for printing. If a :chooser program is specified in the load balance queue printcap entry, then it is started with the normal filter options and environment variables, supplemented as discussed below. The :chooser program will read a list of candidate queues from its STDIN, write the chosen one to its STDOUT, and then exit. The lpd server checks the :chooser exit code - if it is zero (successful) then the chosen queue is used otherwise the exit code is used for the result value of processing the job. This allows the chooser process to not only control the destination of the job but also to hold, remove, or abort the job handling process. If the :chooser does not specify a queue, then the job is skipped and another job is chosen.

One side effect of the using a chooser program is that while there are jobs that can be processed in the queue the lpd server needs to periodically check to see if a server queue has become available. If it did this continually then a very high load would be put on the system. Instead, the chooser_interval option specifies a maximum time in seconds (default 10 seconds) between the times that the lpd server checks to see if there is an available server.

Normally, the chooser is applied to the first job in the queue. If the job cannot be printed then lpd will wait for the chooser_interval time. However, the chooser can also be used to direct jobs by their characteristics, or other criteria. This means that the entire spool spool queue has to be scanned for work. If the :chooser_scan_queue flag is set to 1, then all of the jobs are tested to see if they can be sent to an appropriate destination.

Edit the printcap file so it have the contents indicated below, create the /tmp/lp2 and /tmp/lp3 files with 0777 permissions. Then create the /tmp/chooser.script with the contents indicated below, and give it 0755 (executable) permissions. Make sure that the path to the head program used in chooser.script is correct. Use checkpc -f to check the printcap, and then use lpc reread to restart the lpd server.

    # printcap
    lp:force_localhost
    lp:server
      :sd=/var/spool/lpd/%P
      :sv=lp2,lp3
      :chooser=/tmp/chooser.script
    lp2:force_localhost
    lp2:server
      :ss=lp
      :sd=/var/spool/lpd/%P
      :lp=/tmp/lp2
    lp3:force_localhost
    lp3:server
      :ss=lp
      :sd=/var/spool/lpd/%P
      :lp=/tmp/lp2
    
    # /tmp/chooser.script
    
    #!/bin/sh
    echo CHOOSER $0 $@ >>/tmp/chooser
    set >>/tmp/chooser
    /usr/bin/head -1
    exit 0


Now run the following commands:

    h4: {280} % lpr /tmp/hi
    h4: {281} % lpq -lll
    Printer: lp@h4  (subservers lp2, lp3)
     Queue: no printable jobs in queue
     Status: CHOOSER selected 'lp3' at 14:02:50.605
     Status: transferring 'papowell@h4+178' 
          to subserver 'lp3' at 14:02:50.614
     Status: transfer 'papowell@h4+178' 
          to subserver 'lp3' finished at 14:02:50.624
     Status: job 'papowell@h4+178' removed at 14:02:50.632
     Status: starting subserver 'lp3' at 14:02:50.632
     Status: waiting for server queue process to exit at 14:02:50.651
     Status: subserver pid 10182 exit status 'JSUCC' at 14:02:52.872
     Status: no more jobs to process in load balance queue at 14:02:52.879
    Server Printer: lp2@h4  (serving lp)
     Queue: no printable jobs in queue
    Server Printer: lp3@h4  (serving lp)
     Queue: no printable jobs in queue
     Status: waiting for subserver to exit at 14:02:50.748
     Status: subserver pid 10183 starting at 14:02:50.820
     Status: accounting at start at 14:02:50.821
     Status: opening device '/tmp/lp3' at 14:02:50.833
     Status: printing job 'papowell@h4+178' at 14:02:50.834
     Status: processing 'dfA178h4.private', size 3, format 'f', \
          IF filter 'none - passthrough' at 14:02:50.838
     Status: printing finished at 14:02:50.839
     Status: accounting at end at 14:02:50.839
     Status: finished 'papowell@h4+178', status 'JSUCC' at 14:02:50.841
     Status: subserver pid 10183 exit status 'JSUCC' at 14:02:50.843
     Status: lp3@h4.private: job 'papowell@h4+178' printed at 14:02:50.856
     Status: job 'papowell@h4+178' removed at 14:02:50.871


As you see from the example above, the CHOOSER selected lp3 for use. Let us look at the /tmp/chooser file and see how the chooser.script program was run:

    CHOOSER -Apapowell@h4+113 -CA -D2000-06-01-14:02:13.313 -Hh4.private \
       -J/tmp/hi -Lpapowell -Plp -Qlp -aacct -b3 -d/var/tmp/LPD/lp \
       -hh4.private -j113 -kcfA113h4.private -l66 -npapowell -sstatus \
       -t2000-06-01-14:02:13.379 -w80 -x0 -y0 acct
    USER=papowell
    LD_LIBRARY_PATH=/lib:/usr/lib:/usr/5lib:/usr/ucblib
    HOME=/home/papowell
    PRINTCAP_ENTRY=lp
     :chooser=/var/tmp/LPD/chooser
     :lp=/tmp/lp
     :sd=/var/tmp/LPD/lp
     :server
     :sv=lp2,lp3
    
    lp2=change=0x0
     done_time=0x1
     held=0x0
     move=0x0
     printable=0x0
     printer=lp2
     printing_aborted=0x0
     printing_disabled=0x0
     queue_control_file=control.lp2
     server=0
     spooldir=/var/tmp/LPD/lp2
    lp3=change=0x0
     done_time=0x2
     held=0x0
     move=0x0
     printable=0x0
     printer=lp3
     printing_aborted=0x0
     printing_disabled=0x0
     queue_control_file=control.lp3
     server=0
     spooldir=/var/tmp/LPD/lp3
    PS1=$
    OPTIND=1
    PS2=>
    SPOOL_DIR=/var/tmp/LPD/lp
    LOGNAME=papowell
    CONTROL=Hh4.private
    Ppapowell
    J/tmp/hi
    CA
    Lpapowell
    Apapowell@h4+113
    D2000-06-01-14:02:13.313
    Qlp
    N/tmp/hi
    fdfA113h4.private
    UdfA113h4.private


As you can see, the program is invoked with the same options as a normal filter. In addition, the printcap information for each server queue is passed in an environment variable with the name of the server queue. This means that if there is information needed by the chooser program to test for the availability of hardware, etc., this can be placed in the printcap information.

One of the limitations of using the :chooser program is that you may have a high overhead associated with running the program. The LPRng software provides support for linking in a user provided routine that does the same thing as the :chooser program. This routine has the following API or interface:

    Printcap Option: chooser_routine
       chooser_routine@  - default - do not use chooser routine
       chooser_routine   - use chooser routine
    
    Configuration:
       configure --with-chooser_routine=name --with-user_objs=objectfile.o
          defines the CHOOSER_ROUTINE compilation option to name
          includes the objectfile.o in the library.
    
    extern int CHOOSER_ROUTINE( struct line_list *servers,
        struct line_list *available, int *use_subserver );
      servers:    all subserver queues for this load balance queue
      available:  subserver queues to choose from
      use_subserver: chosen subserver queue
      RETURNS:
         0 - use the 'use_subserver' value as index into servers
             list for server to use
         != 0 - set job status to value returned.


See the LPRng/src/common/lpd_jobs.c and LPRng/src/common/user_objs.c files for details of the servers, available, and user_subserver parameters. The user_objs.c file provides a simple template that can be used as a starting point for a more complex routine. You should modify the code in the user_objs.c file and then use the configure options shown above to cause the user_objs.c file to be compiled and linked into the LPRng executables.