SYNOPSIS
       parallel processes

       ssh_parallel [-h] [-d TIME] [-o OPTION] [-s TIME] [-n NICE] [-v] direc-
       tory machine1 processes1 [machine2 processes2 [...]]

DESCRIPTION
       parallel and ssh_parallel run lines passed in on the standard input  in
       a  BASH(1) shell, in parallel.  parallel runs the commands on the local
       machine, and ssh_parallel runs the commands using SSH on any number  of
       machines. Passwordless SSH must be set up for the ssh_parallel to oper-
       ate. The program terminates when the final child process terminates.


PARALLEL
       Command line arguments:

       processes
              number of processes to run in parallel


       In the event of fork(2) failing, parallel will not loose the  line:  it
       will keep retrying.



SSH_PARALLEL
       Command line arguments:

       -h     print a help message

       -d TIME
              Check mark a machine as dead if it is unresponsive for TIME sec-
              onds.  This will cause the job to be rerun.  The  test  is  per-
              formed  using  the  ssh(1)  options  of  "ServerInterval 10" and
              "ServerAliveCountMax (5+TIME)/10", so the granularity can not be
              finer  than  10  seconds. There is no default: if this option is
              not specified, then these options will not  be  passed  to  SSH,
              ssh's own defaults will be used.

       -o OPTION
              pass OPTION to ssh(1)

       -s TIME
              set sleep timeout to TIME seconds

       -n NICE
              run processes with priority NICE

       -v     verbose operation


       machine2 processes2 ...
              Additional machine to run processes on.



       There  are several retryable errors which can occur. If ssh(1), process
       spawning, or the change to the working directory on the remote  machine
       fails,  then  the  line will be saved and re-executed. These conditions
       are indicated by the return codes 255, 254 and 253 respectively.  Since
       ssh(1)  returns  the return code of the remote executing script, if the
       remote script returns any of these error codes, it will be  rerun,  and
       the corresponding error reported.

       Note  that  some versions of ssh(1) do not follow the documentation and
       return 1, not 255 on certain errors (connection refused,  no  route  to
       host).  If ssh_parallel is built on a machine with broken ssh(1) then a
       return code of 1 is also used to indicate failure.


       A failure of ssh(1) or cd can indicate an intermittent network problem.
       As  a result, the machine which failed is put to sleep for TIME seconds
       (default 10). Each available process on a machine counts as a different
       machine.




       ssh(1) is executed as follows:
               ssh -n -o "PasswordAuthenticaton no" user@host commandline

       This  causes  ssh to fail instead of promoting for the passsword. Since
       this can be due to intermittent network problems,  it  is  a  retryable
       error.  This  also  causes ssh to disconnect from stdin, so remote pro-
       cesses can not read in read from stdin either.


EXAMPLES
       Convert all PNG files to JPEG files using 4 simultaneous processes:
               ls | grep '\.png$' | sed -e 's/\(.*\)png/convert  &  \1jpg/'  |
              parallel 4

       Perform the same operation on bigserver and hugeserver, using more pro-
       cesses:
               ls  |  grep  '\.png$' | sed -e 's/\(.*\)png/convert & \1jpg/' |
              ssh_parallel $PWD bigserver 8 hugeserver 16



AUTHOR
       Edward Rosten