Unix command line queue utility

justin_vanw 10 years ago |

   $ mkfifo ~/.jobqueue
   $ parallel :::: ~/.jobqueue

now, from another terminal:

   echo ls > ~/.jobqueue

Tada!

for a remote job queue, use nc instead of a fifo

for a job queue with a buffer, use a file, and run $ parallel :::: <(tail -f thefile) you can even truncate the file from time to time.

Of course, this doesn't capture the context you are in, such as your current directory or the value of shell variables.

kovek 10 years ago | |

Here is an alternative:

  tty1$ mkfifo /tmp/jobqueue
  tty1$ eval $(tail -n 1 /tmp/jobqueue)

And then, in another terminal:

  tty2$ echo ls > /tmp/jobqueue

giis 10 years ago | |

What's 'parallel'? Is it a *nix command? Looks like its not available on Fedora box

teraflop 10 years ago | | |

Yes, it's a GNU tool. Try "yum install parallel".

http://www.gnu.org/software/parallel/

kovek 10 years ago |

This stackoverflow answer[0] shows how you can queue jobs even after you start running other processes in your queue. Essentially, you can do:

  sleep 2; echo "A" && sleep 2 ; echo "B" && sleep 2 ; echo "C" 
  # Quickly Ctrl-Z
  fg && sleep 2 ; echo "D"

and you will see A\nB\nC\nD\n slowly printed out in your terminal.

[0]: http://superuser.com/a/345455

CyberShadow 10 years ago |

Could you please add some usage examples or elaborate on when this tool would be useful? For example,

> building several targets of a Makefile

As opposed to `make && sudo make install` ?

> downloading multiple files one at a time

As opposed to `(wget url1 &) ; (wget url2 &)` ?

> or simply as a glorified `nohup`

As opposed to `nohup` ?

cozzyd 10 years ago | |

I imagine it's useful if you don't know all the commands you want to queue at the beginning.

For example:

    nq ./my_awesome_analysis /data/complicated_file_with_long_name.dat

    # um... what's the next file? 

    ls /data 
    nq ./my_awesome_analysis /data/even_more_complicated_path.dat

    man rsync
    nq rsync -avh --progress out* somewhere_else.net:

jessaustin 10 years ago | | |

You don't need nq for that. Just kick off ./my_lame_analysis in the background whenever you want.

cozzyd 10 years ago | | |

Yes but then they will execute at the same time instead of sequentially.

shanemhansen 10 years ago | | |

How about:

    ./my_awesome_analysis /data/complicated_file_with_long_name.dat &
    # do some stuf
    wait
    ./my_awesome_analysis /data/even_more_complicated_path.dat

e40 10 years ago | | |

It's still not equivalent, and it's more cumbersome.

PeterWhittaker 10 years ago | |

Thank you for making the comment I came to make!

The shells have had queueing for, well, a very, very long time....

jschwartzi 10 years ago |

Which systems guarantee monotonic behavior for gettimeofday()? I can't think of any. I hope there's no behavior in nq which relies on monotonicity other than file naming, because otherwise you might run into some obscure issues.

pwg 10 years ago |

How does this differ from task spooler:

http://vicerveza.homeunix.net/~viric/soft/ts/

dsl 10 years ago |

cat list.txt | xargs -n 1 -P 20 wget

Queues up 20 concurrent downloads with wget, for example.

icebraining 10 years ago | |

You could also use a for loop. But as others pointed out, that doesn't allow you to add items to the queue after it has started.

stonogo 10 years ago | | |

It does if you use a fifo instead of a text file. This is the difference between basic Unix literacy and actual Unix proficiency. I'm not being insulting; very few people take the time to learn the tools Unix provides. And naturally, a developer is going to be more interested in writing tools than most.

icebraining 10 years ago | | |

I don't think you can use a fifo with xargs, though it'd definitively work with a loop. Even with text files, using tail -f.

iokanuon 10 years ago | | |

    tail -n+0 -f list.txt | xargs -n 1 -P 20 wget