Unix command line queue utility | Hacker News Reader

Unix command line queue utility

github.com

52 pointsevandrix19 comments

   $ 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.

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

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

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

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

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

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` ?

Thank you for making the comment I came to make!

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

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:

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

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

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

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

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.

How does this differ from task spooler:

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

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

Queues up 20 concurrent downloads with wget, for example.

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.

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.

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.

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

Reply on news.ycombinator.com