Worker Tutorial

If you’ve set up your FireServer, this tutorial will help you to:

  • Set up a remote worker that connects to the FireServer to retrieve and execute jobs

Like the previous tutorial, our purpose is to get you set up as quickly as possible; it isn’t intended to demonstrate the features of FireWorks or explain things in great detail. This tutorial can be safely completed from the command line, and requires no programming.

Launch a Rocket on a worker machine (FireWorker)

In the Introductory tutorial, we entered a Firework (job) in the LaunchPad (database) on the FireServer (central server). We then launched a Rocket that fetched the Firework from the database and executed it, all within the same machine.

A more interesting use case of FireWorks is to store FireWorks in the FireServer, but execute them on one or several outside ‘worker’ machine (FireWorkers). For example, you may have one node on a supercomputing resource dedicated to running MongoDB, but want to run FireWorks you have set up on that machine on a different supercomputing resource (e.g. running fireworks on ORNL or ALCF using a NERSC Fireserver). We’ll next configure a worker machine.

Install FireWorks on the FireWorker

On the worker machine, follow the instructions listed at Basic FireWorks Installation.

Reset the FireWorks database

  1. Back at the FireServer, let’s reset our database and add a Firework:

    lpad reset
cd <INSTALL_DIR>/fw_tutorials/worker
lpad add fw_test.yaml

Make sure to keep the MongoDB running on the FireServer, and do not launch a Rocket yet!

Connect to the FireServer from the FireWorker

The FireWorker needs to know the login information for the FireServer. On the FireWorker,

  1. Navigate to the worker tutorial directory:

    cd <INSTALL_DIR>/fw_tutorials/worker

    where <INSTALL_DIR> is your FireWorks installation directory.

  2. Modify the my_launchpad.yaml to contain the credentials of your FireServer. In particular, the host parameter must be changed to the IP address of your FireServer.

    Tip

    If you do not know the IP address of your FireServer and you are on a Linux machine, you can try running /sbin/ifconfig.

    Note

    The name my_launchpad.yaml is a special filename that contains your database credentials. By default, FireWorks checks for this file in the current directory. You can also specify its location manually using the -l parameter of lpad, or you can set up your configuration to set the location of this file once and for all.

  3. Confirm that you can query the Launchpad hosted on the FireServer from your FireWorker:

    lpad get_fws -i 1 -d all

    This should print out the description of a Firework that is READY to run.

    Tip

    If you cannot connect to the database from a remote worker, you might want to check your Firewall settings and ensure that port 27017 (the default Mongo port) is open/forwarded on the central server. For Macs, you might try the Port Map application to easily open ports. If you’re still having problems, you can use telnet to check if a port is open: telnet <HOSTNAME> <PORTNAME>, where <HOSTNAME> is your FireServer hostname and <PORTNAME> is your Mongo port (probably 27017).

Configure your FireWorker

The FireWorker file contains information about this worker’s configuration. Staying in the worker tutorial directory on the FireWorker, modify your my_fworker.yaml by changing the name parameter to something that will help you identify the worker that ran your Firework later on. For example, you might want to use the hostname of the worker machine.

Note

The name my_fworker.yaml is a special filename that contains your FireWorker’s credentials. By default, FireWorks checks for this file in the current directory. You can also specify its location manually using the -w parameter of lpad, or you can set up your configuration to set the location of this file once and for all.

Launch a Rocket on the FireWorker

  1. Staying in the worker tutorial directory on your FireWorker, type:

    rlaunch singleshot

    This should successfully launch a rocket that finds and runs your Firework from the central server.

    Tip

    Remember that we are getting database and FireWorker credentials automatically from my_launchpad.yaml and my_fworker.yaml.

  2. Confirm that the Firework was run:

    lpad get_fws -i 1 -d all

You should notice that the Firework is listed as being COMPLETED. In addition, the name parameter under the launches.fworker field should match the name that you gave to your FireWorker in my_fworker.yaml. If you have multiple FireWorkers, this can help you identify where your job ran later on.

Running rapidfire mode on the FireWorker

Just like on the central server, you can run in rapidfire mode on the FireWorker to process many jobs.

  1. Staying in the worker tutorial directory on your FireWorker, clean up your directory:

    rm FW.json howdy.txt

  2. Add three more FireWorks. Let’s do this from the FireWorker this time instead of the FireServer:

    lpad add fw_test.yaml
lpad add fw_test.yaml
lpad add fw_test.yaml

  3. Run Rockets in rapidfire mode:

    rlaunch rapidfire

You’ve now run multiple jobs on your FireWorker! You could even try running the Rocket Launcher in --nlaunches infinite mode - then, you would have FireWorker that continuously ran new jobs added to the LaunchPad on the FireServer.

Setting Machine-specific or worker-specific parameter via the env variable

From v0.7.7, the FireWorker file now supports the env key. As its name implies, this key allows you to specify machine-specific (or more accurately, worker-specific) environment settings. For example, a particular command called in your Firetasks may be called “command” in machine 1 and “command_v1.2” in machine 2. You can then abstract out this command by specifying the differences in the FireWorker file:

# For the FireWorker file on machine 1
env:
    command: command

# For the FireWorker file on machine 2
env:
    command: command_v1.2

The env can then be accessed within your Firetasks as the “_fw_env” variable in the fw_spec. For example, the run_task method for your Firetask may be something like:

def run_task(fw_spec):
    subprocess.call(fw_spec["_fw_env"]["command"])

This provides a clean way to write machine-agnostic Firetasks with an abstraction of machine-specific commands and settings. Note that you can also use dfferent fw_env settings on the same machine if you run multiple job launch scripts using different Workers on that machine.

Next Steps

A central FireServer and one or more FireWorkers pulling jobs in rapidfire mode might be all that you need to automate your application. However, if your FireWorker is a shared resource you might want to run jobs through an external queuing system rather than directly run rlaunch on your FireWorker. A description of how to run through a queue is given here: Launching Rockets through a queue. Or, you might return to the home page and pursue a different tutorial.