Skip to content

Latest commit

 

History

History
121 lines (95 loc) · 4.47 KB

README.md

File metadata and controls

121 lines (95 loc) · 4.47 KB
Raw

Setting up to run FireWorks

One-time setup

To use FireWorks, your computer (to run the lpad command) and the Firetask worker nodes need to access a MongoDB database holding the workflow information.

Creating a MongoDB database

The team has a MongoDB server in the Allen Center project on Google Cloud. It's useful for running FireWorks in Google Cloud and on our local computers but we run that server only when needed. (See How to run the Whole Cell Model on the Google Cloud Platform for details.) You can access the service securely via an ssh tunnel with your Google login. Creating a database there just takes the steps in the next section.

That MongoDB server in Google Cloud is not accessible to the open Intenet or to Sherlock. To run workflows on Sherlock or elsewhere, you can create an account and database on MongoDB Atlas. See the instructions in the MAT project forum to set it up including using the lpad init -u command to create a my_launchpad.yaml that configures FireWorks to connect to it.

CAUTION: Your Atlas database will be reachable to the entire Internet. So use a password vault to generate and save long, random passwords for the account. Never reuse passwords.

Config YAML files for Fireworks

Fireworks needs a launchpad config file, my_launchpad.yaml. When running with a job queue via qlaunch (such as with SLURM on Sherlock, not with worker nodes on Google Compute Engine), it also needs a queue-adapter file, my_qadapter.yaml.

  1. Create the my_launchpad.yaml file.

    • Atlas MongoDB cluster: See the MAT project forum on how to use the lpad init -u command to create a my_launchpad.yaml looking something like this (after the host setting, are the rest defaults that could be omitted?):

      authsource: admin
host: mongodb+srv://<atlas-username>:<atlas-password>@<atlas-cluster>/wcEcoli?retryWrites=true&w=majority
logdir: null
mongoclient_kwargs: {}
name: null
password: null
port: null
ssl: false
ssl_ca_certs: null
ssl_certfile: null
ssl_keyfile: null
ssl_pem_passphrase: null
strm_lvl: INFO
uri_mode: true
user_indices: []
username: null
wf_user_indices: []

      (The tool wholecell/fireworks/initialize.py could become useful again if updated.)

  • Google Cloud MongoDB database: Run the script runscripts/cloud/mongo-ssh.sh to open an ssh tunnel whenever you want to access the server. (You can leave it running in one terminal tab then ^C it when you're done, or run runscripts/cloud/mongo-ssh.sh bg to run it in a background process and kill it when you're done.)

    Construct my_launchpad.yaml like this:

    host: localhost
port: 27017
name: <your-username-or-another-unique-name-for-your-database>
username: null
password: null

    These values ask FireWorks to connect to localhost:27017, which the mongo-ssh.sh tunnel will forward to port 27017 on the Google Compute Engine MongoDB server.

    name will be your database name.

    We're not using user authentication on this server since the ssh tunnel already requires Google login.

  1. Run

    lpad reset

    This will connect to the MongoDB server then prompt you to type Yes to confirm creating and initializing the database. If it fails to connect, check that the mongo-ssh.sh tunnel is still running.

    If you want additional launchpad databases, create a launchpad yaml file for each one. To use a launchpad config filename besides the default my_launchpad.yaml, pass it as a command option like lpad -l gce_launchpad.yaml webgui.

  2. If you need a queue-adapter to run FireWorks workers on Sherlock, write my_qadapter.yaml from the template wholecell/fireworks/templates/my_qadapter.yaml. Or you can run the interactive script python -m wholecell.fireworks.initialize, but beware that it overwrites my_launchpad.yaml and my_qadapter.yaml.

The launchpad database keeps the status of current and past workflows. You can rerun past workflows without re-uploading them, archive or delete workflows, and mix multiple workflows in the same database, but the simplest approach is to lpad reset it and upload a new workflow each time.