Skip to content

htcondor/scitokens-credmon

Repository files navigation

HTCondor Scitokens CredMon

The HTCondor Scitokens CredMon monitors and refreshes credentials that are being managed by the HTCondor CredD. This package also includes a Flask application that users can be directed to in order to obtain OAuth2 tokens from services for which an HTCondor pool administrator has configured clients. The CredMon will then monitor and refresh these OAuth2 tokens, and HTCondor can use and/or send the tokens with users' jobs as requested by the user.

*** NOTICES ***

  • Development: Development of the HTCondor Scitokens CredMon has been moved to the HTCondor source repository under src/condor_credd/condor_credmon_oauth.
  • Installation: Starting with HTCondor version 8.9.9, the condor_credmon_oauth package should be installed from the HTCondor Enterprise Linux repository. Older versions should be uninstalled first.
  • Documentation: Set up and administration of the credmon is now covered in the HTCondor Administrator's Manual, and usage of OAuth2 credentials is now covered in the HTCondor User's Manual. The remaining documentation here covers the local issuer mode of the credmon and the Docker image, which are currently undocumented in the HTCondor manual.

Prerequisites

  • HTCondor 8.8.2+
  • Python 2.7+
  • HTTPS-enabled web server (Apache, nginx, etc.)
  • WSGI server (mod_wsgi, uWSGI, gunicorn, etc.)

Docker Container

This repository provides a Docker image for users who want to experiment with a personal HTCondor install with the Scitokens CredMon installed. For details, see the instructions for using the Docker container below.

Local Credmon Mode

In the "local mode", the credmon will use a provided private key to sign a Scitoken directly, bypassing any OAuth callout. This is useful in the case where the admin wants a less-complex setup than a full OAuth deployment.

The following condor configuration directives set up the local credmon mode:

# The credential producer invoked by `condor_submit`; causes the credd to be invoked
# prior to the job being submitted.
SEC_CREDENTIAL_PRODUCER = /usr/bin/scitokens_credential_producer

# Path to the private keyfile
# LOCAL_CREDMON_PRIVATE_KEY = /etc/condor/scitokens-private.pem

# The issuer location; relying parties will need to be able to access this issuer to
# download the corresponding public key.
# LOCAL_CREDMON_ISSUER = https://$(FULL_HOSTNAME)

# The authorizations given to the token.  Should be of the form `authz:path` and
# space-separated for multiple authorizations.  The token `{username}` will be
# expanded with the user's Unix username.
# LOCAL_CREDMON_AUTHZ_TEMPLATE = read:/user/{username} write:/user/{username}

# The lifetime, in seconds, for a new token.  The credmon will continuously renew
# credentials on the submit-side.
# LOCAL_CREDMON_TOKEN_LIFETIME = 1200

# Each key must have a name that relying parties can look up; defaults to "local"
# LOCAL_CREDMON_KEY_ID = key-es356

Docker Container Setup

We assume that this container is running with the hostname schedd.client.address Register the client with a Scitokens server. When registering this client, the callback URL should be

https://schedd.client.address:443/return/scitokens

You will need to enter one of more audiences and one or more scope request templates for each audience when you register the client. These values will be used to set scitokens_oauth_resource and scitokens_oauth_permissions in the condor submit file when you run a job.

Record the values of client ID and client secret provided by the server and use them to set the build arguments below.

Build the Docker image

Obtain an X509 host certificate and key pair and put them in the directory docker in this repository. Then build the image with:

docker build \
  --build-arg SCITOKENS_CLIENT_ID='myproxy:oa4mp,2012:/client_id/xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx' \
  --build-arg SCITOKENS_CLIENT_SECRET='xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx' \
  --build-arg SCITOKENS_AUTHORIZATION_URL=https://scitokens.server.address:443/scitokens-server/authorize \
  --build-arg SCITOKENS_TOKEN_URL=https://scitokens.server.address:443/scitokens-server/token \
  --build-arg SCITOKENS_USER_URL=https://scitokens.server.address:443/scitokens-server/userinfo \
  --rm -t scitokens/htcondor-submit .

Run the Docker image

Edit docker-compose.yml to set the hostname and domainname to be the name of the machine on which this container will run so that it is visible from the outside world.

docker-compose up -d

Testing the CredMon with the Docker image

The docker container has an unpriveleged user named submitter who can submit jobs to the schedd. To log into the container as this user, run the following command from the host:

docker exec -it scitokenscredmon_scitokens-htcondor_1 /bin/su -l submitter

Once logged in, you will find a submit file names test.sub. Edit this file and set the arguments scitokens_oauth_resource to one of the audiences that the Schedd is configured to access, and set scitokens_oauth_permissions to a valid scope for that audience. For example, of you registered my.host.org as an audience with a scope template of read:/public/** in the SciTokens server, then

scitokens_oauth_resource = my.host.org
scitokens_oauth_permissions = read:/public

would be valid configurations to get a SciToken for read:/public access on my.host.org.

Once you have edited the submit file, you can submit it in the usual way with

condor_submit test.sub

You will provided with a URL to visit the CredMon to authorize the HTCondor to obtain the required SciToken. Once you have done this, you can re-submit the job by running

condor_submit test.sub

Monitor the statis of the job in the usual way by looking at the job userlog or condor_q. Once the job completes, the output file will contain a dump of:

  • The job environment as printed by env. Look for the value of _CONDOR_CREDS which is set to the path to the directory where the tokens reside.
  • A listing of the directory specified by _CONDOR_CREDS. Look for a scitokens.use file that contains the token.
  • A dump of all the files in the directory _CONDOR_CREDS, including scitokens.use. This should contain JSON that contains an access_token containing the SciToken itself.