Skip to content

Dev_Administrator Procedures

Jump to bottom
Adrien Pavão edited this page Oct 28, 2021 · 25 revisions

Administrator procedures

Re-build the Docker containers

If you change from one branch to another (e.g. Python2 version to Python3), you need to re-build the containers using:

docker-compose build --no-cache
docker-compose up -d

Delete all contents of database

In codalab-competitions folder, on the CodaLab instance:

docker-compose up -d
docker-compose exec django bash
python manage.py reset_db
python manage.py migrate
python manage.py syncdb --migrate
exit
docker-compose restart
docker-compose up -d

Add yourself to "staff" and "superuser"

This gives access to health/simple_status and more.

docker-compose exec django bash
# python manage.py shell_plus
>>> user = ClUser.objects.get(username="your username")
>>> user.is_superuser = True
>>> user.is_staff = True
>>> user.save()

Erase user

/!/ Always check the user's email address to ensure you won't erase an account under the demand of another user /!/

docker-compose exec django bash
# python manage.py shell_plus
>>> user = ClUser.objects.get(username="user name")    # or (email="user email")
>>> user.delete() # ??? Don't seem to work at all ???

Change username

Hint: check if the request is sent from the email address linked to the user account.

docker-compose exec django bash
# python manage.py shell_plus
>>> user = ClUser.objects.get(username="new_username") # check if 'new_username' already exists
>>> user = ClUser.objects.get(username="old_username")    # or (email="user email"), good to check if the person who contacted us owns the account
>>> user.username = 'new_username'
>>> user.save()

Some platform admin functionalities

How to flush the default queue?

Log into each worker to check if they are stuck:

docker logs -f compute_worker

If one worker is stuck then restart it:

docker stop compute_worker
docker rm compute_worker
big command from the wiki

Go to RabbitMQ interface: https://<CODALAB-URL>/admin_monitoring_links/

In "Queues" tab we find the queue and restart it

This one is the default queue (virtual host: "/"):

default_queue

To determine the workers linked to this queue, click on its name, then open the "Consumers" tab

Delete all submissions from one user

From the shell plus:

def delete_submission_files(submission):
    from apps.web.utils import BundleStorage
    url = submission.s3_file
    if not url or url == '':
        logger.error("Received an invalid url to convert to a key: {}".format(url))
        return
    # Remove the beginning of the URL (before bucket name) so we just have the path(key) to the file
    sub_file_key = url.split("{}/".format(settings.AWS_STORAGE_PRIVATE_BUCKET_NAME))[-1]
    # Path could also be in a format <bucket>.<url> so check that as well
    sub_file_key = sub_file_key.split("{}.{}/".format(settings.AWS_STORAGE_PRIVATE_BUCKET_NAME, settings.AWS_S3_HOST))[-1]
    BundleStorage.delete(sub_file_key)
    data_file_attrs = [
        'inputfile',
        'runfile',
        'output_file',
        'private_output_file',
        'stdout_file',
        'stderr_file',
        'history_file',
        'scores_file',
        'coopetition_file',
        'detailed_results_file',
        'prediction_runfile',
        'prediction_output_file',
        'prediction_stdout_file',
        'prediction_stderr_file',
        'ingestion_program_stdout_file',
        'ingestion_program_stderr_file'
    ]
    for data_attr in data_file_attrs:
        attr_obj = getattr(submission, data_attr)
        storage = attr_obj.storage
        if attr_obj.name and attr_obj.name != '':
            logger.info("Attempting to delete storage file: {}".format(attr_obj.name))
            storage.delete(attr_obj.name)
my_subs = CompetitionSubmission.objects.filter(participant__user__username='pavao')  # OR CompetitionSubmission.bojects.filter(participant__user__email="<your-email>")
print(my_subs)  # Ensure they're all yours and look correct
for submission in my_subs:
    delete_submission_files(submission)
# One completed succesfully
my_subs.delete()

SSL certificate

Renewed SSL certificate should be installed before the expiration.

In the case it's not done on time:

Il suffit de copier le certificat au bon endroit, comme spécifié dans le fichier .env. Put SSL certificates in ./certs/ and they are mapped to /app/certs in the container.

To renew in case it's needed:

sudo openssl genrsa -out private.key 2048
sudo openssl req -new -out cert-codalab.csr -key private.key

Then copy/paste the content of cert-codalab.csr in GoDaddy (in "renew certificate")

5 minutes later, you'll be able to download a zip containing the new certificate.

  • Concatenate the big certificate and [...]bundle[...].crt together into [...].chained.crt
  • Rename the big certificate to codalab.crt (or how it is called in .env file)

Then restart docker-compose.

Admins URL

Remark: for RabbitMQ and Flower, the port depends on the configuration of the instance (.env file). You can get this links on "Admin Monitoring Links" under your profile section.

Clone this wiki locally