Skip to content

svthalia/bapc-infra

Repository files navigation

Setting up a domjudge system from scratch

Domserver

Running Ubuntu 20.04 LTS on AWS EC2 t2.micro (for testing purposes). Using official Docker images from https://hub.docker.com/r/domjudge/domserver/

Installation

  1. Install docker-compose via sudo apt install docker-compose
  2. Make sure docker will run directly after booting the machine: sudo systemctl enable docker
  3. Place the docker-compose.yml file in the correct directory, together with the correct .env files. For example, in /opt/domjudge/
  4. sudo docker-compose up to boot up all services. Use -d to run it in the background

This will do a couple of things.

  • An nginx proxy server will be started, routing all web traffic on port 80 and 443 to the designated docker container based on that container's virtual hostname (set as environment variable). The folder nginx will contain persistent configuration files.

    If you want to reset your configuration because of some problem, you can simply remove this nginx folder and restart. This will recreate all configurations. Note that this will also trigger new requests for certificates, so you should not do this too often to avoid rate-limiting by letsencrypt.

  • A lets-encrypt nginx companion docker will start, providing SSL certificates for the proxy server will start.

  • A mariadb database container will start, to be used by domjudge (with custom configurations from db_conf.

  • A custom domjudge domserver docker will be built (based on the official domjudge/domserver container) with some extra configurations in the start script:

    • Symfony's TRUSTED_PROXIES will be updated to correctly work with the nginx proxy server
    • Support for an external CUPS_SERVER will be set up
    • Note that we have chosen for a bare installation, so we will not have any demo data but start with a clean database.
  • An ICPC tools Contest Data Server docker will be created, to connect with the domserver. In order to correctly function, this server needs some additional setup, though.

nginx proxy and HTTPS

We use an nginx-proxy to forward multiple domains to the correct services, as well as to upgrade to HTTPS using letsencrypt. For domjudge, this is fully supported. The Contest Data Server, however, does not support this fully. It has been said that running the CDS behind an nginx proxy will break stuff (especially websockets), but at least in our configuration it does not (or we have not found anything yet that we use). The only thing that will break, is that the CDS will not show correct ip's for the connected clients in its webinterface, but we can live with that...

DOMjudge configuration

To continue setup, we first need to do some configurations in domjudge:

  • The domjudge initial admin password will be printed in the docker logs. Note that if you reboot the containter, but without wiping the database, the old admin password will NOT be overwritten with this new value. Resetting the password can be done by running docker exec -it domjudge_domjudge_1 /opt/domjudge/domserver/webapp/bin/console domjudge:reset-user-password admin
  • In order for the ICPC Contest Data Server to communicate with Domjudge, it needs API credentials. Therefore, an API reader/writer user account must be created in domjudge. We will need those credentials later
  • We also need to create a service account for the judgehosts that will be judging our submissions. Probably, domjudge will already create a user for that purpose, but verify that this indeed has happened. You also might want to set the password (read further to see how to configure judgehosts).
  • Make sure to double check the executables for the languages you want to support. For example, if you want to use cpython for judging python submissions, your python run.sh file should call python3 while PyPy requires pypy3. Note that these settings depend on the exact judgehost that you are running (read further).
  • With ./create_database_dump.sh and ./restore_database_dump.sh <file> you can create and restore backups
  • In domjudge/domjudge-images you find all images that domjudges uses in its interface. Here you can place affiliation logos, team photos and a banner.png to be shown on the static public scoreboard. Check the domjudge documentation for further instructions.

Also note that https://github.com/DOMjudge/domjudge-scripts might contain useful scripts.

Configuring a contest

When everything is setup, you can import the contest and specify the specific settings. When importing problem archives, make sure to be logged in as a user with Jury permissions and associated to a team. This way, after import all test cases that are specified in the problem archive will be submitted and judged for verification in order to check the correct working of the judgehosts. In the "judging verifier", you can then see whether all test cases were correctly verified, or that you might hit some timelimits while that should not be the case.

For further instructions, we refer to Domjudge's admin documentation.

CDS configuration

If the contests are correctly set up in Domjudge, we can configure the Contest Data Server to correctly read from Domjudge.

It is important to know that, while domjudge supports users creating multiple contests from the webinterface without changing configurations or requiring server access, the CDS does require you to change some configuration file for the contests you want it to work on.

Tthe cdsConfig.xml file is mounted as volume and contains information about the contests that the CDS will work with. Make sure that each contest the CDS must use is configured in here, with the correct API credentials for DOMjudge. An example file could be:

Note that /contest-data is also mounted as volume to be the root for all Contest Data Packages.

After the configuration has changed, make sure to reboot the CDS with docker-compose restart cds.

Contest Data Package

For each contest, the Contest Data Server will pull most of its data from domjudge via the API. However for some features, additional data is required that are not provided by domjudge: team pictures, affiliation logos, contest logo and banner, photos and logos for usage in presentations, staff-members, floor-plan, affiliation coordinates. All these data can be manually written to the mounted contest data package.

CUPS configuration

The CUPS server is based on a simple debian image running a container. It does not use the nginx-proxy for SSL (as this would not work), but the service is exposed on port 631 regardless of hostname. The certificates are mounted as a volume, which is a bit hacky, but it works.

All configuration files are persitently mounted at cups-config/ but can also be managed via the webinterface via HTTP(s) (at port 631). Here, printers can be configured. If the CUPS_SERVER variable of Domjudge is set to this server, all prints from domjudge will be sent to the specified (or if not the default) print queue on this server. There are numerous ways to set this up, as you can specify multiple print queues in CUPS and they will all be available to Domjudge.

Note that this CUPS server is not only accessible to Domjudge: any device can use this print server. To change this, the cupsd.conf configuration file can be adapted.

Don't forget to add a default print server for the cups system (this can be done via the web interface on port 631).

At the Radboud University

At the Radboud University, with the Peage printing system, it is possible to configure printers directly to this CUPS printserver. For best results, you want to have Direct Printing access to several printers so you can bypass FollowMe printing during a contest. At the Radboud University, this requires a separate functional print account (with a separate e-number login) although probably things could theoretically also work with a student number and KUARIO account.

The printers should be added as Windows SAMBA printers (this is just smb). The exact URL to use should include the username, password and domain(!) to use. For example, these are 3 printers at the RU:

smb://RU\eXXXXXXXX:[email protected]/KM-0026-direct
smb://RU\eXXXXXXXX:[email protected]/KM-0040-direct
smb://RU\eXXXXXXXX:[email protected]/FollowMe

Note that you might have the encode the password if you cannot authenticate (and please, please, do not forget the RU\ domain, that took ages to debug). As driver we used the driver provided by university, FollowMe.ppd (which is the print driver for the most advanced Konica Minolta MFP available).

Note that you do not print to a printer directly, but print to a virtual print server that the university runs. For Direct Printing this virtual queue does immediately print to the device, so you do not notice this too much. However if you are printing large documents or large numbers of copies, it will take very long (sometimes even more than 10 minutes!) for printers to start receiving the print job. This is because the university virtually proof-prints the jobs before releasing them to the device, to determine the number of pages and print costs. And that takes very long.

Also note that the virtual print server of the RU (or the Konica Minolta machines, but I do not expect that) does NOT support the banner page / job sheet option of CUPS! Somehow, if it is setup, the rest of the printjob is ignored. Otherwise this was the ideal way to realize watermarked print jobs for each team.

VPN tunnel

At the Radboud University, the print server is not accesible from the public network, but only via the local network or VPN. If the domserver is running outside the local network, for example on AWS, you can set up a VPN tunnel to access the printing system.

First make sure the print service can be found by adding the correct entries to /etc/hosts. At the RU, the print server print.hosting.ru.nl is at 131.174.123.28 so we need to add this manual entry 131.174.123.28 print.hosting.ru.nl to /etc/hosts.

Then, make sure the openvpn-ca-science.ovpn file contains route 131.174.123.28 255.255.255.255 so all traffic to the print server is routed through the VPN tunnel.

Use the CNCZ instructions for setting up a VPN tunnel with OpenVPN and the specific openvpn-ca-science.ovpn file in this repo to set up a VPN and tunnel specific traffic via this VPN tunnel. As this needs a Science account for authentication, you need to run this command manually during the contest.

Domjudge print command

To enable the domjudge web printing interface for team members and jury members, a print command should be added. Have a look at the domjudge documentation for what can be setup. Use for example:

enscript -b "Location: [location] - Team [teamid] [teamname]|| [original] - Page $% of $=" -a 0-10 -f Courier9 $([ ! -z [language] ] && echo "--highlight=$(echo [language] | sed 's/py[23]/python/')") [file] 2>&1

This will use the default printer on the CUPS_SERVER configured for domjudge, and print the team location, id and name, as well as a page number on the page, and it will set up language highlighting (with a fix for python 2 and 3).

Add --printer $(echo [location] | sed -e 's/HG075.*/BAPC-north/' -e 's/HG.*/BAPC-south/' -e '/^BAPC/ !s/.*/BAPC-printing/') if you want to select a specific printer based on the team location (here the print queue BAPC-north will be used for team locations that start with HG075., BAPC-south for all other HG locations, and BAPC-printing for all locations that dont start with HG. BAPC-south, BAPC-north and BAPC-printing here refer to the printer names set up in the cups web interface.

Judgehosts

Our judgehosts are run in EC2 instances in an Auto Scaling configuration. The instances themselves run a modified version of the judgehost Docker image. To create an autoscaling group like this you first need to create a template image that will be used as the disk for all the judgehosts. For this you can use any type of instance, I used a t2.micro instance, with Ubuntu 20.04 LTS.

Log in to this image and add any ssh keys you want to use.

To start the Docker container at boot, another systemd service is used:

# /etc/systemd/system/judgehost.service
[Unit]
Description=DOMJudge judgehost
After=update-judgehost.target

[Service]
ExecStart=/usr/bin/docker run --restart=on-failure --network host --privileged -v /sys/fs/cgroup:/sys/fs/cgroup:ro --name judgehost -e DOMSERVER_BASEURL=https://domjudge.thalia.nu/ -e JUDGEDAEMON_USERNAME=judgehost -e JUDGEDAEMON_PASSWORD=THEPASSWORDFORTHEJUDGEHOSTUSER domjudge/judgehost:latest
Type=exec
ExecStop=/usr/bin/docker stop judgehost
Restart=on-failure

[Install]
WantedBy=multi-user.target

You will probably want to replace the environment variables, and possibly the Docker image used for your contest. To make sure the Docker container can be started quickly, you should pull it now by running docker pull domjudge/judgehost:latest . As with the other service, you can enable this service to start at boot with: systemctl enable judgehost.service.

This is all you need to add to the template image, so now you should remove any temporarly files and possibly the line added to /etc/hosts and shutdown. Once shutdown you can create an AMI via this menu (Create Image option):

screenshot from EC2 manager, showing the create image button

Then you will have to create a Launch Template that uses this AMI. No special settings are needed here, but you will need to choose an instance type—we use the c5.large type. You can set a larger disk size too, we use 20GB. Now you can create the Auto Scaling Group from this Launch Template, and you can set it to use the latest version of the Launch Template so you don't have to change this when you update it. This Auto Scaling Group is useful because you can set the amount of judgehosts you want available and they will just appear, additionally you can use spot instances which are cheaper. To make it even easier you can use the following command:

aws autoscaling set-desired-capacity --auto-scaling-group-name bapc-runner --desired-capacity 1

Judgehost Docker image

For the 2020 BAPC prelims it was recommended to run pypy3 for Python 3 submissions. The 7.3.0 DOMjudge server and judgehost did not include the pypy3 command, and the PR for this would only be included in version 8.0.0 of DOMjudge. We included the changed dj_make_chroot in the build of our own Dockerfile and overwite the included dj_make_chroot with the new version that includes pypy3. Otherwise the judgehost Docker container is built as usual.

Useful commands

  • docker-compose up -d to start all services (remove -d to receive full logs realtime)

  • docker-compose down to stop all services. Note that this will also remove all data from the database, while stop will maintain that data.

  • docker-compose stop <service> to stop a single service (similar with start or restart).

  • docker-compose build --no-cache <directory> to do a full rebuild (as this will not automatically happen if you just docker-compose up)

  • docker system prune to remove old data

  • docker-compose logs <service> to check the logs

  • docker-compose exec -it <service> bash to enter the container of a service

  • docker-compose cp <service>:<path> ./ to copy some file from inside your container, even if the container is not running.

  • When creating your own problem archives, they must be zipped in the correct wy in order for domjudge to parse it correctly: zip ../X.zip -r *. Note that you must zip the individual files recursively. zip X.zip -r X/ will not work.

  • aws autoscaling set-desired-capacity --auto-scaling-group-name bapc-runner --desired-capacity 1 to set the judging capacity

Credits

We would like to give credits to: