ARC competition
What is ARC about?
The goal is to solve never-before-seen puzzles.
This is a contest that happens annually in December.
Each participant upload a docker image containing their puzzle solving code.
The contest organizers runs all the docker images.
Input/output is provided via the /data dir. Input is the puzzles. Output is the predicted solutions.
What is the arc-competition dir?
This dir contain tools that automate constructing of a docker image.
PROMPT> rake
rake buildx-build # Create a docker image with the LODA-RUST executable inside
rake buildx-create # Create a docker builder named 'my_loda_builder'
rake payload # Prepare the contents of the /root dir
rake remove-buildx-instance # Remove the buildx instance - after building the docker image, it's no longer needed
rake run # Runs the LODA-RUST process in an isolated container, in an enviroment similar to the ARCathon submission specification
rake save-tar # Export docker image to tar file
rake shell # Runs an interactive bash shell inside the docker image, so it's possible to troubleshoot
The payload dir is copied into the /root dir inside the docker image, which is the $HOME dir.
The secret_data dir is mounted at /data inside the docker image. This is how the docker image communicates with the outside world.
The /data/solution/solution_notXORdinary.json contains the predicted solutions. While the puzzle solver code is running, it continuously updates this file with its findings.
Deploy docker image
These are the steps to create a docker image and submit it to the contest.
Step 1 - Bump version of LODA-RUST
In the loda-rust-cli/Cargo.toml
Insert todays date, like this: version = "1984.12.31"
Don't commit the file in this step. Instead commit after the docker container has been created and verified.
Step 2 - Delete old docker containers and old docker images
Open the Docker Desktop app.
First manually remove old containers.
Secondly manually remove old images.
Step 3 - Login to Docker
Open the Docker Desktop app.
Navigate to Images -> REMOTE REPOSITORIES and click Sign in to the docker account.
In the terminal app.
PROMPT> docker login
Authenticating with existing credentials...
Login Succeeded
Logging in with your password grants your terminal complete access to your account.
For better security, log in with a limited-privilege personal access token. Learn more at https://docs.docker.com/go/access-tokens/
PROMPT>
Step 4 - Delete old buildx instance
PROMPT> rake remove-buildx-instance
will purge
my_loda_builder removed
did purge
PROMPT>
Step 5 - Populate payload directory
This is the data that is stored inside the docker image, such as program files, analytics data.
PROMPT> cargo build --release -p loda-rust-cli
PROMPT> loda-rust analytics-arc
PROMPT> rake payload
Step 6 - Create buildx instance
In order to cross compile for multiple architectures.
PROMPT> rake buildx-create
Step 7 - Create the docker image
This takes around 30 minutes to compile! Emulating x86 on ARM is slow.
PROMPT> rake buildx-build
Step 8 - Save the docker image to a tar file
PROMPT> rake save-tgz
latest: Pulling from username/loda-rust-cli
bb263680fde1: Pull complete
6055b99811ee: Pull complete
db6ade30b079: Pull complete
Digest: sha256:9c93f5982d4f85b8bc3e6b78fa4b39de4d04ac63b49bf9445bbdcddd7da61660
Status: Downloaded newer image for username/loda-rust-cli:latest
docker.io/username/loda-rust-cli:latest
PROMPT> ls -la
-rw-r--r-- 1 neoneye staff 38066792 Jul 19 15:56 docker_lodarust_arc_2023-07-19T13:56:04Z.tgz
PROMPT>
Step 9 - Run the docker image and see if it works
Manually copy around 60 json files from ARC/data/training to secret_data/training.
Check that this amount of json files are roughly also what is shows up when running the executable.
PROMPT> rake run
1984-01-01T12:06:54Z - Start of program
env::consts::ARCH: x86_64
env::consts::OS: linux
initial program_item_vec: 66
initial model_item_vec.len: 63
snip output
Press CTRL-C to stop it.
Verify that the architecture is x86_64. This is the only platform that is accepted for ARCathon.
Great this looks like the content of the secret_data has been mounted correct and the file has been discovered correct.
Now the .tgz can be uploaded to the contest.
Step 10 - Publish the docker image
Add the docker image .tgz file to the arcathon-docker-image repository.
After git push has finished.
Obtain the url for the docker image .tgz file, that looks like this:
https://github.com/neoneye/arcathon-docker-image/raw/main/ARCathon2023/2023-07-19T15-56.tgz
Step 11 - Check that the docker image url actually downloads the file
Paste the docker image url into the browser.
Verify that a +100mb file gets downloaded.
Verify that the file can get loaded by docker:
PROMPT> docker load < 2023-07-19T15-56.tgz
Verify that the docker image can run:
PROMPT> docker run --platform linux/amd64 --mount type=bind,source="$(pwd)"/secret_data,target=/data neoneye/loda-rust-cli:latest
prints out lots of stuff
Press CTRL-C to stop it.
We have verified that the url works, and that the docker image is runnable.
Delete the downloaded file again.
Step 12 - Submission
Great. This docker image is ready to be submitted.