dave
e16b11fe29
All checks were successful
Gitea/docker-artifact/pipeline/head This commit looks good
|
||
---|---|---|
repobot | ||
templates | ||
.dockerignore | ||
.gitignore | ||
Dockerfile | ||
Jenkinsfile | ||
README.md | ||
requirements.txt | ||
setup.py |
docker-artifact
Software repository server
Artifactd provides an HTTP API for repository management. Supported repository formats are:
- Python (Pypi)
- Apt
- Generic tarball
Quickstart
- Pull or build the image
docker run -it --rm -e 'DATABASE_URL=mysql+pymysql://...' -e 'S3_URL=http://...' -p 8080:8080 artifact
The in-container webserver will listen on port 8080 by default. Database url is passed directly to sqlalchemy, but only
mysql is tested. S3_URL is in the form of https?://keyname:keysecret@endpoint_url/bucket_name
. Amazon S3 is supported
but minio is the preferred backend.
Examples
Upload python package:
curl -vv -F 'f=@pyircbot-4.0.0.post3-py3.5.egg' 'http://localhost:8080/addpkg?provider=pypi&reponame=reponame&name=pyircbot&version=4.0.0'
Install python packages:
pip3 install -i http://host/repo/pypi/reponame/ --trusted-host host <packages>
Upload apt package:
curl -vv -F 'f=@python3_3.6.7-1~18.04_amd64.deb' 'http://host/addpkg?provider=apt&reponame=reponame&name=python3&version=3.6.7-1~18.04&dist=bionic'
Install apt packages:
wget -qO- http://host/repo/apt/reponame/pubkey | apt-key add - && \
echo "deb http://host/repo/apt/reponame bionic main" | tee -a /etc/apt/sources.list && \
apt-get update
Upload generic tarball:
curl -F 'f=@cpython-3.8.0b1.tar.gz' 'http://host/addpkg?provider=tar&reponame=cpython&name=cpython&version=3.8.0b1'
CLI
Building on the rest endpoints above:
Apt:
rpcli -s http://localhost:8080 upload -y apt -f extpython-python3.6_3.6.7_amd64.deb_trusty -r reponame -p extpython-python3.6 -i 3.6.7 -a dist=trusty
Python:
rpcli -s http://localhost:8080 upload -y pypi -f tensorflow-2.0.0a0-cp37-cp37m-manylinux1_x86_64.whl -r reponame -p tensorflow -i 2.0.0a0
Tarball:
rpcli -s http://localhost:8080 upload -y tar -f ~/Downloads/cpython-3.8.0b1.tar.gz -r cpython -p cpython -i 3.8.0b1
Notes
- Repos are created automatically when a package is added to them.
- Repo URLs are structured as:
/repo/<provider>/<name>
. URLs at and below this level are handled directly by the provider. - In the apt provider, only binary-amd64 packages are supported. No source, binary-i386 or other groups
- In the apt provider, every repo has only one component, named "main"
- The apt provider will generate a gpg key per repo upon repo creation
- The repo contents can be browsed on the web
- This uses my fork of python-dpkg, from here, which is not automatically
installed via
setup.py
due to pip limitations. - The apt provider includes a convenience shell script:
apt-get update && \
apt-get install -y wget gnupg && \
wget -qO- http://host/repo/apt/reponame/dists/trusty/install | bash -x /dev/stdin
Todo
- CLI tool (for adding packages only)
- 'Simple' cli tool (shell script fetchable from the server for adding packages)
- Centralize deleting packages
- Rpm Support
- Auth
- Support using existing GPG keys for apt
- Nicer UI
- Json API
- deb need to be able to slice package in repos by: component (arbitrary names), index (binary-amd64, binary-i386, source)
- can already slice packages by: repo, dist
- Move copysha256 somewhere generic
- Have the server dictate the S3 root path to the provider plugins
- Assert that submitted package names and file names are sane
- Assert that submitted files smell like the type of file that is intended
- Global & per-provider options:
- option to block overwriting
- Standardize what is returned from provider's web_addpkg
- Standardize some fields of provider's schema (name, version)
- Delete repos if empty (with option to disable per provider)
- Centralize the jinja template environment
- need a way for providers to register jinja filters though