The dcosdev cli provides the one convenient entrypoint for developing operator services (sdk services) or basic service (marathon services).
Start with cloning the dcosdev repository.
git clone https://github.com/mesosphere/dcosdev.git
cd dcosdev
dcosdev uses minio as its asset repository.
Use the following two commands to install the minio service and to make it available via the dc/os public agent.
dcos package install marathon-lb --yes
dcos package install minio --yes
In your browser enter the following address.
http://<public-agent-ip>:9000
The minio credentials are minio / minio123.
Create a bucket named artifacts and set its policy to Read.
Note: Currently dcosdev has only been tested with python2 !
Install the dcosdev cli using the following command, or if you don't want to mess around with the python environment on your local machine then build a docker image as described in the next step.
pip install .
The following command builds a docker image that includes dcosdev and the dcoscli.
docker build --no-cache -f misc/Dockerfile -t mydcosdev .
In your workspace create a project folder for your new service.
mkdir myservice
cd myservice
If you chosen to build the docker image in the previous step then you use the following commands before using dcosdev.
docker run -ti --rm -e PROJECT_PATH=$(pwd) -v $(pwd):/myservice -v /var/run/docker.sock:/var/run/docker.sock mydcosdev bash
cd /myservice
Note: All dcosdev cli commands have to run from the root of your project folder.
Here the help output for dcosdev.
dcosdev -h
dcos service development tools.
Usage:
dcosdev operator new <name> <sdk-version>
dcosdev basic new <name>
dcosdev up
dcosdev build java
dcosdev test <dcos-url> [--dcos-username=<username>] [--dcos-password=<password>]
dcosdev release <package-version> <release-version> <s3-bucket> [--universe=<universe>]
dcosdev operator add java-scheduler
dcosdev operator add tests
dcosdev operator upgrade <new-sdk-version>
dcosdev (-h | --help)
dcosdev --version
Options:
-h --help Show this screen.
--version Show version.
--universe=<universe> Path to a clone of https://github.com/mesosphere/universe (or universe fork)
--dcos-username=<username> dc/os username [default: bootstrapuser]
--dcos-password=<password> dc/os password [default: deleteme]
Before you continue make sure you have the MINIO_HOST environment variable set.
export MINIO_HOST=<public-agent-ip>
dcosdev gives you the choice to create an operator services (sdk services) or basic services (marathon services).
Creating a new operator service.
dcosdev operator new myservice 0.42.1
Your myservice project will now have the following folder file structure.
myservice
|- svc.yml
|- universe
|- package.json
|- marathon.json.mustache
|- config.json
|- resource.json
If you use configuration templates in the svc then they live in the same folder as the svc.yml file.
myservice
|- svc.yml
|- myconfiguration.yml
|- universe
|- package.json
|- marathon.json.mustache
|- config.json
|- resource.json
dcosdev up will also upload the templates to the asset repository. You will have to configure the uri to the template in the resource.json file, and have a fetch statement in the marathon.json.mustache.
Use the following command to add gradle and java resources to your project for custom scheduler development.
dcosdev operator add java-scheduler
Your myservice project will now have the following folder file structure.
myservice
|- svc.yml
|- java
| |- scheduler
| |- build.gradle
| |- settings.gradle
| |- src
| |- /main/java/com/mesosphere/sdk/engine/scheduler/Main.java
|- universe
|- package.json
|- marathon.json.mustache
|- config.json
|- resource.json
You build your custom scheduler using the following command.
dcosdev build java
Note: You can add further gradle java projects as peer folders to the scheduler folder under java. TBD introduce similar support for GO.
Use the following command to add integration tests to your project.
dcosdev operator add tests
Your myservice project will now have the following folder file structure. Two tests will be added, sanity tests foldered service name deployment, overlay tests overlay networking. You will have to add further tests.
myservice
|- svc.yml
|- tests
| |- test_overlay.py
| |- test_sanity.py
| |- ...
|
|- universe
|- package.json
|- marathon.json.mustache
|- config.json
|- resource.json
You run the tests using the following command.
dcosdev test <dcos-url> [--dcos-username=<username>] [--dcos-password=<password>]
Creating a new basic service.
dcosdev basic new myservice
Your myservice project will now have the following folder file structure.
myservice
|- cmd.sh
|- universe
|- package.json
|- marathon.json.mustache
|- config.json
|- resource.json
You upload your service assets to the asset repository using the following command. This is the same command for operator and basic services.
dcosdev up
You will see the following output, showing you the dcos cli commands to take it from here.
after 1st up: dcos package repo add myservice-repo --index=0 http://minio.marathon.l4lb.thisdcos.directory:9000/artifacts/myservice/myservice-repo.json
dcos package install myservice --yes
dcos package uninstall myservice
dcos package repo remove myservice-repo
Clone the Mesosphere Universe repo to your local file system, and create a branch.
git clone https://github.com/mesosphere/universe.git
cd universe
git checkout -b myservice
If this is the 1st release of myservice, then you will have to create the myservice folder, i.e. repo/packages/M/myservice before using the release command. If its not the 1st release check the folder number of the last release in the myservice folder. The release version number to use next will have to be greater then the previous one.
Change back to your myservice project folder.
Use the dcosdev release ... command to upload the release artifacts to s3, and to add a new folder with universe files to the myservice universe branch.
dcosdev release 0.1.0-1.0.0 0 <s3-bucket> --universe=<Path to clone of https://github.com/mesosphere/universe >
Change back to your local universe. Commit the changes, push the branch, and and on github create a pull request.
...
git push origin myservice
...
Note: If you dont have permissions to push a branch to the universe, then you will have to go the universe fork route. Do the changes in your fork and then create a pull request.