Page History

...

1. Go to https://jenkins.mxnet-ci.amazon-ml.com/job/restricted-mxnet-cd/job/mxnet-stable-release-pipeline/ and make sure you have write access to the Jenkins system.
2. Click "Configure" and update the "Branch specifier" to point to the GitHub branch/tag and click "Save" (The GitHub repository URL is set to apache/incubator-mxnet, if you are releasing using a fork, update this URL too)
   1. For branch, use <branchName>
   2. For tag, use refs/tags/<tagName>
3. Click "Build with Parameters" from the left pane, modify the following fields as needed and click on "Build"
   1. MXNET_VARIANTS : Make sure all supported cu* flavors are present
   2. RELEASE_BUILD" : Keep the box Unchecked
   3. CD_RELEASE_JOB_NAME : Do not modify this field
   4. VERSION : Mention the MXNet release tag here e.g. 1.8.0, 2.0.0 (Make sure the tag exists in MXNet github repositoryThis tag is only used in naming python docker images.)
4. This triggers the release pipelines for different cpu and cu* flavors which can be found here https://jenkins.mxnet-ci.amazon-ml.com/job/restricted-mxnet-cd/job/mxnet-stable-release-job/ . You can click on "Open Blue Ocean" from the left pane to check pipeline status. There are 3 pipelines that get trigerred:
   1. First pipeline builds the libmxnet binaries for different flavors. Only after this pipeline completes (takes around 2.5 hours), the other two pipelines are triggered, that run in parallel.
   2. Second pipeline "python/pypi" publishes the pip wheels to https://dist.mxnet.io/python/.
      
   3. Third pipeline "python/docker" publishes the python docker images to https://gallery.ecr.aws/w6z5f7h2/mxnet/python.
5. If any job in any of the three pipelines fails, go through the corresponding logs and either
   1. Re-trigger the pipeline in case it's a flaky failure or
      
   2. Root cause and fix the issue and then re-trigger the pipeline.
6. Once all pipelines are green, Python wheels and Python docker images for all variants are published to their respective targets. Now we need to move these released artifacts to their final locations so that they are available to users:
   1. Python wheels (for this release) from https://dist.mxnet.io/python/ go to PyPI.
      1. Identify the wheels by the version number, date created, and variant string
      2. Download the above wheels, give them another test if needed, and upload them to PyPI
      3. For more info regarding publishing pip wheels to Pypi, please contact zhasheng@apache.org
   2. Python docker images (for this release) from https://gallery.ecr.aws/w6z5f7h2/mxnet/python go to dockerhub.
      
      1. Identify the docker images by the version number, date created, and variant string
      2. Pull each docker image to local (docker pull public.ecr.aws/w6z5f7h2/mxnet/python:<tag>)
      3. Tag downloaded images (docker tag …)
      4. Login to dockerhub (docker login ...)
         
      5. Push images (docker push …)

...