[CI] Jenkin CI server (xgboost-ci.net) down for emergency maintenance

[hcho3]

All Jenkins jobs are now failing due to broken SSL connections, so git clone is failing and also S3 upload is failing. I suspect this has to do with outdated software packages for the worker images we are using.

At any rate, I will need to re-generate the worker images (AMIs), so I'm shutting down the server for a few hours for a day. I will do my best to bring the server back to normal.

@dmlc/xgboost-committer

[hcho3]

"Wouldn't Docker solve the issue of outdated packages"? At the Checkout stage, the CI worker uses the system Git to checkout the XGBoost repository. The Docker image is not available yet, since the Dockerfile is itself part of the repository. Hence, over time, we would still be affected by outdated system packages. Also, unfortunately, we cannot use Docker for Windows yet (NVIDIA/nvidia-docker#429).

[trivialfis]

Thanks for looking into this.

[hcho3]

Update: git clone is now working correctly.

https://xgboost-ci.net/blue/organizations/jenkins/xgboost/detail/PR-4924/2/pipeline/12

[hcho3]

Note to myself: List of software packages to set up Linux workers.

• System-wide update (apt-get upgrade)
• Python 3.x + PIP (python3, python3-pip)
• awscli Python package
• Git
• Docker CE
• OpenJDK 8
• CUDA (for GPU only)
• NVIDIA docker (for GPU only)

[hcho3]

Update: S3 transfer is now working too.

https://xgboost-ci.net/blue/organizations/jenkins/xgboost-win64/detail/PR-4924/1/pipeline/71/

[hcho3]

It looks like Jenkins CI is back to normal. However, I will keep a close eye on it today and tomorrow to see everything is all right.

Note. As a bonus, the Linux GPU workers are now using shiny new Turing GPUs (G4 instances)