kaizhang/hpc-cluster

Automate the process of building HPC clusters.

Build images

To execute the entire build workflow, run:

./buildClient run --config recipe/wangcluster_config.yaml

The steps of workflow can be visualized by:

./buildClient view workflow.html

To build/rebuild a specific image, run:

./buildClient delete Build_compute
./buildClient run --config recipe/wangcluster_config.yaml --select Build_compute

Add new machines

First add the machine specifications to the machines.txt file. The file is TAB-delimited.

To add machines, run addMachines --machines machines.txt. This will add new machines to the server but will not overwrite the existing ones.

To overwrite the existing machines, use addMachines --machines machines.txt --overwrite.

You can also add specific machines using the --include argument.

The machine information will also be appended to the file /etc/slurm/slurm.conf. But you need to manually assign the computers to different partitions and delete redundant records.

How to perform update

1. First update the frontend and reboot: yum -y update && reboot.

2. Rebuild the bootstraping kernel: wwbootstrap `uname -r` .

3. Rebuild the images

4. Readd the machines using --overwrite.

5. Restart the compute nodes.

Configuration files

• /etc/group
• /etc/passwd
• /etc/shadow
• /etc/auto.home
• /etc/auto.master
• /etc/auto.share
• /etc/autofs.conf
• /etc/slurm/slurm.conf