Page Comparison

...

• Code Block
  export MGMT_SERVER_IP=173.31.23.23 export SLURM_CONF_DIR=/opt/slurm/etc

Compute Environment

An AMI will be needed. The steps below give an example of how an AMI could be created in the AWS Console leveraging the existing slurm cluster and its resources and configuration. Or if the compute resources in the slurm cluster are based on an existing AMI, that can be used as wellTo prepare the nested compute node we need to process an AMI. If you have an existing AMI for your compute environment we can leverage that. If an AMI needs to be created from an existing compute node the following steps walk through that process.

The key concepts to keep in mind about getting a good AMI is that it should boot fast and do little to no work, for example in bootstrapping or user_data. Also, it should have everything required to run the workflows and authenticate users.

...

• Allocate a compute node:

  • Code Block
    salloc -N 1 -J ami-creation --no-shell --exclusive --nodelist=<NodeNanme>

• When the job is allocated, gather some information on the node running the job:

  • The salloc command above should have output a JOB_ID.

  • The squeue command should show the JOB_ID running on particular node.

  • Issue the following command to capture information about that node:

    • Code Block
      scontrol show node <NODENAME>

    • Look for the NodeAddr= field in the output to find the Private IPv4 Address of the node running the ami-creation job.

  • Navigate to AWS console EC2 Instances page and search for the Private IPv4 Address.

  • Select that instance and click “Create an image” from the Actions button in the upper right corner.

  • AMI Creation requires a variable amount of time to complete: it may take 15 minutes or longer in some regions at certain times of day. Information requested from the AWS Console to fulfill this request is fairly intuitive: Name the AMI in a memorable way and generally takes several minutes to complete. Give the AMI a unique name, optionally a description, and accept the default values provided by the prompts.

  • NOTE : this action will reboot the node and kill our job, which is expected.

...