Feature Request: mem_used_percent metric for compute nodes

[Waqiah]

Currently the mem_used_percent metric is available for the head node only on the dashboard as explained here:
https://docs.aws.amazon.com/parallelcluster/latest/ug/cloudwatch-alarms-v3.html

Looking to collect the mem_used_percent metric (using boto3 library) from AWS parallel cluster compute nodes. Is there a workaround to achieve this in the meantime, would configuring a CloudWatch Agent to collect the metric from the compute nodes work?

[joehellmersNOAA]

I do this in the OnNodeConfigured custom action script for the compute nodes

1. Install/Update the CloudWatch agent yum install amazon-cloudwatch-agent -y
2. Configure a file to collect metrics for the memory (and other things) putting it in /root/cwagent_mem_config.json for example
3. Update the CloudWatch for this configuration: amazon-cloudwatch-agent-ctl -a fetch-config -m ec2 -c file:/root/cwagent_mem_config.json -s

In the configuration file I change the namespace to be associated with the stack name I'm using for the cluster since I often will have multiple clusters in the same AWS account.
"metrics": { "namespace": "STACK_NAME-compute",