Feature request: include user instance tags with job

[aerogt3]

I've got a use case where, when launching jobs, I'd like to tag the instances that are used for that job with specific tags, ideally immediately upon instance launch.

For example, if submitting jobs to slurm, with the job names JOB_ONE and JOB_TWO, I would want to have the instances tagged jobname=JOB_ONE and jobname=JOB_TWO for costing purposes. Is there any way this could be done via parallelcluster at instance launch time? I can perform the tagging inside the job, but this then takes several minutes after the nodes are launched.

[hgreebe]

There is currently no parallelcluster supported way to do this. I have added this request to the back log.

[maestro7879]

Would also like to see this, same use case. Cost accounting.

[aerogt3]

Is there any chance of this getting addressed in a specific release? Not having this presents quite a number of UX challenges.

For example: due to the re-queueing of jobs in slurm, jobs often switch between Q and R states many times, even though the job is not actually running or getting hardware allocated. So it becomes very difficult to determine if a job is truly running. To even attempt this you need to login to the head node, which makes it very slow.

However, if instances got user-defined tags per job on instance launch {Key:jobName, Value:JOB_ONE}, then this problem is solved in a fast and reliable way - JOB_ONE is known to be running with 100% certainty if a describe instances call filtering by tag jobName=JOB_ONE returns any instances.

And there are others as well! In general, user-defined tags can be specified at launch when running single EC2 instances, and those instances contain those tags immediately. So I can imagine a lot of EC2 customers that use or move to parallelCluster will appreciate having the same behavior, as it enables tools and workflows for single ec2 instances to very easily be adapted to parallelcluster jobs.