Service Definition Including Target-Specific Variables

[nicolas-tallet]

On many customer platforms, we have (at least) two distinct partitions, one being CPU-only, the other one being GPU-accelerated.

Several services are required to execute on each of the two partitions but with a specific set of variables - these variables are therefore, in a sense, target-specific.

For instance, here is a service aiming at running the STREAM benchmark:

services:
  stream:
    after: []
    desc: STREAM
    fanout: 324
    require: []
    require_filter: [prolog]
    services:
      x3145:
        actions:
          stream,step2:
            cmd: |
              "%STEP2_STREAM_X3145_SRCDIR/job.sh" \
                "%STEP2_RUNDIR/$(hostname --short)/$(date +"%%FT%%T").stream" \
                "%STEP2_STREAM_X3145_THRESHOLD"
        target: "@model:X3145"
      x3515:
        actions:
          stream,step2:
            cmd: |
              "%STEP2_STREAM_X3515_SRCDIR/job.sh" \
                "%STEP2_RUNDIR/$(hostname --short)/$(date +"%%FT%%T").stream" \
                "%STEP2_STREAM_X3515_THRESHOLD"
    target: "@model:X3515"
    timeout: 60

where the working directory and the performance threshold will depend on the target.

It seems to me that duplication of the service declaration could be avoided if this was possible to declare target-specific variables.

Does this appear to be a fair request for enhancement or am I doing things wrong here?

Thanks.

[cedeyn]

Hello Nicolas,
There is no plan to provide such feature yet.

This is not the first time that we have this reported need. But Milkcheck doesn't manage any per node configuration, it relies on clustershell. To manage specific configuration on nodes, we use nodesets and groups and use that in cmd part.
It implies to have nodeset configuration synchronized on all of your nodes.

Documentation on nodeset groups available here:
https://clustershell.readthedocs.io/en/latest/config.html#node-groups

I hope it helps you.

Regards,

Aurélien