Hidden Phases for GMT overhead

[ArneTR]

When using the GMT it happens sometimes that one must do work that one feels does not really belong to the emissions of the software.

Examples include:

• Running a database migration
• sleeping between steps
• Warming up caches
• Creating helper directories or copying files around

One could argue that these phases should be hidden and the associated energy and carbon cost to be deducted from the GMT phase.

In #1135 I have supplied a possible visual implementation to highlight the idea.

The associated usage scenario would be similar to this:

... 
flow:
  - name: Hide me
    hidden: true # this is the new flag introduced
    container: gcb-alpine-stress
    commands:
# Alpine does not have stress, so we use stress-ng
# We need the -q flag because otherwise it will write debug to STDERR
      - type: console
        command: sleep 0
        note: Starting Stress


  - name: Stress
    container: gcb-alpine-stress
    commands:
# Alpine does not have stress, so we use stress-ng
# We need the -q flag because otherwise it will write debug to STDERR
      - type: console
        command: stress-ng -c 1 -t 5 -q
        note: Starting Stress

  - name: Hide me 2
    hidden: true # this is the new flag introduced
    container: gcb-alpine-stress
    commands:
# Alpine does not have stress, so we use stress-ng
# We need the -q flag because otherwise it will write debug to STDERR
      - type: console
        command: sleep 0
        note: Starting Stress

This discussion shall open up a debate wether this is a useful feature or more a horrible UX for users.

Pro

• Some stuff really is only needed to make GMT work with a measurement. Example: Copying files to a specific location as GMT makes some paths in the container read-only. With this feature these operations can be included and do not skew the measurement

Con

• For a new user it is not really clear why some phases are hidden and some are not. The result of deducting the cost of the hidden phases from the total RUNTIME would be that the sum of the parts would not equal anymore the single phases. This can very confusing.

• No carbon accounting can be done easily anymore with GMT. At the moment if one wants to understand how much carbon and energy all measurements have taken effectively one could just sum up all runs in GMT. Some people actually do that with our API bc they want to know what the measurement costs aka trying to understand the overhad cost of constantly measuring the software. Since the hidden phases have actually occured it would be weird to exclude them here ...

Happy for feedback on this!

@ribalba

[ribalba]

I really like the idea of this. It makes the output a lot clearer. I wouldn't deduct the cost for hidden operations. I would only make it an UI change. Sometimes I still want to see what the resource consumption was of something I have hidden. What is the advantage of removing the usage?