Reduce parallelism and split validation per language

Today, there is one validation workflow per building block (new quickstarts) and within each there is a multiplication matrix for programming language and protocol/SDK. It helps to avoid validation to fail due to left over state from a previous scenario, but this causes too many jobs to be triggered, exhausting the available workers for other workflows.

Instead, split the workflows per programming language. Within each workflow, multiplex only per protocol and have a loop within the workflow to run validation for each building block. This way, there would still be isolation for each building block but reduced number of jobs triggered.

Summary:

• Move from workflow file per building block to workflow per language in .github/workflows
• Matrix is only for [sdk, http] (variant)
• Building blocks run in sequence in a single step with a loop that understands the folder structure, so when a new building block quickstart is added for a language, it is automatically picked up without requiring code changes:

for building_block in `ls -1`; do # get list of building blocks from folders or from env variable in global.env
  # TODO: add check if folder exists.
  cd $building_block/$language/$variant
  make validate 
  cd ../../../
done

[paulyuk]

Note @ASHIQUEMD is signing on to drive this. Thank you MD!

[paulyuk]

I like it that there’s still some split so that:

• one workflow isn’t too long
• the isolation prevents 1 flakey issue taking down the whole matrix

The split by Building block has been good so that any new building block we introduce, does not affect the others that are in good shape. We’ll add building blocks more often than adding languages. Also you sort of have to “test in main” with these things, so i think authoring will get harder. but if it makes daily runtime better I’m still good with trying something and optimizing that way.

We should definitely solve too many jobs being triggered.