Parallel execution with matrix of containers?

[pthexton]

Firstly, thanks for the work you guys have done on cirrus-cli and cirrus-tart !

Is it possible to be able leverage macOS's support for running 2 macOS VMs in parallel? For example, if I have a task I want to run on both macOS Monterey and macOS Ventura (currently beta), I can configure it as

task:
  name: Automation
  matrix:
    macos_instance:
      image: ventura
    macos_instance:
      image: monterey
...

This works, but doing a cirrus run performs the task on each OS one after the other. Presumably this may be awkward as the engine would need to know it can only run 2 at a time (should somebody configure, say, 4 different containers in the matrix section), but I was just wondering if this can already be done and I've just missed it in the documentation.

[fkorotkov]

Unfortunately, at the moment it's not possible to run tasks in parallel via CLI. The task itself is pretty complicated given that the executer will not only need to respect depends_on requirements like right now but also it will need to know about each task's requested resources. For example, check Tart VM's configuration to see CPU/Memory setup and see what is provided by the host.

And there might be a config with a mix of different instances like containers requesting different amount of CPUs and more than 2 macos_instance so the macOS limitations should also be taken in consideration...

Having said so, I think for 99% of the cases it will be good enough to introduce --parallel flag which first will run everything in parallel regardless depends_on and CPU/memory/virtualization resources. Innovator will be responsible to make sure all tasks passed to cirrus run --parallel can be actually run in parallel. And after some time depends_on support can be added. Lastly resource awareness can be added for truly hassle free experience.