pyinfra performance repo questions

I recently made some benchmark tests to compare sakes and Ansibles performance, and came across the pyinfra-performance repo, so decided to throw in pyinfra as well.

The results I got were a bit different from the pyinfra performance repo:

• Ansible ran faster than pyinfra for hosts < 50 for a single shell command, and hosts < 10 for multiple commands

Note, forks wasn't set for Ansible in the pyinfra performance repo, so it was limited to 5, whereas pyinfra ran with many more processes.
Perhaps I misconfigured pyinfra/Ansible somehow, the pyinfra implementation is here https://github.com/alajmo/sake-performance/tree/main/pyinfra.

Anyway, I also spotted some different behavior to how Sake/Ansible handles CLI output compared to pyinfra:

• it seems pyinfra output is directed to stderr, so it was a bit difficult to utilize gnutime for benchmarking, for instance:
  • sake: sake run ping -s servers --limit=8 > /dev/null prints nothing
  • ansible: ansible-playbook ping-playbook.yaml -i inventory.py > /dev/null prints nothing
  • pyinfra: pyinfra inventory.py pyinfra-ping.py > /dev/null prints

        --> Loading config...
    
        --> Loading inventory...
    
        --> Connecting to hosts...
        /bin/zsh: can't open input file: echo 0.0.0.0 | grep -q '^dedi[0-9]\+$'
            [host_0] Could not connect ([Errno None] Unable to connect to port 10000 on 0.0.0.0)
        --> pyinfra error: No hosts remaining!
    

• I kept getting the error /bin/zsh: can't open input file: echo 0.0.0.0 | grep -q '^dedi[0-9]\+$', even though the tasks were running just fine.

Also, I couldn't find any documentation but is this description of how pyinfra works correct:

• Parse config files
• Connect to servers serially and gathers facts
• Do work locally using the facts and figure out which operations to perform on hosts
• Perform operations on hosts in parallel by sending out shell commands

Start by reproducing the commands and output redirection examples from the issue, then compare the pyinfra-performance repository with the linked sake-performance implementation. Review how pyinfra gathers facts, runs operations, and emits CLI output. Done means the benchmark differences, stderr behavior, shell error, and execution flow are documented with clear answers.