Execution Path Coverage Harness

The same CPU-and-memory burn run once through every way horus-runtime can execute work, chained so each path's usage can be measured on its own.

by Temple Compute

Run Execution Path Coverage Harness on Horus

W-41 · Execution Path Coverage Harness

The same CPU-and-memory burn run once through every way horus-runtime can execute work, chained so each path's usage can be measured on its own.

Domain: Engine Showcases

Overview

An acceptance harness for per-task resource measurement. horus-runtime executes work in structurally different ways, and a monitor that measures one does not necessarily measure the others. Each task runs the same workload (touch ~300 MiB, burn ~4 s of CPU) through one execution path; all but docker_container append their label to a chain in a stamp JSON. workflow.yaml has 4 tasks. run.py adds a fifth, python_function, which cannot be expressed in YAML because it takes a live callable. workflow.docker.yaml holds the docker_container task.

Pipeline

Tasks run one at a time so in-process tasks never share a process concurrently.

shell_command        shell + command                 subprocess on the target
shell_python_script  shell + python_script           subprocess, script shipped to target
uv_environment       uv_python_environment           subprocess inside a uv venv
python_exec_string   python + python                 in-process exec() of a code string
python_function      python_function (run.py only)   in-process direct call
docker_container     docker + command (workflow.docker.yaml)   container under the Docker daemon

Inputs / Outputs

• Inputs: none. Each task except the first reads the previous task's stamp.
• results/<task_id>.json: label, python, pid, allocated_mb, cpu_iterations, chain; the script and python_function stamps also carry peak_rss_mb, and the script stamps elapsed_s.
• uv_environment must not set runtime.python; its stamp's python is the venv interpreter, not the system one.
• The in-process tasks share the orchestrator pid, so python_exec_string and python_function stamps carry the same pid.
• docker_container declares no outputs and writes to stdout only.

Parameters

| id | Default | Meaning | |---|---|---| | alloc_mb | 300 | MiB each task allocates and touches (4-4096) | | burn_seconds | 4 | Seconds each task spins one core (1-60) |

run.py's python_function task and workflow.docker.yaml keep fixed 300 MiB / 4 s values.