Skip to main content

Python EXecution ENvironment, scheduler included

Project description

Python EXecution ENvironment

This is a (fairly) simple framework for building anything that requires some kind of ordered execution, such as a build system, a test suite, or similar.

How to use it

See further topics (in this order):

  1. Scheduler
  2. Task metadata
  3. Factory

Current state

The project currently sits idle until I find a use for it. It is not under very active development, but it's also not abandonware.

Feel free to fork it and make enhancements if you wish.

No APIs, names or anything is currently declared as stable, if you use this project in your projects, expect having to rename/rework things some time in the future.


The idea came from limitations of various (non-Python) test suites that used static (ini) files for configuration, limiting options for dynamic execution, not having parallel execution support (or a crappy make-job-like one), etc.

Thus the goal was to create a system which would support

  1. Tree-based structure of tests, no single (flat) or double (main/sub) levels
  2. Dependency tracking and resolution between tests
  3. Parallel execution of tests with resolved dependencies
  4. Dynamic runtime configuration (selecting tests based on OS, etc.)
  5. Multiple results per one test (ie. 'pass', but with problems detected)
  6. Persistent state keeping (for re-runs of tests, for OS reboot)
  7. Some watchdog for keeping test runtime in check
  8. etc.

and the idea was to utilize python as a "configuration" language.

Generalizing this, the above formed into

  1. A "factory-time" logic, which would traverse a tree structure and generate the tests (a.k.a. execution units, python callables) based on some input
  2. An "execution-time" logic, which would use a scheduler to efficiently run the collected tests

Further generalizing, it ultimately turned into an universal execution engine info which you can feed any python callables and, given the right metadata, they get scheduled and run in the right order, with the "factory-time" being essentially just an example way of how to generate execution units.


Needs python 3.6+, mostly due to quality-of-life features. Functionally, it could be ported to python 2.7 as it currently uses only basic threading and multiprocessing features (eg. no futures), at the expense of code readability.

  • Parallel result pipelining
    • Wish
      • A thread-safe process-safe interface to get results from running tasks
      • The user can then spawn a separate custom pool of threads/processes that query this interface and post-process results returned from tasks, in parallel
        • While other tasks are still running, not after everything finishes
    • Why Not
      • There are process-unsafe metadata returned to the user, namely the callable objects that were executed, exception objects, etc.
        • These are safe to pass only to the thread/process that added the task
      • There is a simple workaround: simply take care of the metadata in the thread that added the tasks and pass on picklable data (ie. retvals from the executed callables) to the custom process pool for further post-processing


  • Scheduling groups of exclusivity

    • Pass groups of callables instead of just callables to the scheduler
    • Have only one group running at a time
      • As it finishes, destroy the worker pool, create a new one for the new group
    • Allows to specify nr. of workers/spare and pool type (thread/mp) used on a per group basis!
    • Useful when a set of tasks is very different from the rest, ie. testing virtual machines (5 workers) versus testing syscalls (100 workers), and one global setting cannot cover both.
    • Also useful to logically split mutually exclusive things that need uninterrupted / unconditional access to the OS without defining "claims" to lots of tests.
    • Also useful for tasks that should logically run "together", possibly needing sequential execution due to formal requirements.
  • Adding new tasks on-the-fly

    • While processing results from the iterator, add new tasks
    • Useful when the overall task set depends on result(s) from some tasks
    • pexen.sched.pool support already in place
      • But proper Sched support would need heavy rewrites, removal of sanity checks, etc.
        • I did the rewrite; took 6-8 hours and the code was incomprehensible
        • Refactoring of the whole pexen.sched would be needed
  • A post-processor for factories

    • A task that requires some generic resource, ie. httpd
    • The post-processor would find (somehow) another, best suitable, task that provides it, adding it to the tasks for execution
    • Similar to how Debian packages can depend on a virtual package like mail-transport-agent and have multiple packages (exim4, postfix, etc.) provide it
  • "UseCases" that would glue Factory and Scheduler

    • Ie. "test suite UseCase" for running a test suite, collecting and formatting results, storing xUnit results in a file, supporting resume from an interrupted execution, etc.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pexen-0.0.3.tar.gz (15.6 kB view hashes)

Uploaded Source

Built Distribution

pexen-0.0.3-py3-none-any.whl (16.6 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page