A proof-of-concept serialization, transmission and restoring python runtime.

About

pyteleport is capable of making snapshots of python runtime from (almost) arbitrary state, including locals, globals and stack. It then transforms snapshots into specially designed bytecode that resumes execution from the snapshot state. The bytecode can be run remotely: this way pyteleport teleports python runtime.

Install

pip install pyteleport

Example

Teleport inside REPL:

>>> result = "something important"
>>> from pyteleport import tp_bash
>>> tp_bash("ssh", "remote")
>>> print(result)  # REPL is now running on a remote machine with all data available

Also works from anywhere in the stack:

def a():
    def b():
        def c():
            result = "hello"              # runs locally
            tp_bash(...)                  # teleport here
            return result + " world"      # runs remotely
        return len(c()) + float("3.5")    # c() runs locally, then remotely; the rest runs remotely
    return 5 * (3 + b())                  # b() runs locally, then remotely; sum and product run remotely
assert a() == 87.5                        # a() runs locally, then remotely; == and assert run remotely

API

pyteleport.tp_bash(*args), pyteleport.tp_shell(*args)

A PoC transfer of python runtime state: just specify how to reach the remote shell and pyteleport will do the rest for you.

pyteleport.dump("filename.pyc")

Save the current runtime state as a snapshot.pyc file. You may then resume or replay your python program from the snapshot state by simply running python snapshot.pyc.

How it works

• You invoke teleport in your python script.
• pyteleport collects the runtime state: globals, locals, stack.
• pyteleport dumps the runtime into a specially designed "morph" bytecode.
• The bytecode is transmitted to the target environment and passed to a python interpreter there.
• The remote python runs the bytecode which (1) develops the runtime state into the snapshot and (2) resumes running.
• The local python runtime pipes the stdio of the remote process or resumes independently (fork mode).

Known limitations

This is a proof of concept. The package works with cPython 3.9-3.14.

Current limitations:

• no thread support
• no async support (needs further investigation regarding non-python stack)
• more generally, no native code support
• back-teleport and nested teleport (never tried)

License

LICENSE.md

Useful information

• Experimenting with AWS