cloudpickle-nvtabular.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "41aac980",
   "metadata": {},
   "source": [
    "This is a self-contained notebook to demonstrate potential workflow deserialization failures when pickling functions contained in external modules by reference."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf1bb726",
   "metadata": {},
   "source": [
    "We're going to create a very simple Python module here, consisting of one function.  We're creating it here because we're going to delete it later and want this notebook to be repeatable."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "c6d3875d",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(\"identity.py\", \"w\") as of:\n",
    "    of.write(\"\"\"\n",
    "def identity(col):\n",
    "    return col\n",
    "\n",
    "\"\"\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3399dde",
   "metadata": {},
   "source": [
    "This workflow simply applies the identity function to a column in a dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "2a7c6e52",
   "metadata": {},
   "outputs": [],
   "source": [
    "import nvtabular as nvt\n",
    "import identity\n",
    "\n",
    "wf = nvt.Workflow(\n",
    "    [\"col_a\"] >> nvt.ops.LambdaOp(identity.identity)\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5cbdd99a",
   "metadata": {},
   "source": [
    "Now we'll serialize the workflow."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "7b876bc2",
   "metadata": {},
   "outputs": [],
   "source": [
    "wf.save(\"identity-byref\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "51aec205",
   "metadata": {},
   "source": [
    "We'll also save another copy, this time directing `cloudpickle` to serialize every function in the `identity` module by value."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "a43a9cbd",
   "metadata": {},
   "outputs": [],
   "source": [
    "import cloudpickle\n",
    "\n",
    "cloudpickle.register_pickle_by_value(identity)\n",
    "wf.save(\"identity-byvalue\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f8dcee80",
   "metadata": {},
   "source": [
    "Now we'll clear `identity` from Python's module cache and delete the underlying file.  This effectively sets us up to simulate loading the workflow in a different Python interpreter (and in a context without the underlying module file, such as another host or within a Docker container)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a58e7523",
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "import os\n",
    "\n",
    "del sys.modules[\"identity\"]\n",
    "os.unlink(\"identity.py\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8e5d3e5e",
   "metadata": {},
   "source": [
    "Note that the workflow with `identity` serialized by reference fails to load, but the workflow with `identity` serialized by value is resilient to the absence of `identity.py`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "18af7d1d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Failed to load workflow\n",
      "No module named 'identity'\n"
     ]
    }
   ],
   "source": [
    "wf_ref = None\n",
    "\n",
    "try:\n",
    "    wf_ref = nvt.Workflow.load(\"identity-byref\")\n",
    "except ModuleNotFoundError as mnfe:\n",
    "    print(\"Failed to load workflow\")\n",
    "    print(str(mnfe))\n",
    "\n",
    "wf_ref"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "891ba720",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<nvtabular.workflow.workflow.Workflow at 0x7fef54aa3e20>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "wf_val = nvt.Workflow.load(\"identity-byvalue\")\n",
    "wf_val"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}