NestedCollectionsSerialization.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "e21fe63f-f6a4-47fc-9497-dbb3aa7f4ddc",
   "metadata": {},
   "outputs": [],
   "source": [
    "from distributed.protocol.serialize import Serialize, serialize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "8b30db4e-55ae-4700-a73f-ce77be6eee62",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "buffer = [\n",
    "    np.random.random((4, ))\n",
    "]\n",
    "# If the array is in a simple list that is being serialize, we're just passing the memory view\n",
    "assert type(serialize(buffer[:1])[1][0]) is memoryview\n",
    "serialize(buffer[:1])[1][0] == buffer[0].data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "7a870a80-1dd1-4a29-8c07-5f2e588cd3fd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "buffer = [\n",
    "    Serialize(np.random.random((4, )))\n",
    "]\n",
    "# If we're using a Serialized object, it's also there (but there is additional payload due to the Serialized object on pos0 of the serialize output\n",
    "serialize(buffer[:1])[1][1] == buffer[0].data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "dac7a9ce-fde1-4485-9538-b3a07a3a51c6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "buffer = [\n",
    "    (1, np.random.random((4, )))\n",
    "]\n",
    "# Now, if we have nested objects, i.e. a tuple in a list we're recusing into the collections\n",
    "# However, we're only looking at the first item and check if there is a special serializer registered\n",
    "# for that item. In this case it is just an integer and we're throwing the tuple into a plain pickle.dumps, \n",
    "# i.e. it copies stuff\n",
    "assert type(serialize(buffer[:1])[1][0]) is not memoryview\n",
    "assert type(serialize(buffer[:1])[1][0]) is bytes\n",
    "serialize(buffer)[1][0] == buffer[0][1].data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "6cb4c7dd-dc9b-44c0-b637-3121bfdcc899",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "buffer = [\n",
    "    (np.random.random((4, )), 1)\n",
    "]\n",
    "# IFF, however, the first item is dask-serializable, it will attempt to do so and we have a gain a memoryview\n",
    "assert type(serialize(buffer[:1])[1][0]) is memoryview\n",
    "serialize(buffer)[1][0] == buffer[0][0].data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "2213f606-7184-40b2-8842-944c30315c7c",
   "metadata": {},
   "outputs": [],
   "source": [
    "class _List(list):\n",
    "    pass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "289e7a6b-809e-409d-ae9d-8f1b4e0eb65e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "# If we now wrap the entire thing into a custom list object or any iterable, it will still detect it\n",
    "# If it is on the first level\n",
    "buffer = _List([\n",
    "    np.random.random((4, ))\n",
    "])\n",
    "serialize(buffer)[1][1] == buffer[0].data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "306709be-7747-4a82-8ad7-a6d7d41a9881",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "# However, we no longer recurse into the nested collections and this just pickles stuff\n",
    "buffer = _List([\n",
    "    (np.random.random((4, )), 1)\n",
    "])\n",
    "serialize(buffer)[1][0] == buffer[0][0].data"
   ]
  }
 ],
 "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}