Spark HashPartitioning Fun.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pyspark.sql import SparkSession\n",
    "\n",
    "spark = (\n",
    "    SparkSession.builder\n",
    "    .appName('hashing-repro')\n",
    "    .getOrCreate()\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pyspark.sql import functions as psf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import uuid"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "generate_uuid = psf.udf(lambda row_id: str(uuid.uuid4()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Create a bunch of uuids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Row(id=0, uuid='ff701c08-63ef-48ff-9ccf-4faeb10add4b')"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = spark.range(10000).withColumn('uuid', generate_uuid('id'))\n",
    "test_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_df = test_df.repartition(100, 'uuid')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Write to parquet then read back\n",
    "\n",
    "This seems to be important. In the real world this would be the boundary between batch jobs or something like that..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_df.write.parquet('/tmp/test.parquet', mode='overwrite')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_df = spark.read.parquet('/tmp/test.parquet')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+--------------------+-----+\n",
      "|SPARK_PARTITION_ID()|count|\n",
      "+--------------------+-----+\n",
      "|                   1| 1404|\n",
      "|                   6| 1171|\n",
      "|                   3| 1301|\n",
      "|                   5| 1232|\n",
      "|                   4| 1278|\n",
      "|                   7|  756|\n",
      "|                   2| 1342|\n",
      "|                   0| 1516|\n",
      "+--------------------+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "test_df.groupby(psf.spark_partition_id()).count().show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Grab the first 100 rows\n",
    "\n",
    "This the right way to do this but there shouldn't be any harm to it right?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['27feca96-7e5a-4072-b3ec-b17501d0f9d6',\n",
       " 'cf639bb9-3b4b-4440-918e-86c6d9d60e1f',\n",
       " '5001f034-442f-4988-a57b-b77be8108f16']"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "first_uuids = [row['uuid'] for row in test_df.collect()[:100]]\n",
    "\n",
    "first_uuids[:3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_filtered_df = test_df.filter(test_df.uuid.isin(first_uuids)).repartition(10, 'uuid')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+--------------------+-----+\n",
      "|SPARK_PARTITION_ID()|count|\n",
      "+--------------------+-----+\n",
      "|                   1|  100|\n",
      "+--------------------+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "test_filtered_df.groupby(psf.spark_partition_id()).count().show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Other partition sizes will also produce bad results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+--------------------+-----+\n",
      "|SPARK_PARTITION_ID()|count|\n",
      "+--------------------+-----+\n",
      "|                   1|  100|\n",
      "+--------------------+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "test_filtered_df = test_df.filter(test_df.uuid.isin(first_uuids)).repartition(5, 'uuid').groupby(psf.spark_partition_id()).count().show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}