From RDDs to jointed DataFrames PySpark

Question

I'm looking for a way to combine two DataFrames by key. I started by creating Dataframes from rdds :

Given :

x = sc.parallelize([('_guid_YWKnKkcrg_Ej0icb07bhd-mXPjw-FcPi764RRhVrOxE=', 'FR', '75001'),
                ('_guid_XblBPCaB8qx9SK3D4HuAZwO-1cuBPc1GgfgNUC2PYm4=', 'TN', '8160'),
               ]
              )

y = sc.parallelize([('_guid_oX6Lu2xxHtA_T93sK6igyW5RaHH1tAsWcF0RpNx_kUQ=', 'JmJCFu3N'),
                ('_guid_hG88Yt5EUsqT8a06Cy380ga3XHPwaFylNyuvvqDslCw=', 'KNPQLQth'),
                ('_guid_YWKnKkcrg_Ej0icb07bhd-mXPjw-FcPi764RRhVrOxE=', 'KlGZj08d'),
               ]
              )

My code :

df_x = x.toDF(['id', 'countrycode', 'postalcode'])
df_y = y.toDF(['id_gigya', 'krux'])

df = df_x.join(df_y, df_x.id == df_y.id_gigya, 'fullouter')

which gives :

[Row(id=u'_guid_XblBPCaB8qx9SK3D4HuAZwO-1cuBPc1GgfgNUC2PYm4=', countrycode=u'TN', postalcode=u'8160', id_gigya=None, krux=None),
 Row(id=None, countrycode=None, postalcode=None, id_gigya=u'_guid_oX6Lu2xxHtA_T93sK6igyW5RaHH1tAsWcF0RpNx_kUQ=', krux=u'JmJCFu3N'),
 Row(id=None, countrycode=None, postalcode=None, id_gigya=u'_guid_hG88Yt5EUsqT8a06Cy380ga3XHPwaFylNyuvvqDslCw=', krux=u'KNPQLQth'),
 Row(id=u'_guid_YWKnKkcrg_Ej0icb07bhd-mXPjw-FcPi764RRhVrOxE=', countrycode=u'FR', postalcode=u'75001', id_gigya=u'_guid_YWKnKkcrg_Ej0icb07bhd-mXPjw-FcPi764RRhVrOxE=', krux=u'KlGZj08d')]

It's perfect, but I want a keep a unique id, either 'id_gigya' or 'id', since it's the same id !

With :

df_x.join(df_y, df_x.id == df_y.id_gigya, 'fullouter').drop(df_y.id_gigya).collect()

Or

df_x.join(df_y, df_x.id == df_y.id_gigya, 'fullouter').drop(df_x.id).collect()

I got this :

[Row(id=u'_guid_XblBPCaB8qx9SK3D4HuAZwO-1cuBPc1GgfgNUC2PYm4=', countrycode=u'TN', postalcode=u'8160', krux=None),
 Row(id=None, countrycode=None, postalcode=None, krux=u'JmJCFu3N'),
 Row(id=None, countrycode=None, postalcode=None, krux=u'KNPQLQth'),
 Row(id=u'_guid_YWKnKkcrg_Ej0icb07bhd-mXPjw-FcPi764RRhVrOxE=', countrycode=u'FR', postalcode=u'75001', krux=u'KlGZj08d')]

[Row(countrycode=u'TN', postalcode=u'8160', id_gigya=None, krux=None),
 Row(countrycode=None, postalcode=None, id_gigya=u'_guid_oX6Lu2xxHtA_T93sK6igyW5RaHH1tAsWcF0RpNx_kUQ=', krux=u'JmJCFu3N'),
 Row(countrycode=None, postalcode=None, id_gigya=u'_guid_hG88Yt5EUsqT8a06Cy380ga3XHPwaFylNyuvvqDslCw=', krux=u'KNPQLQth'),
 Row(countrycode=u'FR', postalcode=u'75001', id_gigya=u'_guid_YWKnKkcrg_Ej0icb07bhd-mXPjw-FcPi764RRhVrOxE=', krux=u'KlGZj08d')]

My objectif is to have, anyway, an id by row.. Ideas ? Thx !

Answer 1

Once you have your joined dataset, you can run another select to output specific columns, then convert to rdd, map it to get only non-null IDs:

df.select('id','id_gigya','countrycode','postalcode')\
  .rdd\
  .map(lambda x: Row(id=(x.id if x.id_gigya == None else x.id_gigya), postalcode=x.postalcode, countrycode=x.countrycode))\
  .collect()

which outputs:

[
  Row(countrycode=u'TN', id=u'_guid_XblBPCaB8qx9SK3D4HuAZwO-1cuBPc1GgfgNUC2PYm4=', postalcode=u'8160'),
  Row(countrycode=None, id=u'_guid_hG88Yt5EUsqT8a06Cy380ga3XHPwaFylNyuvvqDslCw=', postalcode=None), 
  Row(countrycode=u'FR', id=u'_guid_YWKnKkcrg_Ej0icb07bhd-mXPjw-FcPi764RRhVrOxE=', postalcode=u'75001'),
  Row(countrycode=None, id=u'_guid_oX6Lu2xxHtA_T93sK6igyW5RaHH1tAsWcF0RpNx_kUQ=', postalcode=None)
]