python - 循环后合并 RDD PySpark

Question

我正在使用 PySpark，并且正在寻找一种修改 4 个 rdd 的方法，这些 rdd 包含在列表中。当我显示我的列表时，我有这样的东西:

因此:

for r in repartionned_rdd:

    print r.collect()

给予:

[(u'_guid_NCw7SuFnCh_mFW3SI3qTvBCbqXKD4mtsdJvWE7HNgNg=', (u'f', u'KSJakOd2|KtC9ZF9h'))]
[(u'_guid_OCs2au-sKnxzPE0uRPDP4hg1vvhgpzRAAYjNWRQpKbw=', (u'f', u'KxrylzuA|KpSXJwH2')), (u'_guid_txH15ULaeUDBC4Z_NlEOj2xoYBFa-08imqIBLfYsKps=', (u'f', u'bda54c71-cd1e-4eb7-856c-ba2e6def30c8|6e189e07-807e-41a2-a60a-b07d894a2905')), (u'_guid_ehCT6NyD9l3q3NV9ZroaWVEo3bnDt4tvbU_fMBrEn1g=', (u'm', u'537D69B4-743A-45B9-BED1-A25AA5926F13|2bb3e466-edc5-4302-b102-3bddb1f8c490|aa4760de-104c-4dc3-94c3-336427f89723')), (u'_guid_9F4Ph5GztLN9IlWNgZWKPMCcT4N3Je6-93iM_130F-c=', (u'f', u'KOQqBzhU|KrDt5GC4')), (u'_guid_nPlE_f-zoOHNYiXJSGXWoVryc1U4Bnfxkow3P0mDUFY=', (u'f', u'Kh3tIZR1|Khs0tRsh|K3geBqb_|KBrVNcDX|Jg2uDy8M|529816a3-ee43-4423-961f-8aedaf25d58c')), (u'331d8410d4924e72b0f0585e888c85ce', (u'f', u'1F37807A-CBEA-4B78-85D7-5A97B37B539E'))]
[(u'28b195c271f14a329235c262e7baecbf', (u'm', u'50c41480-a94e-4afa-a732-b6ed7a057239'))]
[(u'c65ac2064bc14116a363125392dcc6f7', (u'f', u'77e4b9b3-83b4-4553-b274-7a16f553cf05')), (u'171f92200d634d62bdc6685bdb7a94e3', (u'f', u'bdf53cb6-695d-4dde-b0c1-d1a34ebea6f7|a09e4074-c22e-48a1-9976-ee2151b5888c|K1Umlb5M|639B02B4-24AD-4069-99A2-C68E8C8F7F06|KjE3wXIr')), (u'_guid_wQZIzeFxciX9CIHUPeWOF2euOIC0jiOsXVXN98_zCh8=', (u'f', u'F0992237-2598-4B13-AA8A-C37D436B901C|C80D1A89-DD84-4734-838F-128F99EBDD20|KthpuVu0')), (u'_guid_ufOcKO48drwr50yJN26NriX5MLYONwmALxWcmly7oqQ=', (u'f', u'KlY10YxX|KyCVx_km'))]

我的目标是为这个列表中的每个rdd添加一种新的“列”。该行将包含每个 rdd 的唯一索引。我的代码:

for i, rdd in enumerate(repartionned_rdd):

    new_rdd = rdd.map(lambda x : x + (float(i), ))

    print new_rdd.collect()

这给出:

[(u'_guid_NCw7SuFnCh_mFW3SI3qTvBCbqXKD4mtsdJvWE7HNgNg=', 
 (u'f', u'KSJakOd2|KtC9ZF9h'), 0.0)]

[(u'_guid_OCs2au-sKnxzPE0uRPDP4hg1vvhgpzRAAYjNWRQpKbw=', 
 (u'f', u'KxrylzuA|KpSXJwH2'), 1.0), 
 (u'_guid_txH15ULaeUDBC4Z_NlEOj2xoYBFa-08imqIBLfYsKps=', 
 (u'f', u'bda54c71-cd1e-4eb7-856c-ba2e6def30c8|6e189e07-807e-41a2-a60a-b07d894a2905'), 1.0),
 (u'_guid_ehCT6NyD9l3q3NV9ZroaWVEo3bnDt4tvbU_fMBrEn1g=', 
 (u'm', u'537D69B4-743A-45B9-BED1-A25AA5926F13|2bb3e466-edc5-4302-b102-3bddb1f8c490|aa4760de-104c-4dc3-94c3-336427f89723'), 1.0), 
 (u'_guid_9F4Ph5GztLN9IlWNgZWKPMCcT4N3Je6-93iM_130F-c=', 
 (u'f', u'KOQqBzhU|KrDt5GC4'), 1.0), 
 (u'_guid_nPlE_f-zoOHNYiXJSGXWoVryc1U4Bnfxkow3P0mDUFY=', 
 (u'f', u'Kh3tIZR1|Khs0tRsh|K3geBqb_|KBrVNcDX|Jg2uDy8M|529816a3-ee43-4423-961f-8aedaf25d58c'), 1.0), 
 (u'331d8410d4924e72b0f0585e888c85ce', 
 (u'f', u'1F37807A-CBEA-4B78-85D7-5A97B37B539E'), 1.0)]

[(u'28b195c271f14a329235c262e7baecbf', 
 (u'm', u'50c41480-a94e-4afa-a732-b6ed7a057239'), 2.0)]

[(u'c65ac2064bc14116a363125392dcc6f7', 
 (u'f', u'77e4b9b3-83b4-4553-b274-7a16f553cf05'), 3.0), 
 (u'171f92200d634d62bdc6685bdb7a94e3', 
 (u'f', u'bdf53cb6-695d-4dde-b0c1-d1a34ebea6f7|a09e4074-c22e-48a1-9976-ee2151b5888c|K1Umlb5M|639B02B4-24AD-4069-99A2-C68E8C8F7F06|KjE3wXIr'), 3.0),
 (u'_guid_wQZIzeFxciX9CIHUPeWOF2euOIC0jiOsXVXN98_zCh8=', 
 (u'f', u'F0992237-2598-4B13-AA8A-C37D436B901C|C80D1A89-DD84-4734-838F-128F99EBDD20|KthpuVu0'), 3.0),
 (u'_guid_ufOcKO48drwr50yJN26NriX5MLYONwmALxWcmly7oqQ=', 
 (u'f', u'KlY10YxX|KyCVx_km'), 3.0)]

因此，我的 new_rdd 中的每一行都包含一个新列，具体是 rdd 的索引(如代码中所述!)

我现在的目标只是将所有这些新的 rdd 放入一个唯一的 rdd 中。我试过这个:

all_rdds_list =[]

for i, rdd in enumerate(repartionned_rdd):

    new_rdd = rdd.map(lambda x : x + (float(i), ))

    all_rdds_list.append(new_rdd)

但是当我尝试显示我的 rdds 时，我得到了这个:

for x in all_rdds_list:

    print x.collect()

结果:

[(u'_guid_NCw7SuFnCh_mFW3SI3qTvBCbqXKD4mtsdJvWE7HNgNg=', 
 (u'f', u'KSJakOd2|KtC9ZF9h'), 3.0)]

[(u'_guid_OCs2au-sKnxzPE0uRPDP4hg1vvhgpzRAAYjNWRQpKbw=', 
 (u'f', u'KxrylzuA|KpSXJwH2'), 3.0), 
 (u'_guid_txH15ULaeUDBC4Z_NlEOj2xoYBFa-08imqIBLfYsKps=', 
 (u'f', u'bda54c71-cd1e-4eb7-856c-ba2e6def30c8|6e189e07-807e-41a2-a60a-b07d894a2905'), 3.0),
 (u'_guid_ehCT6NyD9l3q3NV9ZroaWVEo3bnDt4tvbU_fMBrEn1g=', 
 (u'm', u'537D69B4-743A-45B9-BED1-A25AA5926F13|2bb3e466-edc5-4302-b102-3bddb1f8c490|aa4760de-104c-4dc3-94c3-336427f89723'), 3.0), 
 (u'_guid_9F4Ph5GztLN9IlWNgZWKPMCcT4N3Je6-93iM_130F-c=', 
 (u'f', u'KOQqBzhU|KrDt5GC4'), 3.0), 
 (u'_guid_nPlE_f-zoOHNYiXJSGXWoVryc1U4Bnfxkow3P0mDUFY=', 
 (u'f', u'Kh3tIZR1|Khs0tRsh|K3geBqb_|KBrVNcDX|Jg2uDy8M|529816a3-ee43-4423-961f-8aedaf25d58c'), 3.0), 
 (u'331d8410d4924e72b0f0585e888c85ce', 
 (u'f', u'1F37807A-CBEA-4B78-85D7-5A97B37B539E'), 3.0)]

[(u'28b195c271f14a329235c262e7baecbf', 
 (u'm', u'50c41480-a94e-4afa-a732-b6ed7a057239'), 3.0)]

[(u'c65ac2064bc14116a363125392dcc6f7', 
 (u'f', u'77e4b9b3-83b4-4553-b274-7a16f553cf05'), 3.0), 
 (u'171f92200d634d62bdc6685bdb7a94e3', 
 (u'f', u'bdf53cb6-695d-4dde-b0c1-d1a34ebea6f7|a09e4074-c22e-48a1-9976-ee2151b5888c|K1Umlb5M|639B02B4-24AD-4069-99A2-C68E8C8F7F06|KjE3wXIr'), 3.0),
 (u'_guid_wQZIzeFxciX9CIHUPeWOF2euOIC0jiOsXVXN98_zCh8=', 
 (u'f', u'F0992237-2598-4B13-AA8A-C37D436B901C|C80D1A89-DD84-4734-838F-128F99EBDD20|KthpuVu0'), 3.0),
 (u'_guid_ufOcKO48drwr50yJN26NriX5MLYONwmALxWcmly7oqQ=', 
 (u'f', u'KlY10YxX|KyCVx_km'), 3.0)]

帮忙？谢谢!

Answer 1

您的方法有两个问题。首先，您使用一个变量，该变量在评估方法分配之前会发生变化。 map 调用是转换，因此仅在您应用操作(例如 collect)时执行。这就是为什么当您在枚举循环内进行收集时，您会看到正确的附加列，但在后面的示例中，它为每个映射选择 i 的最后一个值。

第二个问题是，如果您尝试将 rdd 合并在一起，则应该使用 union 函数而不是 rdd 列表。如果您确实想要一个 rdd 列表，那么您可以将下面的并集替换为之前的列表附加。

full_rdd = None
for i, rdd in enumerate(repartionned_rdd):
    new_rdd = rdd.map(lambda x : x + (float(i),))
    if full_rdd is None:
        full_rdd = new_rdd
    else:
        full_rdd = sc.union([full_rdd, new_rdd])
    # This will force the lazy evaluation to execute now before `i` changes
    full_rdd.count()