Pyspark 错误 self._sock.recv_into(b) socket.timeout：超时

Question

目标是使用 UDF 对行进行分类。我在 windows.

上使用 pyspark

使用过滤器等简单函数或操作似乎可行。

关于如何解决 timeout/socket 故障的任何指示都会有所帮助（请参阅下面的错误）。

数据中没有空值。

from pyspark.sql.functions import udf
from pyspark.sql.types import IntegerType,StringType

def BreakDown(arr_value):
    start_year = arr_value[0]
    start_month = arr_value[1]
    end_year = arr_value[2]
    end_month = arr_value[3]
    curr_year = arr_value[4]
    curr_month = arr_value[5]
    if   (curr_year == start_year) & (curr_month >= start_month) : return 1
    elif   (curr_year == end_year) & (curr_month <= end_month) : return 1
    elif   (curr_year > start_year) & (curr_year < end_year) : return 1
    else: return 0

    
udfBreakDown = udf(BreakDown, IntegerType())

temp = temp.withColumn('include', udfBreakDown(F.struct('start_year','start_month','end_year','end_month','curr_year','curr_month')))

PythonException: An exception was thrown from the Python worker. Please see the stack trace below. Traceback (most recent call last):
File "E:\spark\spark-3.0.1-bin-hadoop2.7\python\lib\pyspark.zip\pyspark\worker.py", line 585, in main File "E:\spark\spark-3.0.1-bin-hadoop2.7\python\lib\pyspark.zip\pyspark\serializers.py", line 593, in read_int length = stream.read(4) File "C:\ProgramData\Anaconda3\lib\socket.py", line 669, in readinto return self._sock.recv_into(b) socket.timeout: timed out

Answer 1

当您可以使用 Spark 内置函数时，请始终避免使用 UDF。您可以使用 when 函数重写您的逻辑，如下所示：

from pyspark.sql import functions as F

def get_include_col():
    c = F.when((F.col("curr_year") == F.col("start_year")) & (F.col("curr_month") >= F.col("start_month")), F.lit(1)) \
        .when((F.col("curr_year") == F.col("end_year")) & (F.col("curr_month") <= F.col("end_month")), F.lit(1)) \
        .when((F.col("curr_year") > F.col("start_year")) & (F.col("curr_year") < F.col("end_year")), F.lit(1)) \
        .otherwise(F.lit(0))
    return c


temp = temp.withColumn('include', get_include_col())

您还可以使用 functools.reduce 动态生成 when 表达式，而无需将它们全部粘贴。例如：

import functools
from pyspark.sql import functions as F

cases = [
    ("curr_year = start_year and curr_month >= start_month", 1),
    ("curr_year = end_year and curr_month <= end_month", 1),
    ("curr_year > start_year and curr_year < end_year", 1)
]

include_col = functools.reduce(
    lambda acc, x: acc.when(F.expr(x[0]), F.lit(x[1])),
    cases,
    F
).otherwise(F.lit(0))

temp = temp.withColumn('include', include_col)

Pyspark 错误 self._sock.recv_into(b) socket.timeout：超时

Pyspark with error self._sock.recv_into(b) socket.timeout: timed out

python

apache-spark

apache-spark-sql

pyspark

pyspark-dataframes