Python Spark submit引发GC内存不足异常

我是新手。我们有下面的配置单元查询，在它上面，我们通过使用spark和python执行pivot操作

下面pyspark脚本执行一些透视操作并写入配置单元表。配置单元查询返回1.4亿行

方法1

from pyspark import SparkContext
from pyspark import HiveContext
from pyspark.sql import functions as F
sc = SparkContext()
hc = HiveContext(sc)
tbl = hc.sql("""
    Select Rating.BranchID
    , Rating.Vehicle
    , Rating.PersonalAutoCov
    , Rating.PersonalVehicleCov 
    , Rating.EffectiveDate
    , Rating.ExpirationDate
    , attr.name as RatingAttributeName
    , Cast(Rating.OutputValue as Int) OutputValue
    , Rating.InputValue
    From db.dbo_pcx_paratingdata_piext_master rating
        Inner Join db.dbo_pctl_ratingattrname_piext_master attr 
            on rating.RatingAttribute = attr.id 
            and attr.CurrentRecordIndicator = 'Y'
    Where 
        rating.CurrentRecordIndicator = 'Y'
    """)
tbl.cache()
pvttbl1 = tbl.groupby("BranchId","Vehicle","PersonalAutoCov","PersonalVehicleCov","EffectiveDate","ExpirationDate")\
    .pivot("RatingAttributeName")\
    .agg({"InputValue":"max", "OutputValue":"sum"})

pvttbl1.createOrReplaceTempView("paRatingAttributes") 
hc.sql("Create table dev_pekindataaccesslayer.createcount as select * from paRatingAttributes") 

当我使用spark submit命令运行上面的脚本时，我的结果是

java.lang.OutOfMemoryError:java堆空间

有时候

java.lang.OutOfMemoryError:超出GC开销限制

我使用的spark提交命令

spark-submit spark_ex2.py --master yarn-cluster --num-executors 15 --executor-cores 50 --executor-memory 100g --driver-memory 100g, --conf `"spark.sql.shuffle.partitions=1000", --conf "spark.memory.offHeap.enabled=true", --conf "spark.memory.offHeap.size=100g",--conf "spark.network.timeout =1200", --conf "spark.executor.heartbeatInterval=1201"`

详细日志：

INFO MemoryStore: Memory use = 1480.9 KB (blocks) + 364.8 MB (scratch space shared across 40 tasks(s)) = 366.2 MB. 
Storage limit = 366.3 MB.
WARN BlockManager: Persisting block rdd_11_22 to disk instead.
WARN BlockManager: Putting block rdd_11_0 failed due to an exception
WARN BlockManager: Block rdd_11_0 could not be removed as it was not found on disk or in memory
ERROR Executor: Exception in task 0.0 in stage 1.0 (TID 10)
java.lang.OutOfMemoryError: Java heap space
        at java.nio.HeapByteBuffer.<init>(HeapByteBuffer.java:57)
        at java.nio.ByteBuffer.allocate(ByteBuffer.java:335)

但是上面的脚本涉及到中间表的创建，这是一个附加步骤。 在方法2中，当我使用相同的spark submit命令保持limit关键字时，命令工作正常

我的方法1出了什么问题？我如何才能使它起作用

---

注意：我已经遵循并尝试了所有建议的conf参数，但仍然没有成功。

以防我最后的评论不清楚。我相信您可能有一些类似于

java-Xmx100G-Xmx2G whateverthesparkmainclassiscaled

，这将导致

-Xmx2G

取代

-Xmx100G

，这将解释内存错误的原因。因此，查看确切的java命令行非常重要

ps-eo args | grep java

可用于查找完整命令line@Tharunkumar雷迪。。您能否在pivot之前向我们展示一些“tbl”的示例数据，以及对其使用pivot操作后的期望值。你能解释一下那部分吗。Pivot是一种昂贵的操作，也许我们可以尝试其他方法来做同样的事情。只是一个想法。@vikrantrana，具有以下配置”--conf spark.thread.appMasterEnv.spark\u HOME=/dev/null“对我来说一切都很好。你知道这个参数的作用吗？@Tharunkumar Reddy。。不知道。我会检查一下，但我在pivot上发现了一些有趣的东西。可能有用。参数

--conf spark.warn.appMasterEnv.spark\u HOME=/dev/null

将环境变量

spark\u HOME

设置为/dev/null。换句话说，它将取消设置任何现有的

SPARK\u HOME

。我不知道为什么这能解决问题。如果我不得不猜测的话，我会说要么你现有的SPARK_HOME中有一个坏的配置文件，要么SPARK_HOME中有一些JAR现在没有加载，这会释放内存。

    from pyspark import SparkContext
    from pyspark import HiveContext
    from pyspark.sql import SQLContext
    from pyspark.sql import functions as F
    sc = SparkContext()
    hc = HiveContext(sc)
    sqlContext = SQLContext(sc)
    tbl = hc.sql("""
       Select Rating.BranchID
          , Rating.Vehicle
         , Rating.PersonalAutoCov
            , Rating.PersonalVehicleCov
           , Rating.EffectiveDate
          , Rating.ExpirationDate
         , attr.name as RatingAttributeName
         , Cast(Rating.OutputValue as Int) OutputValue
         , Rating.InputValue
        From db.dbo_pcx_paratingdata_piext_master rating
           Inner Join db.dbo_pctl_ratingattrname_piext_master attr
              on rating.RatingAttribute = attr.id
             and attr.CurrentRecordIndicator = 'Y'
        Where
           rating.CurrentRecordIndicator = 'Y'
       """)
    tbl.createOrReplaceTempView("Ptable")
    r=sqlContext.sql("select count(1) from Ptable")
    m=r.collect()[0][0]
    hc.sql("drop table if exists db.Ptable")
    hc.sql("Create table db.Ptable as select * from Ptable")
    tb2 = hc.sql("select * from db.Ptable limit "+str(m))
    pvttbl1 = tb2.groupby("BranchId","Vehicle","PersonalAutoCov","PersonalVehicleCov","EffectiveDate","ExpirationDate")\
        .pivot("RatingAttributeName")\
        .agg({"InputValue":"max", "OutputValue":"sum"})

    pvttbl1.createOrReplaceTempView("paRatingAttributes")
    hc.sql("drop table if exists db.createcount")
    hc.sql("Create table db.createcount STORED AS ORC as select * from paRatingAttributes")