spark-submit 如何在集群模式下传递 --driver-class-path？

Posted 2023-04-15

【中文标题】spark-submit 如何在集群模式下传递 --driver-class-path？

【英文标题】：spark-submit how to pass --driver-class-path in cluster mode?

【发布时间】：2020-08-20 16:05:54

【问题描述】：

好吧，如果使用带有驱动程序类路径的 pyspark shell，我可以使用 docker 映像访问 hive 资源：

$ pyspark --driver-class-path /etc/spark2/conf:/etc/hive/conf
Python 3.7.4 (default, Aug 13 2019, 20:35:49)
Using Python version 3.7.4 (default, Aug 13 2019 20:35:49)
SparkSession available as 'spark'.
>>> from pyspark.sql import SparkSession
>>>
>>> #declaration
... appName = "test_hive_minimal"
>>> master = "yarn"
>>>
... sc = SparkSession.builder \
...     .appName(appName) \
...     .master(master) \
...     .enableHiveSupport() \
...     .config("spark.hadoop.hive.enforce.bucketing", "True") \
...     .config("spark.hadoop.hive.support.quoted.identifiers", "none") \
...     .config("hive.exec.dynamic.partition", "True") \
...     .config("hive.exec.dynamic.partition.mode", "nonstrict") \
...     .getOrCreate()
>>> sql = "show tables in user_tables"
>>> df_new = sc.sql(sql)
20/08/20 15:08:50 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
>>> df_new.show()
+-----------+--------------------+-----------+
|   database|           tableName|isTemporary|
+-----------+--------------------+-----------+
|user_tables|              dummyt|      false|
|user_tables|abcdefg...dummytable|      false|

但如果通过 spark-submit 使用相同的脚本，则会出现以下错误：

spark-submit --master local --deploy-mode cluster --name test_hive --executor-memory 2g --num-executors 1 -- test_hive_minimal.py --verbose

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/conda/lib/python3.7/site-packages/pyspark/sql/session.py", line 767, in sql
    return DataFrame(self._jsparkSession.sql(sqlQuery), self._wrapped)
  File "/opt/conda/lib/python3.7/site-packages/pyspark/python/lib/py4j-0.10.7-src.zip/py4j/java_gateway.py", line 1257, in __call__
  File "/opt/conda/lib/python3.7/site-packages/pyspark/sql/utils.py", line 71, in deco
    raise AnalysisException(s.split(': ', 1)[1], stackTrace)
pyspark.sql.utils.AnalysisException: "Database 'user_tables' not found;"

test_hive_minimal.py 是一个检查 hive db 的简单脚本：

from pyspark.sql import SparkSession

appName = "test_hive_minimal"
master = "yarn"
# Creating Spark session
sc = SparkSession.builder \
    .appName(appName) \
    .master(master) \
    .enableHiveSupport() \
    .config("spark.hadoop.hive.enforce.bucketing", "True") \
    .config("spark.hadoop.hive.support.quoted.identifiers", "none") \
    .config("hive.exec.dynamic.partition", "True") \
    .config("hive.exec.dynamic.partition.mode", "nonstrict") \
    .getOrCreate()

sql = "show tables in user_tables"
df_new = sc.sql(sql)
df_new.show()
sc.stop()

我尝试了几种方法，传递 hive.metastore.uris、spark.sql.warehouse.dir 以及将 xml 文件作为 --files 传递。不知何故，我的执行者似乎无法访问配置。有人可以帮忙吗？

更新： 我成功地将 hive-site.xml 作为 --files 传递给集群模式下的 spark-submit，并且日志显示它不再为 Metastore 创建本地 derby.db。但是，现在面临另一个问题如下：

20/08/21 09:59:29 INFO state.StateStoreCoordinatorRef: Registered StateStoreCoordinator endpoint
20/08/21 09:59:31 INFO hive.HiveUtils: Initializing HiveMetastoreConnection version 1.2.1 using Spark classes.
20/08/21 09:59:31 INFO hive.metastore: Trying to connect to metastore with URI thrift://cluster01.cdh.com:9083
20/08/21 09:59:32 ERROR transport.TSaslTransport: SASL negotiation failure
javax.security.sasl.SaslException: GSS initiate failed [Caused by GSSException: No valid credentials provided (Mechanism level: Failed to find any Kerberos tgt)]
        at com.sun.security.sasl.gsskerb.GssKrb5Client.evaluateChallenge(GssKrb5Client.java:211)
        at org.apache.thrift.transport.TSaslClientTransport.handleSaslStartMessage(TSaslClientTransport.java:94)
        at org.apache.thrift.transport.TSaslTransport.open(TSaslTransport.java:271)
        at org.apache.thrift.transport.TSaslClientTransport.open(TSaslClientTransport.java:37)

似乎是 kerberos 问题，但我已经拥有有效的 kerberos 令牌，并且能够通过终端/也通过 docker 的 spark-shell 访问 hdfs。这里需要做什么？在集群上提交时，这不是由纱线自动配置的吗？

【参考方案1】：

我认为你应该在 spark-submit 命令中传递 keytab，这段代码是通过 SSH 运行的？

【讨论】：

keytab 已在 env 中设置。我尝试专门通过但没有帮助。就像所有其他调用一样，这是从 docker 终端内部提交的。

【参考方案2】：

更新： 在 docker 容器上共享 vol mount 并将 keytab/principal 与 hive-site.xml 一起传递以访问 Metastore 后问题已得到解决。

spark-submit --master yarn \
--deploy-mode cluster \
--jars /srv/python/ext_jars/terajdbc4.jar \
--files=/etc/hive/conf/hive-site.xml \
--keytab /home/alias/.kt/alias.keytab \ #this is mounted and kept in docker local path 
--principal [e-mail] \
--name td_to_hive_test \
--driver-cores 2 \
--driver-memory 2G \
--num-executors 44 \
--executor-cores 5 \
--executor-memory 12g \
td_to_hive_test.py