java - 如何在没有网络访问的情况下将 parckage(例如 mmlspark)安装到 CDH 集群？

Question

因为很难连接maven.org在中国，我不能不安装 mmlspark经过

pyspark --packages com.microsoft.ml.spark:mmlspark_2.11:1.0.0-rc1 --repositories=https://mmlspark.azureedge.net/maven

会得到

:::: ERRORS
        Server access error at url https://repo1.maven.org/maven2/com/microsoft/ml/lightgbm/lightgbmlib/2.3.100/lightgbmlib-2.3.100.pom (java.net.ConnectException: Connection timed out (Connection timed out))

        Server access error at url https://repo1.maven.org/maven2/com/microsoft/ml/lightgbm/lightgbmlib/2.3.100/lightgbmlib-2.3.100.jar (java.net.ConnectException: Connection timed out (Connection timed out))


:: USE VERBOSE OR DEBUG MESSAGE LEVEL FOR MORE DETAILS
Exception in thread "main" java.lang.RuntimeException: [unresolved dependency: com.microsoft.ml.lightgbm#lightgbmlib;2.3.100: not found, download failed: com.microsoft.ml.spark#mmlspark_2.11;1.0.0-rc1!mmlspark_2.11.jar, download failed: org.scalatest#scalatest_2.11;3.0.5!scalatest_2.11.jar(bundle), download failed: com.microsoft.cntk#cntk;2.4!cntk.jar, download failed: org.openpnp#opencv;3.2.0-1!opencv.jar(bundle)]
        at org.apache.spark.deploy.SparkSubmitUtils$.resolveMavenCoordinates(SparkSubmit.scala:1308)
        at org.apache.spark.deploy.DependencyUtils$.resolveMavenDependencies(DependencyUtils.scala:54)
        at org.apache.spark.deploy.SparkSubmit.prepareSubmitEnvironment(SparkSubmit.scala:315)
        at org.apache.spark.deploy.SparkSubmit.submit(SparkSubmit.scala:143)
        at org.apache.spark.deploy.SparkSubmit.doSubmit(SparkSubmit.scala:86)
        at org.apache.spark.deploy.SparkSubmit$$anon$2.doSubmit(SparkSubmit.scala:926)
        at org.apache.spark.deploy.SparkSubmit$.main(SparkSubmit.scala:935)
        at org.apache.spark.deploy.SparkSubmit.main(SparkSubmit.scala)
[TerminalIPythonApp] WARNING | Unknown error in handling PYTHONSTARTUP file /opt/cloudera/parcels/CDH-6.2.0-1.cdh6.2.0.p0.967373/lib/spark/python/pyspark/shell.py:

尝试手动安装
我有一个亚马逊 ec2 实例，它可以访问 maven.org ，我下载了所有包并复制到本地CDH集群，路径 /opt/cloudera/parcels/CDH/lib/spark/jars/mmlspark_jars/ ,
并设置配置:
第一的
spark-defaults.conf:

spark.driver.extraClassPath=/opt/cloudera/parcels/CDH/lib/spark/jars/mmlspark_jars/*
spark.executor.extraClassPath=/opt/cloudera/parcels/CDH/lib/spark/jars/mmlspark_jars/*

第二:
spark-env.sh:

export SPARK_CLASSPATH=/opt/cloudera/parcels/CDH/lib/spark/jars/mmlspark_jars/*:$SPARK_CLASSPATH

可以看到jar被加载

但是 import mmlspark仍然 ModuleNotFoundError: No module named 'mmlspark' 通过一些努力
我发现:提取 mmlspark.jar，将 mmlspark 压缩到文件夹内并放入 hdfs( hdfs://test/mmlspark.zip )，加载此 .zip在 pyfiles (--py-files hdfs://test/mmlspark.zip ) 中，可以使 import mmlspark 成功。
我使用 jar 依赖项和 mmlspark.zip 启动一个 pyspark shell:

pyspark --jars "/user/spark/mmlspark_jars/com.github.vowpalwabbit_vw-jni-8.7.0.3.jar,/user/spark/mmlspark_jars/com.jcraft_jsch-0.1.54.jar,/user/spark/mmlspark_jars/com.microsoft.cntk_cntk-2.4.jar,/user/spark/mmlspark_jars/com.microsoft.ml.lightgbm_lightgbmlib-2.3.100.jar,/user/spark/mmlspark_jars/com.microsoft.ml.spark_mmlspark_2.11-1.0.0-rc1.jar,/user/spark/mmlspark_jars/commons-codec_commons-codec-1.10.jar,/user/spark/mmlspark_jars/commons-logging_commons-logging-1.2.jar,/user/spark/mmlspark_jars/io.spray_spray-json_2.11-1.3.2.jar,/user/spark/mmlspark_jars/org.apache.httpcomponents_httpclient-4.5.6.jar,/user/spark/mmlspark_jars/org.apache.httpcomponents_httpcore-4.4.10.jar,/user/spark/mmlspark_jars/org.openpnp_opencv-3.2.0-1.jar,/user/spark/mmlspark_jars/org.scala-lang.modules_scala-xml_2.11-1.0.6.jar,/user/spark/mmlspark_jars/org.scala-lang_scala-reflect-2.11.12.jar,/user/spark/mmlspark_jars/org.scalactic_scalactic_2.11-3.0.5.jar,/user/spark/mmlspark_jars/org.scalatest_scalatest_2.11-3.0.5.jar" --py-files hdfs://test/mmlspark.zip

测试代码

from sklearn import datasets
iris = datasets.load_iris()
X = iris.data
Y = iris.target
df = np.column_stack([X,Y])
df = pd.DataFrame(df)
df.columns = ['f1', 'f2', 'f3', 'f4', 'label']
feature_cols = ['f1', 'f2', 'f3', 'f4']
df = spark.createDataFrame(df)

from pyspark.ml.feature import VectorAssembler
vec_assembler = VectorAssembler(inputCols=feature_cols, outputCol='features')
df1 = vec_assembler.transform(df)

from mmlspark.lightgbm import LightGBMRegressor
model = LightGBMRegressor(objective='quantile',
                          alpha=0.2,
                          learningRate=0.3,
                          numLeaves=31,
                         featuresCol='features',
                         labelCol='label').fit(df1)

错误

Py---------------------------------------------------------------------------
Py4JJavaError                             Traceback (most recent call last)
<ipython-input-55-fe341b86ea18> in <module>
     18                           numLeaves=31,
     19                          featuresCol='features',
---> 20                          labelCol='label').fit(df1)

/opt/cloudera/parcels/CDH/lib/spark/python/pyspark/ml/base.py in fit(self, dataset, params)
    130                 return self.co(params)._fit(dataset)
    131             else:
--> 132                 return self._fit(dataset)
    133         else:
    134             raise ValueError("Params must be either a param map or a list/tuple of param maps, "

/opt/cloudera/parcels/CDH/lib/spark/python/pyspark/ml/wrapper.py in _fit(self, dataset)
    293 
    294     def _fit(self, dataset):
--> 295         java_model = self._fit_java(dataset)
    296         model = self._create_model(java_model)
    297         return self._copyValues(model)

/opt/cloudera/parcels/CDH/lib/spark/python/pyspark/ml/wrapper.py in _fit_java(self, dataset)
    289         :return: fitted Java model
    290         """
--> 291         self._transfer_params_to_java()
    292         return self._java_obj.fit(dataset._jdf)
    293 

/opt/cloudera/parcels/CDH/lib/spark/python/pyspark/ml/wrapper.py in _transfer_params_to_java(self)
    125                 self._java_obj.set(pair)
    126             if self.hasDefault(param):
--> 127                 pair = self._make_java_param_pair(param, self._defaultParamMap[param])
    128                 pair_defaults.append(pair)
    129         if len(pair_defaults) > 0:

/opt/cloudera/parcels/CDH/lib/spark/python/pyspark/ml/wrapper.py in _make_java_param_pair(self, param, value)
    111         sc = SparkContext._active_spark_context
    112         param = self._resolveParam(param)
--> 113         java_param = self._java_obj.getParam(param.name)
    114         java_value = _py2java(sc, value)
    115         return java_param.w(java_value)

/opt/cloudera/parcels/CDH/lib/spark/python/lib/py4j-0.10.7-src.zip/py4j/java_gateway.py in __call__(self, *args)
   1255         answer = self.gateway_client.send_command(command)
   1256         return_value = get_return_value(
-> 1257             answer, self.gateway_client, self.target_id, self.name)
   1258 
   1259         for temp_arg in temp_args:

/opt/cloudera/parcels/CDH/lib/spark/python/pyspark/sql/utils.py in deco(*a, **kw)
     61     def deco(*a, **kw):
     62         try:
---> 63             return f(*a, **kw)
     64         except py4j.protocol.4JJavaError as e:
     65             s = e.java_exception.toString()

/opt/cloudera/parcels/CDH/lib/spark/python/lib/py4j-0.10.7-src.zip/py4j/protocol.py in get_return_value(answer, gateway_client, target_id, name)
    326                 raise Py4JJavaError(
    327                     "An error occurred while calling {0}{1}{2}.\n".
--> 328                     format(target_id, ".", name), value)
    329             else:
    330                 raise Py4JError(

Py4JJavaError: An error occurred while calling o1298.getParam.
: java.util.NoSuchElementException: Param metric does not exist.
    at org.apache.spark.ml.param.Params$$anonfun$getParam$2.apply(params.scala:729)
    at org.apache.spark.ml.param.Params$$anonfun$getParam$2.apply(params.scala:729)
    at scala.Option.getOrElse(Option.scala:121)
    at org.apache.spark.ml.param.Params$class.getParam(params.scala:728)
    at org.apache.spark.ml.PipelineStage.getParam(Pipeline.scala:42)
    at sun.reflect.GeneratedMethodAccessor15.invoke(Unknown Source)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:498)
    at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:244)
    at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357)
    at py4j.Gateway.invoke(Gateway.java:282)
    at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:132)
    at py4j.commands.CallCommand.execute(CallCommand.java:79)
    at py4j.GatewayConnection.run(GatewayConnection.java:238)
    at java.lang.Thread.run(Thread.java:748)
py

在这里，我认为这个错误是因为 mmlspark python port 无法加载 jar ，从而导致 Py4JJavaError。但我不知道，我已经做了我知道的一切。

Answer 1

最后我搞定了。
关键是pass .jar至 pyFiles ，这让我很惊讶python可以阅读.jar .
重击:

pyspark \
--master yarn \
--conf spark.submit.pyFiles=hdfs://pupuxdc/user/spark/mmlspark_jars/…… .jar \
--conf spark.yarn.dist.jars=hdfs://pupuxdc/user/spark/mmlspark_jars/…… .jar

pyspark 代码:

spark_builder = (
    SparkSession
    .builder
    .config("spark.port.maxRetries", 100)
    .appName(app_name))
    
spark = spark_builder.getOrCreate()
jar_files = [...]
for i in jar_files:
    spark.sparkContext.addPyFile(i)

通知， .config('spark.submit.pyFiles=hdfs://pupuxdc/user/spark/mmlspark_jars/…… .jar')不生效。