操作 SparkSQL （spark版本2.3） JOIN使用临时表版本代码,Python代码库,德仔网

代码语言
代码分类
【Python】操作 SparkSQL （spark版本2.3） JOIN使用临时表版本
作者:黎昕 / 发布于2018/4/19/ 1849
# coding: utf-8
"""
使用SPAKR 对资源的操作类
"""
import time
from pyspark.sql import SparkSession
from pyspark.sql.functions import broadcast
from ccnu_resource.settings import MYSQL_CONF


class ResourceError(Exception):
    pass


class SparkBase(object):
    """
    SPARK 基类
    """

    def init_jdbc(self):
        raise NotImplementedError


class SparkSql(SparkBase):
    SPARK_MASTER = 'local[4]'
    APP_NAME = 'spark-t1'

    def __init__(self):
        self.spark = None
        self.jdbc = None

        self.init_spark()
        self.init_jdbc()

    def init_spark(self):
        """
        初始化spark
        :return:
        """
        self.spark = SparkSession.builder \
            .master(self.SPARK_MASTER) \
            .appName(self.APP_NAME) \
            .getOrCreate()

    def init_jdbc(self):
        """
        初始化jdbc
        :return:
        """
        jdbc_url = 'jdbc:mysql://{0}:{1}/{2}'.format(
            MYSQL_CONF['host'],
            MYSQL_CONF['port'],
            MYSQL_CONF['db']
        )

        self.jdbc = self.spark.read \
            .format("jdbc") \
            .option("url", jdbc_url) \
            .option("user", MYSQL_CONF['user']) \
            .option("password", MYSQL_CONF['password'])

    def load_table_df(self, table_name):
        """
        读取数据库表的DATAFRAME
        :param table_name:
        :return: DATAFRAME
        """
        jdbc_df = self.jdbc.option('dbtable', table_name).load()
        return jdbc_df

    def __del__(self):
        """
        对象任务完成后关闭链接
        :return:
        """
        print('关闭spark链接')
        self.spark.stop()


class SparkResource(object):
    def __init__(self, spark_sql=None):
        if not isinstance(spark_sql, SparkSql):
            raise ResourceError('缺少SparkSql实例')

        self.spark_sql = spark_sql

    def get_question_diff_distri(self, faculty=None, subject=None):
        """
        得到某学科和学段下面试题的困难度分布
        :param faculty:
        :param subject:
        :return:
        """
        if not faculty or not subject:
            raise ResourceError('缺少faculty或者subject')

        filter_str = "faculty = {0} and subject = {1}".format(
            faculty, subject
        )

        # 读取表的dataframe
        question_df = self.spark_sql.load_table_df('question')

        # 统计排序
        res_df = question_df.filter(filter_str).groupBy(
            "diff"
        ).count()

        return res_df

    def get_question_freq_top_n(self, n=20, faculty=None, subject=None):
        """
        得到某学科和学段下面的试卷里面的试题使用频繁度
        :param n:
        :param faculty:
        :param subject:
        :return:
        """
        if not faculty or not subject:
            raise ResourceError('缺少faculty或者subject')
        filter_str = "faculty = {0} and subject = {1}".format(
            faculty, subject
        )

        # 读取表的dataframe
        sub_q_df = self.spark_sql.load_table_df('paper_subtype_question')
        question_df = self.spark_sql.load_table_df('question')

        # 生成临时表
        sub_q_df.createOrReplaceTempView('tmp_paper_subtype_question')
        question_df.createOrReplaceTempView('tmp_question')

        # 检索临时表
        df = self.spark_sql.spark.sql(
            'SELECT question_id, faculty, subject '
            'from tmp_paper_subtype_question tpsq '
            'LEFT JOIN tmp_question tq '
            'ON tpsq.question_id = tq.qid '
            'WHERE tq.structure_string IS NOT NULL '
        )

        # 统计排序
        res_df = df.filter(filter_str).groupBy(
            "question_id"
        ).count().sort('count', ascending=False).limit(n)

        return res_df

    def get_knowledge_top_n(self, n=20, faculty=None, subject=None):
        """
        获得某学科和学段下面的试题获得所绑定知识点的TOP数
        :param n:
        :param faculty:
        :param subject:
        :return:
        """
        if not faculty or not subject:
            raise ResourceError('缺少faculty或者subject')

        filter_str = "faculty = {0} and subject = {1}".format(
            faculty, subject
        )

        # 读取表的dataframe
        cog_map_df = self.spark_sql.load_table_df('question_cognition_map')
        question_df = self.spark_sql.load_table_df('question')

        # 生成临时表
        cog_map_df.createOrReplaceTempView('tmp_question_cognition_map')
        question_df.createOrReplaceTempView('tmp_question')

        # 检索临时表
        df = self.spark_sql.spark.sql(
            'SELECT cognition_map_num, faculty, subject '
            'from tmp_question_cognition_map tqcm '
            'LEFT JOIN tmp_question tq '
            'ON tqcm.question_id = tq.qid'
        )

        # 统计排序
        res_df = df.filter(filter_str).groupBy(
            "cognition_map_num"
        ).count().sort('count', ascending=False).limit(n)

        return res_df

    def get_paper_info(self):
        df = self.spark_sql.load_table('paper')
        df.createOrReplaceTempView("tmp_paper")
        sqlDF = self.spark_sql.spark.sql("SELECT id FROM tmp_paper")
        sqlDF.show()

start_time = time.time()
sr = SparkResource(spark_sql=SparkSql())
knowledge_df = sr.get_knowledge_top_n(faculty=3, subject=1)
print(knowledge_df.toJSON().collect())

diff_df = sr.get_question_diff_distri(faculty=3, subject=1)
print(diff_df.toJSON().collect())

question_freq = sr.get_question_freq_top_n(faculty=3, subject=1)
print(question_freq.toJSON().collect())
print(time.time() - start_time)
试试其它关键字
同语言下
可能有用的
黎昕贡献的其它代码(12)