Python ETL—使用cx_Oracle批量或迭代地将大型数据集加载到Oracle数据库中

使用Python将包含10MM记录的数据集加载到Oracle数据库表中。 创建的数据帧没有问题。从cx_或LCE加载数据帧记录计数时出现过大错误

试图在数据帧上循环并通过一次插入100k记录批量加载10MM记录

下面的代码shwon可以工作，但只适用于适合分配内存的小数据集。我需要一个适用于批量和大型数据集的

已经尝试在行上迭代，但这需要很长时间。我们也尝试过加载更小的数据帧——这是可行的，但没有达到目的

还尝试使用Bindarray和数组大小来lad数据帧，但没有任何效果

import pandas as pd
import datetime
import sys
import re
from itertools import groupby, islice, takewhile
import cx_Oracle

format = '%y_%m_%d'

TODAY = datetime.date.today()
add = datetime.timedelta(days=1)
yesterday = datetime.date.today() - add
dd = datetime.date.strftime(TODAY,format)

# connection variables
connection = cx_Oracle.connect("user/Oracle_database_connect_info")
cur = connection.cursor()

# dataframe headers
columns = ['C1','C2','C3','C4']

# -- >> test and sample the file
csv_df = pd.read_csv(r'csvfile_location')

# add record_id for values
csv_df_idx = csv_df.index.values +1
csv_df.insert(0,column = 'RECORD_ID' , value=csv_df_idx)


### TABLE ALREADY CREATED IN DATABASE ###


for index, row in csv_df.iterrows():
    ### Insert and Iterate to inset records
    ### Convert to list for easy load into DB
    csv_df_dataset_lst = csv_df.values.tolist()
    insert_statement = """
    INSERT INTO TEST_LOAD
        ( RECORD_ID ,C1 ,C2 ,C3 ,C4)values (:1,:2,:3,:4,:5)    """

    # control number of records to bind for insert
    # cur.bindarraysize = 100000 # --->>> did not work
    # cur.arraysize = 100000 # --->>> did not work
    cur.executemany(insert_statement,csv_df_dataset_lst)
    connection.commit()
connection.close()

---

我明白了。诀窍是编写一个函数，根据要加载的批的大小将数据帧分块

下面是最终代码

import pandas as pd
import numpy as np
import datetime
import sys
import re
from itertools import groupby, islice, takewhile
import cx_Oracle

format = '%y_%m_%d'

TODAY = datetime.date.today()
add = datetime.timedelta(days=1)
yesterday = datetime.date.today() - add
dd = datetime.date.strftime(TODAY,format)

# connection variables
connection = cx_Oracle.connect("user/Oracle_database_connect_info")
cur = connection.cursor()

# dataframe headers
columns = ['C1','C2','C3','C4']

# -- >> test and sample the file
csv_df = pd.read_csv(r'csvfile_location')

# add record_id for values
csv_df_idx = csv_df.index.values +1
csv_df.insert(0,column = 'RECORD_ID' , value=csv_df_idx)


### TABLE ALREADY CREATED IN DATABASE ###


# set batch size ie record count
batch_size = 100000

# create chunker function to separate the dataframe into batches
# Note: last batch will contain smallest amout of records.
def chunker(seq,size):
    return(seq[pos:pos+size] for pos in range(0,len(seq),size))


insert_statement = """
    INSERT INTO TEST_LOAD
        ( RECORD_ID ,C1 ,C2 ,C3 ,C4)values (:1,:2,:3,:4,:5)    """

# Optional use cursor.prepare so Oracle DB avoids compiling the insert statement over and over
try:
    cur.prepare(insert_statement)
except cx_Oracle.DatabaseError as Exception:
    printf('Failed to prepare insert cursor')
    printException(Exception)
    exit(1)

for i in chunker(csv_df,batch_size):
    ### Insert and Iterate to inset records
    ### Convert to list for easy load into DB
    csv_df_dataset_lst = csv_df.values.tolist()

    cur.executemany(insert_statement,csv_df_dataset_lst)
    connection.commit()
    # record counter to monitor the loading.
    number_of_records_loaded = cur.execute("""SELECT COUNT(*), SYSDATE FROM TEST_LOAD GROUP BY SYSDATE""")
    record_out = cur.fetchall()
    for row in record_out:
        print(row)
connection.close()

---

使用Cursor.prepare（）和Cursor.executemany（）。您能分享一下在为大型数据集执行此代码时收到的错误消息吗？cx_Oracle.DatabaseError:DPI-1015:数组大小10000000太大了。我想明白了。在python中生成一个函数来分块数据帧，然后在该分块上迭代。由于cx_Oracle使用语句缓存，使用explict cursor.prepare可能没有多大帮助。运行查询来计算记录只会增加开销-为什么需要它？为什么不使用executemany（）的batchErrors模式？如果您知道数据大小，可以通过预定义内存来提高效率。您可能想看一下cx_Oracle文档—并不是说有任何一种“正确”的方法可以做到这一点。