Oracle Client 安装
1、环境
日期:2019年8月1日
公司已经安装好Oracle服务端
Windows版本:Windows10专业版
系统类型:64位操作系统,基于x64的处理器
Python版本:Python 3.6.4 :: Anaconda, Inc.
2、下载网址
https://www.oracle.com/database/technologies/instant-client/downloads.html
3、解压至目录
解压后(这里放D盘)
4、配置环境变量
控制面板\系统和安全\系统 -> 高级系统设置 -> 环境变量
新建ORACLE_HOME,值为包解压的路径
编辑PATH,添加%ORACLE_HOME%
Navicat连接测试
cx_Oracle
安装命令
conda install cx_Oracle
基础代码
import cx_Oracle def execute(query): db = cx_Oracle.connect('用户名/密码@IP/ServiceName') cursor = db.cursor() cursor.execute(query) result = cursor.fetchall() cursor.close() db.close() return result def commit(sql): db = cx_Oracle.connect('用户名/密码@IP/ServiceName') cursor = db.cursor() cursor.execute(sql) db.commit() cursor.close() db.close()
封装成类
from cx_Oracle import Connection # conda install cx_Oracle from conf import CONN, Color class Oracle(Color): def __init__(self, conn=CONN): self.db = Connection(*conn, encoding='utf8') # 用户名 密码 IP/ServiceName self.cursor = self.db.cursor() def __del__(self): self.cursor.close() self.db.close() def commit(self, sql): try: self.cursor.execute(sql) self.db.commit() except Exception as e: self.red(e) def fetchall(self, query): self.cursor.execute(query) return self.cursor.fetchall() def fetchone(self, query, n=9999999): self.cursor.execute(query) for _ in range(n): one = self.cursor.fetchone() if one: yield one def fetchone_dt(self, query, n=9999999): self.cursor.execute(query) columns = [i[0] for i in self.cursor.description] length = len(columns) for _ in range(n): one = self.cursor.fetchone() # tuple yield {columns[i]: one[i] for i in range(length)} def read_clob(self, query): self.cursor.execute(query) one = self.cursor.fetchone() while one: try: yield one[0].read() except Exception as e: self.red(e) one = self.cursor.fetchone() def db2sheet(self, query, prefix): df = pd.read_sql_query(query, self.db) if 'url' in df.columns: df['url'] = "'" + df['url'] df.to_excel(prefix.replace('.xlsx', '')+'.xlsx', index=False) def db2sheets(self, queries, prefix): writer = pd.ExcelWriter(prefix.replace('.xlsx', '')+'.xlsx') for sheet_name, query in queries.items(): df = pd.read_sql_query(query, self.db) if 'url' in df.columns: df['url'] = "'" + df['url'] df.to_excel(writer, sheet_name=sheet_name, index=False) writer.save() def tb2sheet(self, table): sql = "SELECT * FROM " + table self.db2sheet(sql, table) def insert(self, dt, tb): for k, v in dt.items(): if isinstance(v, str): dt[k] = v.replace("'", '').strip() ls = [(k, v) for k, v in dt.items() if v is not None] sql = 'INSERT INTO %s (' % tb + ','.join(i[0] for i in ls) + \ ') VALUES (' + ','.join('%r' % i[1] for i in ls) + ')' self.commit(sql) def insert_clob(self, dt, tb, clob): for k, v in dt.items(): if isinstance(v, str): dt[k] = v.replace("'", '').strip() # 把超长文本保存在一个变量中 # declare = "DECLARE variate CLOB := '%s';\n" % dt[clob] join = lambda x: '||'.join("'%s'" % x[10922*i: 10922*(i+1)] for i in range(len(x)//10922+1)) # 32768//3 declare = "DECLARE variate CLOB := %s;\n" % join(dt[clob]) dt[clob] = 'variate' ls = [(k, v) for k, v in dt.items() if v is not None] sql = 'INSERT INTO %s (' % tb + ','.join(i[0] for i in ls) + ') VALUES (' +\ ','.join('%r' % i[1] for i in ls) + ');' sql = declare + 'BEGIN\n%s\nEND;' % sql.replace("'variate'", 'variate') self.commit(sql) def update(self, dt_update, dt_condition, table): sql = 'UPDATE %s SET ' % table + ','.join('%s=%r' % (k, v) for k, v in dt_update.items()) \ + ' WHERE ' + ' AND '.join('%s=%r' % (k, v) for k, v in dt_condition.items()) self.commit(sql) def truncate(self, tb): self.commit('truncate table ' + tb) db_read = Oracle() fetchall = db_read.fetchall fetchone = db_read.fetchone read_clob = db_read.read_clob if __name__ == '__main__': query = ''' '''.strip() for i in fetchone(query, 99): print(i)
conf
CONN = ('用户名', '密码', 'IP/ServiceName') conn = '用户名/密码@IP/ServiceName'
文本字符串查询
class INSTR(Oracle): """文本字符串查询""" def highlight_instr(self, table, field, keyword, clob=True): sql = "SELECT %s FROM %s WHERE INSTR(%s,'%s')>0" % (field, table, field, keyword) if clob: for i in self.read_clob(sql): self.highlight(i, keyword) else: for i, in self.fetchone(sql): self.highlight(i, keyword) def regexp_instr(self, table, field, pattern, regexp=True, clob=True): sql = "SELECT %s FROM %s WHERE INSTR(%s,'%s')>0" % (field, table, field, pattern) sql = sql.replace('INSTR', 'REGEXP_INSTR') if regexp else sql if clob: for i in self.read_clob(sql): yield i else: for i, in self.fetchone(sql): yield i
一个简单的建表示例
-- 建表 CREATE TABLE table_name ( serial_number NUMBER(10), collect_date DATE, url VARCHAR2(255), long_text CLOB, price NUMBER(10)-- 若需要精确到小数点2位,按分存储,/100还原到元 ); -- 给表添加备注 COMMENT ON TABLE table_name IS '中文表名'; -- 给表字段添加备注 COMMENT ON COLUMN table_name.serial_number IS '编号'; COMMENT ON COLUMN table_name.collect_date IS '日期'; COMMENT ON COLUMN table_name.url IS 'URL'; COMMENT ON COLUMN table_name.long_text IS '长文本'; COMMENT ON COLUMN table_name.price IS '价钱'; -- 插入 INSERT INTO table_name(collect_date) VALUES (DATE'2019-08-23'); INSERT INTO table_name(long_text) VALUES ('a'); INSERT INTO table_name(long_text) VALUES ('b'); -- 查询 SELECT * FROM table_name WHERE TO_CHAR(long_text) in ('a','b'); -- 查建表语句(表名大写) SELECT dbms_metadata.get_ddl('TABLE','TABLE_NAME') FROM dual; -- 删表 DROP TABLE table_name;
sqlalchemy
import os # 解决【UnicodeEncodeError: 'ascii' codec can't encode character】问题 os.environ['NLS_LANG'] = 'AMERICAN_AMERICA.AL32UTF8' # os.environ['NLS_LANG'] = 'SIMPLIFIED CHINESE_CHINA.UTF8' from cx_Oracle import makedsn from sqlalchemy import create_engine, Column, String, Integer from sqlalchemy.ext.declarative import declarative_base from sqlalchemy.orm import sessionmaker # 连接数据库(ORA-12505: TNS:listener does not currently know of SID given in connect descriptor) ip = '' port = '' tnsname = '' # 实例名 uname = '' # 用户名 pwd = '' # 密码 dsnStr = makedsn(ip, port, service_name=tnsname) connect_str = "oracle://%s:%s@%s" % (uname, pwd, dsnStr) # 创建连接引擎,这个engine是lazy模式,直到第一次被使用才真实创建 engine = create_engine(connect_str, encoding='utf-8') # 创建对象的基类 Base = declarative_base() class Student(Base): # 表名 __tablename__ = 'student' # 表字段 sid = Column(String(20), primary_key=True) age = Column(Integer) # 建表(继承Base的所有表) Base.metadata.create_all(bind=engine) # 使用ORM操作数据库 Session = sessionmaker(bind=engine) # 创建ORM基类 session = Session() # 创建ORM对象 tb_obj = Student(sid='a6', age=18) # 创建表对象 session.add(tb_obj) # 添加到ORM对象(插入数据) session.commit() # 提交 session.close() # 关闭ORM对象 # 删表(继承Base的所有表) Base.metadata.drop_all(engine)
报错处理
DPI-1047: 64-bit Oracle Client library cannot be loaded
首先操作系统位数、python位数、cx_Oracle版本要对应上;另外可能缺【Visual C++】
每次装完后,要重启pycharm和python
ORA-12170: TNS:Connect timeout occurred
打开终端ping一下
检查【主机名或IP地址】、【服务名或SID】、【用户名】和【密码】是否填对
中文乱码
encoding=‘utf8'
ORA-00972: identifier is too long
insert语句中出现'之类的字符
解决方法:将可能报错的字符替换掉
ORA-64203: Destination buffer too small to hold CLOB data after character set conversion.
select TO_CHAR(long_text) from table_name,目标缓冲区太小,无法储存CLOB转换字符后的数据
解决方法:不在SQL用TO_CHAR,改在Python中用read(如上代码所示)
ORA-01704: string literal too long
虽然CLOB可以保存长文本,但是SQL语句有长度限制
解决方法:把超长文本保存在一个变量中(如上代码所示)
PLS-00172: string literal too long
字符串长度>32767(215-1)
解决方法:使用'||'来连接字符串(如上代码所示)
ORA-00928: missing SELECT keyword
INSERT操作时,表字段命名与数据库内置名称冲突,如:ID、LEVEL、DATE等
解决方法:建立命名规范
cx_Oracle.DatabaseError: ORA-12505: TNS:listener does not currently know of SID given in connect descriptor
使用sqlalchemy时的报错
原因可能是目标数据库是集群部署的,可以咨询一下DBA,或见上面代码from cx_Oracle import makedsn
UnicodeEncodeError: 'ascii' codec can't encode character
使用sqlalchemy时的报错,插入中文字符引起
解决方法是设置os.environ['NLS_LANG']
更多关于Python连接Oracle之环境配置、实例代码及报错解决方法请查看下面的相关链接