oracle监控

python代码

#!/usr/bin/env python
# -*- coding: UTF-8 -*-
import subprocess
import sys
import re

def run_cmd(cmd, cwd=None, runas=None):
    if not sys.platform.startswith('win') and runas and runas != 'root':
        cmd = 'su - {} -c "{}"'.format(runas, cmd)
    proc = subprocess.Popen(cmd,
                            stdout=subprocess.PIPE,
                            stderr=subprocess.STDOUT,
                            shell=True,
                            cwd=cwd)
    return proc

def get_os_oracle_user():
    cmd = ''' ps -ef | grep ora_ | grep -v grep | awk '{ print $1 }' | uniq '''
    code, res = run_cmd(cmd)
    if code == 0:
        os_oracle_user = res
        return os_oracle_user
    else:
        print "获取运行database的操作系统用户失败...", res
        sys.exit(1)

def get_grid_user():
    cmd = '''ps -ef | grep asm_pmon | grep -v grep |uniq'''
    return_code, result = run_cmd(cmd)
    if return_code == 0 and result:
        os_oracle_user = result.split('\n')[0].split()[0].strip()
        return os_oracle_user
    else:
        print "获取运行database的操作系统用户失败...", result
        sys.exit(1)

class Oracle_init():
    """所有监控oracle的基类"""
    def __init__(self, user='monitoruser', password='123456'):
        try:
            if user == 'sysdba':
                self.user = '/ as sysdba'
            elif user and password:
                self.user = '{0}/{1}'.format(user, password)
            else:
                self.user = None
            self.os_oracle_user=get_os_oracle_user()
        except Exception as err:
            print err
            sys.exit(1)

    def run_cmd(self,sid,cmd):
        cmd = r'''su - {} -c "export ORACLE_SID={}
        sqlplus -S /nolog <<EOF
        conn /as sysdba;
        {}
        exit;
        EOF"'''.format(self.os_oracle_user, sid,cmd)
        code, res = run_cmd(cmd)
        error=self.check_error(res)
        if error:
            print error
            sys.exit(1)
        else:
            return res

    def check_error(self,res):
        if re.search(r'ORA-01017', res):
            error = '账号或密码错误'
        elif re.search(r'ORA-01034', res):
            error = '数据库不可用'
        elif re.search('ORA-\d+:[\s\S]*', res):
            error = re.search('ORA-\d+:[\s\S]*', res).group()
        else:
            error = None
        return error

# 获取所有实例名
def get_sids():
    sids = ''
    cmd = "ps -ef|grep -v grep |grep ora_pmon|awk '{print $NF}'|sed 's/ora\_pmon\_//'|uniq"
    code, res = run_cmd(cmd)
    if not code and res:
        sids = res.replace('\n', ' ')
    return sids

########判断ohasd状态##############
def get_oracleOhasdStatus():
    try:
        cmd = "ps -ef | grep ohasd.bin|grep -v grep"
        code,res=run_cmd(cmd)
        if code:
            return False
        else:
            return True
    except Exception as err:
        sys.stderr.write(err.__str__() + '\n')
        return False

################监听日志大小##############
def get_oracle_listenerLog(sid):
    try:
        global is_cluster
        if is_cluster:
            user=get_grid_user()
        else:
            user= get_os_oracle_user()
        cmd = 'su - {} -c "export ORACLE_SID={};' \
              'du -sm $ORACLE_BASE/diag/tnslsnr/`hostname`/listener/trace/listener.log"'.format(user,sid)
        code,res=run_cmd(sid, cmd)
        if not res:
            print '未获得数据'
            sys.exit(1)
        else:
            return res.split()[0]
    except Exception as err:
        sys.stderr.write(err.__str__() + '\n')
        return -1

################监听trace文件# ##############
def get_oracleTrace(sid):
    try:
        global is_cluster
        if is_cluster:
            user = get_grid_user()
            cmd = 'su - {} -c "export ORACLE_SID={};cat $ORACLE_BASE/diag/asm/+asm/$ORACLE_SID/trace/alert_+ASM1.log |egrep \'ORA-|ERROR\'"'.format(
                user, sid)
        else:
            user = get_os_oracle_user()
            cmd='su - {} -c "export ORACLE_SID={};cat $ORACLE_BASE/diag/rdbms/$DB_UNIQUENAME/$ORACLE_SID/trace |egrep \'ORA-|ERROR\'"'.format(
                user, sid)

        code, res = run_cmd(sid, cmd)
        if not res:
            print '未获得数据'
            sys.exit(1)
        else:
            return res.split()[0]
    except Exception as err:
        sys.stderr.write(err.__str__() + '\n')
        return -1

if __name__=='__main__':
    output = "metric=%s|value=%d|type=%s|tags=%s"
    print output % ("system.file.number", get_oracleOhasdStatus(), 'gauge', '')

    obj = Oracle_init(user='monitoruser', password='123456')

    sql_cmd = "select value from v\\\$parameter where name='cluster_database';"
    sids=get_sids()
    res=obj.run_cmd(sids[0],sql_cmd)
    res=res.split('\n')
    length=len(res)
    is_cluster=None
    for i in range(-1,-1-length,-1):
        if not i :
            continue
        if i.strip()=='False':
            is_cluster=False
            break
        if i.strip()=='True':
            is_cluster=True
            break
    if is_cluster is None:
        print '无法判断是否是集群'
        sys.exit(1)

    for sid in sids:
        print output % ("system.file.number", get_oracle_listenerLog(sid), 'gauge', '')
        print output % ("system.file.number", get_oracleTrace(sid), 'gauge', '')