python2.6写看门狗并发邮件

top -b 就可以去掉特殊字符。

python写看门狗很合适，永不死，以及方便获取各种信息。

若程序死掉，获取一些信息，然后重启程序，发邮件。

发送的邮件为html格式，比文本更方便阅读。

#!/usr/bin/python2.6# -*- coding: utf-8 -*-import time;import os;import sys;import signal;import socket;# set the default encoding to utf-8# reload sys model to enable the getdefaultencoding method.reload(sys);# using exec to set the encoding, to avoid error in IDE.exec("sys.setdefaultencoding('utf-8')");assert sys.getdefaultencoding().lower() == "utf-8";#TODO: check python version, >=2.6 requires.def handler(signum, frame):    print 'usr press ctrl+c, exit';    sys.exit(0)signal.signal(signal.SIGINT, handler)# the command to get and check the My server statuscheck_My_server_command = "/etc/init.d/My-server status";# when server stopped, restart it.start_My_server_command = "/etc/init.d/My-server start";# the interval to check the server.check_My_server_interval_seconds = 5;# sleep seconds when main function exceptionmain_function_exception_sleep_seconds = 10;# email to user listto_user = ["[e-mail]", "[e-mail]"]# constsMy_server_dir = "/usr/local/My-server";My_server_version_command = "%s/objs/My_server"%(My_server_dir);My_server_conf_file="%s/conf/My_server.conf"%(My_server_dir);My_server_log_prefix = "mylog_server_system";import smtplibfrom email.mime.text import MIMETextdef send_mail(smtp_server, username, password, to_user, subject, content):    msg = MIMEText(content, _subtype='html', _charset='gb2312');    msg['Subject'] = subject;    msg['From'] = username;    msg['To'] = ";".join(to_user);        try:        smtp = smtplib.SMTP();                smtp.connect(smtp_server);         smtp.login(username, password);        smtp.sendmail(username, to_user, msg.as_string());        smtp.close();                return 0;    except Exception, ex:        print "send email failed: %s"%(ex);        return 1;if len(sys.argv) <= 3:    print """Usage: %s <email_smtp_server> <email_username> <email_password>        email_smtp_server the smtp server. eg. smtp.126.com        email_username    the email user to login to smtp server. eg. [e-mail]        email_password    the email user password to login. e.g xxxxxFor example:        %s smtp.126.com [e-mail] xxxxx"""%(sys.argv[0], sys.argv[0]);    sys.exit(1);(email_smtp_server, email_username, email_password) = sys.argv[1:];print "email_smtp_server=%s, email_username=%s, email_password=%s"%(email_smtp_server, email_username, email_password);def strip_execute_command_get_output(cmd):    output = execute_command_get_output(cmd);    output = [x.strip() for x in output];    return output;    def execute_command_get_output(cmd):    try:        pipe = os.popen(cmd);        output = pipe.readlines();        pipe.close();        return output;    except Exception, ex:        print "execute cmd %s failed: %s"%(cmd, ex);        return "error";hostname = socket.gethostname();iplist = strip_execute_command_get_output('''ifconfig 2>/dev/null |grep "inet addr"|awk '{print $2}'|awk -F ':' '{print $2}'|grep -v "127.0.0.1"''');sysinfo = "%s %s"%(sys.platform, sys.version.replace("\n", " "));print "hostname=%s, iplist=%s, sysinfo=%s"%(hostname, iplist, sysinfo);def generate_content_then_start_server():    basic_cmd = time.strftime('%Y-%m-%d %X', time.localtime());    basic_info = "%s\n%s\n%s"%(hostname, " ".join(iplist), sysinfo);    version_cmd = "%s -v 2>&1"%(My_server_version_command);    version_info = execute_command_get_output(version_cmd);    ls_cmd = "ls -lh %s 2>&1; ls -lh %s/log 2>&1"%(My_server_dir, My_server_dir);    ls_info = execute_command_get_output(ls_cmd);    lsbrelease_cmd = "lsb_release -a 2>&1";    lsbrelease_info = execute_command_get_output(lsbrelease_cmd);    uname_cmd = "uname -a 2>&1";    uname_info = execute_command_get_output(uname_cmd);    ccamr_cmd = "ccamr list 2>&1";    ccamr_info = execute_command_get_output(ccamr_cmd);    ps_cmd = "ps aux|grep My|grep -v Myd|grep -v grep 2>&1";    ps_info = execute_command_get_output(ps_cmd);    df_cmd = "df -h 2>&1";    df_info = execute_command_get_output(df_cmd);    free_cmd = "free -m 2>&1";    free_info = execute_command_get_output(free_cmd);    conf_cmd = "cat %s 2>&1"%(My_server_conf_file);    conf_info = execute_command_get_output(conf_cmd);    top_cmd = "top -c -n 1 -b 2>&1";    top_info = execute_command_get_output(top_cmd);    vmstat_cmd = "vmstat 1 3 2>&1";    vmstat_info = execute_command_get_output(vmstat_cmd);    iostat_cmd = "iostat 1 3 2>&1";    iostat_info = execute_command_get_output(iostat_cmd);        print "dump info success";        # old logs    old_logs = strip_execute_command_get_output("(cd %s/log/ && ls |grep %s) 2>&1"%(My_server_dir, My_server_log_prefix));        # start server.    start_cmd = "%s  2>&1"%(start_My_server_command);    start_info = execute_command_get_output(start_cmd);    print "try start server completed";        # get the system start logs    newlogs_cmd = "(cd %s/log/ && ls |grep %s"%(My_server_dir, My_server_log_prefix);    for i in old_logs:        newlogs_cmd = "%s| grep -v %s"%(newlogs_cmd, i);    newlogs_cmd = "%s| xargs cat) 2>&1"%(newlogs_cmd);    newlogs_info = execute_command_get_output(newlogs_cmd);        content = '''        <style>            div.cmd{                background-color: #00E000;                font-size: 15px;                font-weight: bold;                padding: 2px 3px 2px 5px;            }            div.info{                background-color: #CCCCCC;                margin-top: 0px;                margin-bottom: 10px;                margin-left: 10px;                font-size: 12px;                padding: 2px 3px 2px 5px;            }        </style>        <!-- basic -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- version -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- ls -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- lsbrelease -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- uname -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- ccamr -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- ps -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- df -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- free -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- conf -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- top -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- vmstat -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- iostat -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- start server -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>        <!-- server log -->        <div>            <div class='cmd'>%s</div>            <div class='info'>%s</div>        </div>    '''%(basic_cmd, "<pre>%s</pre>"%"".join(basic_info),     version_cmd, "<pre>%s</pre>"%"".join(version_info),     ls_cmd, "<pre>%s</pre>"%"".join(ls_info),     lsbrelease_cmd, "<pre>%s</pre>"%"".join(lsbrelease_info),     uname_cmd, "<pre>%s</pre>"%"".join(uname_info),     ccamr_cmd, "<pre>%s</pre>"%"".join(ccamr_info),    ps_cmd, "<pre>%s</pre>"%"".join(ps_info),     df_cmd, "<pre>%s</pre>"%"".join(df_info),     free_cmd, "<pre>%s</pre>"%"".join(free_info),     conf_cmd, "<pre>%s</pre>"%"".join(conf_info),     top_cmd, "<pre>%s</pre>"%"".join(top_info),     vmstat_cmd, "<pre>%s</pre>"%"".join(vmstat_info),     iostat_cmd, "<pre>%s</pre>"%"".join(iostat_info),     start_cmd, "<pre>%s</pre>"%"".join(start_info),     newlogs_cmd, "<pre>%s</pre>"%"".join(newlogs_info)    );        #print content;    return content;def main():    while True:        print "start check server status";        ret = os.system("%s  > /dev/null 2>&1"%(check_My_server_command));        if ret != 0:            print "server crash, ret=%s"%(ret);            subject = "[MyServer] watchdog server %s crash"%(hostname);            content = generate_content_then_start_server();            print "email to %s"%(to_user);            if 0 == send_mail(email_smtp_server, email_username, email_password, to_user, subject, content):                print "email success";        else:            print "server is ok";        time.sleep(check_My_server_interval_seconds);while True:    try:        main();    except Exception, ex:        print "main function exception: %s"%(ex);    time.sleep(main_function_exception_sleep_seconds);