def job_info():
req = request.values
id = int(req['id']) if ('id' in req and req['id']) else 0
info = JobList.query.filter_by(id=id).first()
if not info:
return redirect(GlobalUrlService.buildHomeUrl("/job/index/index"))
info = ModelHelper.model2Dict(info)
server_info = JobServer.query.filter_by(id=info['server_id']).first()
cate_info = JobCategory.query.filter_by(id=info['cate_id']).first()
server_env_map = CommonConstant.server_env_map
run_status_map = CommonConstant.run_status_map
info['next_run_time'] = DateHelper.getDateOnTimestamps(
info['next_run_time'], '%Y-%m-%d %H:%M')
info['env_name'] = server_env_map.get(info['env_id'])
info['run_status_desc'] = run_status_map.get(info['run_status'])
info['job_status_desc'] = job_status_map.get(info['status'])
info['server_name'] = server_info.name
info['cate_name'] = cate_info.name if cate_info else ''
info['run_interval_desc'] = DateHelper.formatBeautyTime(
info['run_interval'] * 60)
user_map = ModelHelper.getDictFilterField(
User,
select_field=User.id,
id_list=[info['owner_uid'], info['relate_uid']])
##获取最近5天运行记录
log_list = JobRunLog.query.filter_by(job_id=id).order_by(
JobRunLog.id.desc())[0:5]
log_data = []
if log_list:
for item in log_list:
tmp_data = ModelHelper.model2Dict(item)
tmp_data['status_desc'] = CommonConstant.job_log_status_map[
tmp_data['status']]
tmp_data['duration'] = ""
if DateHelper.getCurrentTime(date=tmp_data['end_time']
) == CommonConstant.DEFAULT_DATETIME:
tmp_data['end_time'] = "未知"
tmp_data['duration'] = time.time() - time.mktime(
tmp_data['start_time'].timetuple())
else:
tmp_data['duration'] = tmp_data['end_time'].timestamp(
) - tmp_data['start_time'].timestamp()
tmp_data['duration'] = DateHelper.formatBeautyTime(
tmp_data['duration'])
log_data.append(tmp_data)
return UtilHelper.renderView(
"home/job/index/info.html", {
"info": info,
"log_list": log_data,
"user_map": user_map,
"job_level_map": CommonConstant.job_level_map,
})
def alertRunTimeJudge(self, job_info, job_run_start_time):
#常驻Job不用判断时间
if job_info.job_type == CommonConstant.default_status_pos_2:
return True
job_run_total_time = int( time.time() - job_run_start_time)
job_id = job_info.id
alert_temp = "[运行时间异常]您的job(job_{0} {1})在{2}开始运行,预计时长{3} - {4}分钟,但是运行了{5}分钟"
alert_content = ""
if job_run_total_time < (job_info.threshold_down * 60 - 10 * 60): # 运行时长低于下限值，报警（阀值后面-10分钟）
alert_content = alert_temp.format(job_id, job_info.name,
DateHelper.getDateOnTimestamps(job_info.next_run_time,format="%Y-%m-%d %H:%M"),
job_info.threshold_down, job_info.threshold_up,DateHelper.formatBeautyTime( job_run_total_time ) )
elif job_run_total_time > ( job_info.threshold_up * 60 + 15 * 60 ):
alert_content = alert_temp.format(job_id, job_info.name,
DateHelper.getDateOnTimestamps(job_info.next_run_time,format="%Y-%m-%d %H:%M"),
job_info.threshold_down, job_info.threshold_up,DateHelper.formatBeautyTime( job_run_total_time ) )
if alert_content:
JobService.saveAlertLog( job_id , alert_content)
return True
def handleItem(self, item, host):
pid_file = self.getPidPath('job_%s.pid' % item['id'])
pid = 0
if self.checkPidExist(pid_file):
# 文件存储的是python的进程id，找真真运行的需要使用关键词查找
pid = self.findPidByKw(item['id'])
app.logger.info("%s : %s" % (pid_file, pid))
# 判断进程本身是否存在，但是由于删除pid在进程结束之后 会有延迟，所以需要临时增加删除功能
run_flag = os.path.exists('/proc/%s' % pid) if pid > 0 else False
'''
1=>该结束未结束
2=>该运行未运行
3=>执行命令异常
4=>Job平台标识正在运行，但是在机器上没有发现job进程
5=>Job平台标识没有运行，但是在机器上发现了job进程
6=>运行时间过短
'''
alert_content = ''
if item['run_status'] == CommonConstant.default_status_true and run_flag:
# 数据库中标记没有运行，但是在机器上发现了进程
alert_content = "您的job(job_{0} {1})运行状态异常,可能原因:{2}"\
.format(item['id'], item['name'], "Job平台标识没有运行，但是在机器%s上发现了job进程" % host )
os.system('ls -al /proc/%s' % pid)
os.system('ps -ef |grep tmp_job_%s' % item['id'])
elif item[
'run_status'] == CommonConstant.default_status_pos_2 and not run_flag:
# 重新检查job的状态，避免运行时间较短的job误报警
time.sleep(2)
job_info = JobList.query.filter_by(id=item['id']).first()
if job_info.run_status == CommonConstant.default_status_pos_2:
alert_content = "您的job(job_{0} {1})运行状态异常[Job平台会自动修复],可能原因:{2}" \
.format(item['id'], item['name'], "Job平台标识正在运行，但是在机器%s上没有发现job进程" % host )
elif item['run_status'] == CommonConstant.default_status_true and item[
'next_run_time'] < self.current_time:
# 该运行未运行
alert_content = "您的JOB(job_{0} {1})应该在{2}运行,但是现在还没有运行" \
.format(item['id'], item['name'], DateHelper.getDateOnTimestamps(item['next_run_time'], '%Y-%m-%d %H:%M'))
elif item['run_status'] == CommonConstant.default_status_pos_2 and item[
'job_type'] != CommonConstant.default_status_pos_2:
# 常驻Job不判断时间
# 对于固定运行时长，如果运行时间大于原计划时间的120%就报警；动态运行时长，超出上限就报警
run_duration_time = int(time.time() - item['next_run_time'])
# 使用动态预估运行时间确定是否报警，加了15分钟的缓冲时间
if run_duration_time > (item['threshold_up'] +
15) * 60: # 运行时长高于上限值，报警
run_duration_min = round(run_duration_time / 60, 1)
alert_content = "[运行超时]您的job(job_{0} {1})在{2}开始运行,预计时长{3}-{4}分钟,但是目前仍然在运行(运行了{5}分钟)"\
.format(item['id'], item['name'], DateHelper.getDateOnTimestamps(item['next_run_time'], '%Y-%m-%d %H:%M'),
item['threshold_down'], item['threshold_up'],run_duration_min )
# 自动修复
if 'Job平台标识正在运行' in alert_content and 'Job平台会自动修复' in alert_content:
try:
##下一分钟
# next_date = datetime.datetime.now() + datetime.timedelta(minutes=1)
# next_date = next_date.replace(second=0)
# next_run_time = int(time.mktime(next_date.timetuple()))
# params = {
# "next_run_time": next_run_time,
# "run_status": CommonConstant.default_status_true,
# "status": CommonConstant.default_status_true
# }
# JobList.query.filter_by(id=item['id']).update(dict(params))
# db.session.commit()
app.logger.info("您的job(job_{0} {1}) 自动修复成功~~".format(
item['id'], item['name']))
except:
app.logger.info("您的job(job_{0} {1}) 自动修复失败，错误原因：{2}~~".format(
item['id'], item['name'], self.getErrMsg()))
##存储到报警日志表
if alert_content:
app.logger.info(alert_content)
JobService.saveAlertLog(item['id'], alert_content)
return True
def job_index():
req = request.values
cate_id = int(req.get("cate_id", CommonConstant.default_status_false))
owner_uid = int(req.get("owner_uid", CurrentUserService.getUid()))
env_id = int(req.get("env_id", CommonConstant.default_status_false))
server_id = int(req.get("server_id", CommonConstant.default_status_false))
status = int(req.get("status", CommonConstant.default_status_neg_99))
display_status = int(
req.get("display_status", CommonConstant.default_status_neg_99))
job_type = int(req.get("job_type", CommonConstant.default_status_neg_99))
job_level = int(req.get("job_level", CommonConstant.default_status_neg_99))
kw = req.get("kw", "").strip()
page = int(req.get("p", 1))
query = JobList.query
if RBACService.checkDataPrivilege("all"):
pass
else:
query = query.filter_by(owner_uid=CurrentUserService.getUid())
if cate_id:
query = query.filter_by(cate_id=cate_id)
if env_id:
query = query.filter_by(env_id=env_id)
if owner_uid:
query = query.filter_by(owner_uid=owner_uid)
if server_id:
query = query.filter_by(server_id=server_id)
if status > CommonConstant.default_status_neg_99:
query = query.filter_by(status=status)
if display_status > CommonConstant.default_status_neg_99:
query = query.filter_by(is_del=display_status)
if job_type > CommonConstant.default_status_neg_99:
query = query.filter_by(job_type=job_type)
if job_level > CommonConstant.default_status_neg_99:
query = query.filter_by(job_level=job_level)
if kw:
if kw.isdigit():
query = query.filter_by(id=int(kw))
else:
query = query.filter(
or_(JobList.name.ilike('%{}%'.format(kw)),
JobList.command.ilike('%{}%'.format(kw))))
page_params = {
"total": query.count(),
"page_size": CommonConstant.PAGE_SIZE,
"page": page,
"display": CommonConstant.PAGE_DISPLAY
}
pages = UtilHelper.iPagination(page_params)
offset = (page - 1) * CommonConstant.PAGE_SIZE
limit = CommonConstant.PAGE_SIZE * page
list = query.order_by(JobList.id.desc())[offset:limit]
data = []
server_map = ModelHelper.getDictFilterField(JobServer)
cate_map = ModelHelper.getDictFilterField(JobCategory)
staff_map = ModelHelper.getDictFilterField(User)
server_env_map = CommonConstant.server_env_map
run_status_map = CommonConstant.run_status_map
job_type_map = CommonConstant.job_type_map
job_level_map = CommonConstant.job_level_map
if list:
for item in list:
tmp_data = ModelHelper.model2Dict(item)
tmp_server_info = ModelHelper.model2Dict(
server_map.get(tmp_data['server_id']))
tmp_cate_info = ModelHelper.model2Dict(
cate_map.get(tmp_data['cate_id']))
tmp_owner_staff_info = ModelHelper.model2Dict(
staff_map.get(tmp_data['owner_uid']))
tmp_relate_staff_info = ModelHelper.model2Dict(
staff_map.get(tmp_data['relate_uid']))
tmp_data['next_run_time'] = DateHelper.getDateOnTimestamps(
tmp_data['next_run_time'], '%Y-%m-%d %H:%M')
tmp_data['env_name'] = server_env_map.get(tmp_data['env_id'])
tmp_data['run_status_desc'] = run_status_map.get(
tmp_data['run_status'])
tmp_data['job_status_desc'] = job_status_map.get(
tmp_data['status'])
tmp_data['server_name'] = tmp_server_info.get("name")
tmp_data['cate_name'] = tmp_cate_info.get("name", '')
tmp_data['owner_name'] = tmp_owner_staff_info.get("name", '')
tmp_data['relate_name'] = tmp_relate_staff_info.get("name", '')
tmp_data['run_interval_desc'] = DateHelper.formatBeautyTime(
tmp_data['run_interval'] * 60)
data.append(tmp_data)
sc = {
'kw': kw,
'cate_id': cate_id,
'owner_uid': owner_uid,
'env_id': env_id,
'server_id': server_id,
'status': status,
'display_status': display_status,
'job_type': job_type,
'job_level': job_level,
}
set_flag = RBACService.checkPageRelatePrivilege("set")
ops_flag = RBACService.checkPageRelatePrivilege("ops")
return UtilHelper.renderView(
"home/job/index/index.html", {
"list": data,
"pages": pages,
"job_status_map": job_status_map,
"server_env_map": server_env_map,
"server_map": server_map,
"staff_map": staff_map,
"cate_map": cate_map,
"display_status_map": display_status_map,
"job_type_map": job_type_map,
"job_level_map": job_level_map,
"sc": sc,
"set_flag": set_flag,
"ops_flag": ops_flag,
})
def job_set():
if UtilHelper.isGet():
req = request.values
id = int(req['id']) if ('id' in req and req['id']) else 0
act = req.get("act", "").strip()
info = {'owner_uid': CurrentUserService.getUid()}
if id > 0:
info = JobList.query.filter_by(id=id).first()
info = ModelHelper.model2Dict(info)
info['next_run_time'] = DateHelper.getDateOnTimestamps(
info['next_run_time'], '%Y-%m-%d %H:%M')
if act == "copy":
info['id'] = 0
info['status'] = 0
info['name'] = "【复制】" + info['name']
server_list = JobServer.query.order_by(JobServer.id.desc()).all()
user_list = User.query.order_by(User.id.desc()).all()
cate_list = JobCategory.query.order_by(JobCategory.id.desc()).all()
return UtilHelper.renderView(
"home/job/index/set.html", {
"info": info,
"server_list": server_list,
"user_list": user_list,
"cate_list": cate_list,
"server_env_map": CommonConstant.server_env_map,
"job_type_map": CommonConstant.job_type_map,
"job_status_map": job_status_map,
"job_level_map": CommonConstant.job_level_map,
})
req = request.values
id = int(req['id']) if ('id' in req and req['id']) else 0
cate_id = int(req.get("cate_id", "0").strip())
name = req.get("name", "").strip()
env_id = int(req.get("env_id", "0").strip())
server_id = int(req.get("server_id", "0").strip())
owner_uid = int(req.get("owner_uid", "0").strip())
relate_uid = int(req.get("relate_uid", "0").strip())
command = req.get("command", "").strip()
command_kill = req.get("command_kill", "").strip()
job_type = int(req.get("job_type", "0").strip())
job_level = int(req.get("job_level", "1").strip())
status = int(req.get("status", "0").strip())
next_run_time = req.get("next_run_time", "").strip()
run_interval = int(req.get("run_interval", "0").strip())
threshold_down = int(req.get("threshold_down", "0").strip())
threshold_up = int(req.get("threshold_up", "0").strip())
note = req.get("note", "").strip()
if cate_id < 1:
return UtilHelper.renderErrJSON("请选择所属分类~~")
if not ValidateHelper.validLength(name, 1, 15):
return UtilHelper.renderErrJSON("请输入符合规范的名称~~")
if env_id < 1:
return UtilHelper.renderErrJSON("请选择运行环境~~")
if server_id < 1:
return UtilHelper.renderErrJSON("请选择运行服务器~~")
if owner_uid < 1:
return UtilHelper.renderErrJSON("请选择Job负责人~~")
if relate_uid < 1:
return UtilHelper.renderErrJSON("请选择Job相关人~~")
if not ValidateHelper.validLength(command, 5):
return UtilHelper.renderErrJSON("请输入Job命令~~")
if job_type < 1:
return UtilHelper.renderErrJSON("请选择Job类型~~")
if not ValidateHelper.validDate(next_run_time,
r'^\d{4}-\d{2}-\d{2} \d{2}:\d{2}$'):
return UtilHelper.renderErrJSON("请选择调度时间~~")
if job_type != CommonConstant.default_status_pos_2:
if run_interval < 1:
return UtilHelper.renderErrJSON("请输入运行间隔~~")
if threshold_down < 0:
return UtilHelper.renderErrJSON("请输入预估最短运行时长~~")
if threshold_up < 1:
return UtilHelper.renderErrJSON("请输入预估最长运行时长~~")
else:
run_interval = threshold_down = threshold_up = 0
info = JobList.query.filter_by(id=id).first()
if info:
model_job = info
##正在运行的Job不能编辑运行时间
if info.run_status == CommonConstant.default_status_pos_2:
if info.command != command:
return UtilHelper.renderErrJSON("正在运行的Job不能修改运行命令~~")
if info.job_type != job_type:
return UtilHelper.renderErrJSON("正在运行的Job不能修改类型~~")
if info.status != status:
return UtilHelper.renderErrJSON("正在运行的Job不能修改调度状态~~")
if info.next_run_time != DateHelper.getTimestamps(next_run_time +
":00"):
return UtilHelper.renderErrJSON("正在运行的Job不能修改调度时间~~")
else:
model_job = JobList()
##只有需要调度的才需要判断时间
if model_job.run_status != CommonConstant.default_status_pos_2 and status \
and DateHelper.getTimestamps(next_run_time + ":00") < time.time():
return UtilHelper.renderErrJSON("调度时间不能小于当前时间~~")
model_job.name = name
model_job.env_id = env_id
model_job.server_id = server_id
model_job.owner_uid = owner_uid
model_job.relate_uid = relate_uid
model_job.job_type = job_type
model_job.job_level = job_level
model_job.cate_id = cate_id
model_job.command = command
model_job.command_kill = command_kill
model_job.next_run_time = DateHelper.getTimestamps(next_run_time + ":00")
model_job.run_interval = run_interval
model_job.threshold_up = threshold_up
model_job.threshold_down = threshold_down
model_job.note = note
model_job.status = status
if status and (not info or not info.run_status):
model_job.run_status = CommonConstant.default_status_true
db.session.add(model_job)
db.session.commit()
return UtilHelper.renderSucJSON({"id": model_job.id})