需求:kill 掉yarn上超时的任务,实现不同队列不同超时时间的kill机制,并带有任务名的白名单功能
此为python脚本,可配置crontab使用
# _*_ coding=utf-8 _*_
# !/usr/bin/python
import re
import commands
import time
run_app_arr = []
timeout_app_arr = []
ONE_HOURE = 1
THREE_HOURE = 3
TEST_QUEUE_NAME = ['hue', 'etl-test']
ONLINE_QUEUE_NAME = ['default']
KILL_WHITE_LIST = ['org.apache.spark.sql.hive.thriftserver.HiveThriftServer2']
DINGDING_URL = 'xxx'
ding_cmd = """ curl %s -H 'Content-Type: application/json' -d '{"msgtype": "text", "text": {"content": "== YARN OVERTIME JOB KILL 告警 ==\n\n 当前时间: %s \n kill_app_id: %s \n kill_app_name: %s \n kill_app_queue: %s "}}' """
f = None
try:
f = open('/home/hadoop/autokillhadoopjob/check_timeout_job.log', 'a')
commond = '. /etc/profile && yarn application -list | grep "http://" |grep "RUNNING" |cut -f1,2,5'
# 获得正在运行job的id,name,queue 加到 run_app_arr
status, output = commands.getstatusoutput(commond)
f.write('#' * 50 + '\n')
f.write('=> start_time: %s \n' % (time.strftime('%Y-%m-%d %H:%M:%S')))
if status == 0 :
for line in output.split('\n'):
if line.startswith('application_'):
app_line = re.split('\t', line)
running_app_id = app_line[0].strip()
running_app_name = app_line[1].strip()
app_queue = app_line[2].strip()
# 根据所在队列 筛选出app加到数组中
if app_queue in TEST_QUEUE_NAME or app_queue in ONLINE_QUEUE_NAME:
run_app_arr.append((running_app_id, running_app_name, app_queue))
else:
f.write('yarn -list 执行失败. status: %s.'%(status))
# 遍历所有队列的running job,如有超时加到timeout_app_arr
for run_app in run_app_arr:
running_app_id = run_app[0]
running_app_name = run_app[1]
running_app_queue = run_app[2]
commond = ". /etc/profile && yarn application -status " + running_app_id + "| grep 'Start-Time' | awk -F ':' '{print $2}'"
status, output = commands.getstatusoutput(commond)
if status == 0:
for line in output.split('\n'):
start_timestamp = line.strip()
if start_timestamp.isdigit():
# 计算任务耗时
elapsed_time = time.time() - int(start_timestamp) / 1000
cost_time = round(elapsed_time / 60 / 60, 2)
f.write('=> cost_time: %sh \n' % (cost_time))
# print cost_hour
# 筛选出超时的job 加到数据组中/过滤掉白名单任务
if running_app_name not in KILL_WHITE_LIST:
if (running_app_queue in TEST_QUEUE_NAME and cost_time > ONE_HOURE) \
or (running_app_queue in ONLINE_QUEUE_NAME and cost_time > THREE_HOURE):
# if cost_hour > 0:# 测试
f.write('=> timeout app => %s # %s # %s\n' % (running_app_id, running_app_name, running_app_queue))
timeout_app_arr.append((running_app_id, running_app_name, running_app_queue))
else:
f.write('yarn -status 执行失败. status: %s.'%(status))
if len(timeout_app_arr) == 0:
f.write('=> no timeout job.\n')
# kill掉超时的job 并dingding报警
for kill_app in timeout_app_arr:
kill_app_id = kill_app[0]
kill_app_name = kill_app[1]
kill_app_queue = kill_app[2]
commond = '. /etc/profile && yarn application -kill ' + kill_app_id
status, output = commands.getstatusoutput(commond)
if status == 0:
f.write('=> kill app sucessfully: %s # %s # %s.\n' % (kill_app_id, kill_app_name, kill_app_queue))
current_time = time.strftime('%Y-%m-%d %H:%M:%S')
cmd = ding_cmd % (DINGDING_URL, current_time, kill_app_id, kill_app_name, kill_app_queue)
commands.getstatusoutput(cmd)
else:
f.write('=> kill app failed: %s # %s # %s.\n' % (kill_app_id, kill_app_name, kill_app_queue))
f.write('=> stop_time: %s \n' % (time.strftime('%Y-%m-%d %H:%M:%S')))
except Exception as e:
f.write('=> Exception: %s \n' % (e.message))
finally:
if f:
f.close()