我们的项目中用apschedule作为核心定时调度模块。所以对apschedule进行了一些调查和源码级的分析。
1、为什么选择apschedule？
听信了一句话，apschedule之于python就像是quartz之于java。实际用起来还是不错的。
2、安装
# pip安装方式
$ pip install apscheduler
# 源码编译方式
$ wget https://pypi.python.org/pypi/APScheduler/#downloads
$ python setup.py install
3、apschedule有四个主要的组件
1）trigger - 触发器
2）job stores - 任务存储（内存memory和持久化persistence）
3）executor - 执行器(实现是基于concurrent.futures的线程池或者进程池)
4）schedulers - 调度器(控制着其他的组件，最常用的是background方式和blocking方式)
先上一个例子
# -*- coding:utf-8 -*-
import redis
from datetime import datetime, timedelta
from apscheduler.schedulers.background import BackgroundScheduler
from apscheduler.jobstores.redis import RedisJobStore
from apscheduler.jobstores.sqlalchemy import SQLAlchemyJobStore
from apscheduler.executors.pool import ThreadPoolExecutor, ProcessPoolExecutor
from apscheduler.events import EVENT_JOB_MAX_INSTANCES, EVENT_JOB_ERROR, EVENT_JOB_MISSED
class ScheduleFactory(object):
def __init__(self):
if not hasattr(ScheduleFactory, '__scheduler'):
__scheduler = ScheduleFactory.get_instance()
self.scheduler = __scheduler @staticmethod
def get_instance():
pool = redis.ConnectionPool(
host='10.94.99.56',
port=6379,
)
r = redis.StrictRedis(connection_pool=pool)
jobstores = {
'redis': RedisJobStore(2, r),
'default': SQLAlchemyJobStore(url='sqlite:///jobs.sqlite')
}
executors = {
'default': ThreadPoolExecutor(max_workers=30),
'processpool': ProcessPoolExecutor(max_workers=30)
}
job_defaults = {
'coalesce': False,
'max_instances': 3
}
scheduler = BackgroundScheduler(jobstores=jobstores, executors=executors, job_defaults=job_defaults, daemonic=False)
return scheduler
说明：上例中，scheduleFactory被实现为一个单例模式，保证new出的对象全局唯一
4、对scheduler的选择
这里只给出两个场景：
1）BackgroundScheduler：这种方式在创建scheduler的父进程退出后，任务同时停止调度。适用范围：集成在服务中，例如django。
2）BlockingScheduler：这种方式会阻塞住创建shceduler的进程，适用范围：该程序只干调度这一件事情。
选择完调度器之后
1）scheduler.start() 启动调度器
2）scheduler.shutdown() 停止调度器，调用该方法，调度器等到所有执行中的任务执行完成再退出，可以使用wait=False禁用
程序变为如下样子
class ScheduleFactory(object):
def __init__(self):
if not hasattr(ScheduleFactory, '__scheduler'):
__scheduler = ScheduleFactory.get_instance()
self.scheduler = __scheduler @staticmethod
def get_instance():
pool = redis.ConnectionPool(
host='10.94.99.56',
port=6379,
)
r = redis.StrictRedis(connection_pool=pool)
jobstores = {
'redis': RedisJobStore(2, r),
'default': SQLAlchemyJobStore(url='sqlite:///jobs.sqlite')
}
executors = {
'default': ThreadPoolExecutor(max_workers=30),
'processpool': ProcessPoolExecutor(max_workers=30)
}
job_defaults = {
'coalesce': False,
'max_instances': 3
}
scheduler = BackgroundScheduler(jobstores=jobstores, executors=executors, job_defaults=job_defaults, daemonic=False)
# scheduler = BlockingScheduler(jobstores=jobstores, executors=executors, job_defaults=job_defaults, daemonic=False)
return scheduler def start(self):
self.scheduler.start() def shutdown(self):
self.scheduler.shutdown()
5、对jobstores的选择
大的方向有两个：
1）非持久化
可选的stores：MemoryJobStrore
适用于你不会频繁启动和关闭调度器，而且对定时任务丢失批次不敏感。
2）持久化
可选的stores：SQLAlchemyJobStore， RedisJobStore，MongoDBJobStore，ZooKeeperJobStore
适用于你对定时任务丢失批次敏感的情况
jobStores初始化配置的方式是使用一个字典，例如
jobstores = {
'redis': RedisJobStore(2, r),
'default': SQLAlchemyJobStore(url='sqlite:///jobs.sqlite')
}
key是你配置store的名字，后面在添加任务的使用，可以指定对应的任务使用对应的store，例如这里选用的都是key=default的store。
def add_job(self, job_func, interval, id, job_func_params=None)
self.scheduler.add_job(job_func, jobstore='default', trigger='interval', seconds=interval, id=id, kwargs=job_func_params, executor='default', next_run_time=next_run_time, misfire_grace_time=30)
6、executor的选择
只说两个，线程池和进程池。默认default是线程池方式。这个数是执行任务的实际并发数，如果你设置的小了而job添加的比较多，可能出现丢失调度的情况。
同时对于python多线程场景，如果是计算密集型任务，实际的并发度达不到配置的数量。所以这个数字要根据具体的要求设置。
一般来说我们设置并发为30，对一般的场景是没有问题的。
executors = {
'default': ThreadPoolExecutor(max_workers=30),
'processpool': ProcessPoolExecutor(max_workers=30)
}
同样在add_job的时候，我们可以选择对应的执行器
def add_job(self, job_func, interval, id, job_func_params=None)
self.scheduler.add_job(job_func, jobstore='default', trigger='interval', seconds=interval, id=id, kwargs=job_func_params, executor='default', next_run_time=next_run_time, misfire_grace_time=30)
7、trigger的选择
这是最简单的一个了，有三种，不用配置
1、date - 每天的固定时间
2、interval - 间隔多长时间执行
3、cron - 正则
8、job的增删改查接口api可以参看手册
http://apscheduler.readthedocs.io/en/latest/userguide.html#choosing-the-right-scheduler-job-store-s-executor-s-and-trigger-s
9、问题fix
1）2017-07-24 14:06:28,480 [apscheduler.executors.default:120] [WARNING]- Run time of job "etl_func (trigger: interval[0:01:00], next run at: 2017-07-24 14:07:27 CST)" was missed by 0:00:01.245424
这个问题对应的源码片段是
def run_job(job, jobstore_alias, run_times, logger_name):
"""
Called by executors to run the job. Returns a list of scheduler events to be dispatched by the
scheduler. """
events = []
logger = logging.getLogger(logger_name)
for run_time in run_times:
# See if the job missed its run time window, and handle
# possible misfires accordingly
if job.misfire_grace_time is not None:
difference = datetime.now(utc) - run_time
grace_time = timedelta(seconds=job.misfire_grace_time)
if difference > grace_time:
events.append(JobExecutionEvent(EVENT_JOB_MISSED, job.id, jobstore_alias,
run_time))
logger.warning('Run time of job "%s" was missed by %s', job, difference)
continue logger.info('Running job "%s" (scheduled at %s)', job, run_time)
try:
retval = job.func(*job.args, **job.kwargs)
except:
exc, tb = sys.exc_info()[1:]
formatted_tb = ''.join(format_tb(tb))
events.append(JobExecutionEvent(EVENT_JOB_ERROR, job.id, jobstore_alias, run_time,
exception=exc, traceback=formatted_tb))
logger.exception('Job "%s" raised an exception', job)
else:
events.append(JobExecutionEvent(EVENT_JOB_EXECUTED, job.id, jobstore_alias, run_time,
retval=retval))
logger.info('Job "%s" executed successfully', job) return events
这里面有个参数是misfire_grace_time，默认是1s，如果任务的实际执行时间与任务调度时间的时间差>misfire_grace_time，就会warning并且跳过这次任务的调度！！！
为什么会发生这个问题？
1）executor并发度不够，你添加的任务太多
2) misfire_grace_time，还是太小了
2）如果你使用的trigger=interval，并且设置了misfire_grace_time=30这种的话，如果你首次启动的时间是10:50那么调度间隔和实际执行可能有1分钟的误差
怎么解决这个问题呢，你可以通过next_run_time设置首次调度的时间，让这个时间取整分钟。例如
def add_job(self, job_func, interval, id, job_func_params=None):
next_minute = (datetime.now() + timedelta(minutes=1)).strftime("%Y-%m-%d %H:%M")
next_run_time = datetime.strptime(next_minute, "%Y-%m-%d %H:%M")
self.scheduler.add_job(job_func, jobstore='default', trigger='interval', seconds=interval, id=id, kwargs=job_func_params, executor='default', next_run_time=next_run_time, misfire_grace_time=30)
3）2017-07-25 11:02:00,003 [apscheduler.scheduler:962] [WARNING]- Execution of job "rule_func (trigger: interval[0:01:00], next run at: 2017-07-25 11:02:00 CST)" skipped: maximum number of running instances reached (1)
对应的源码为
for job in due_jobs:
# Look up the job's executor
try:
executor = self._lookup_executor(job.executor)
except:
self._logger.error(
'Executor lookup ("%s") failed for job "%s" -- removing it from the '
'job store', job.executor, job)
self.remove_job(job.id, jobstore_alias)
continue run_times = job._get_run_times(now)
run_times = run_times[-1:] if run_times and job.coalesce else run_times
if run_times:
try:
executor.submit_job(job, run_times)
except MaxInstancesReachedError:
self._logger.warning(
'Execution of job "%s" skipped: maximum number of running '
'instances reached (%d)', job, job.max_instances)
event = JobSubmissionEvent(EVENT_JOB_MAX_INSTANCES, job.id,
jobstore_alias, run_times)
events.append(event)
submit_job的源码
with self._lock:
if self._instances[job.id] >= job.max_instances:
raise MaxInstancesReachedError(job) self._do_submit_job(job, run_times)
self._instances[job.id] += 1
这是什么意思呢，当对一个job的一次调度的任务数>max_instances，会触发这个异常，并终止调度。例如对一个批次的调度，比如job1，在10:00这次的调度，执行的时候发现有两个任务被添加了。这怎么会发生呢？会。可能09:59分的调度没有成功执行，但是持久化了下来，那么在10：00会尝试再次执行。
max_instances默认是1，如果想让这种异常放过的话，你可以设置max_instances大一些，比如max_instances=3
10、如果你想监控你的调度，那么apschedule提供了listener机制，可以监听一些异常。只需要注册监听者就好
def add_err_listener(self):
self.scheduler.add_listener(err_listener, EVENT_JOB_MAX_INSTANCES|EVENT_JOB_MISSED|EVENT_JOB_ERROR) def err_listener(ev):
msg = ''
if ev.code == EVENT_JOB_ERROR:
msg = ev.traceback
elif ev.code == EVENT_JOB_MISSED:
msg = 'missed job, job_id:%s, schedule_run_time:%s' % (ev.job_id, ev.scheduled_run_time)
elif ev.code == EVENT_JOB_MAX_INSTANCES:
msg = 'reached maximum of running instances, job_id:%s' %(ev.job_id)
rs = RobotSender()
rs.send(
"https://oapi.dingtalk.com/robot/send?access_token=499ca69a2b45402c00503acea611a6ae6a2f1bacb0ca4d33365595d768bb2a58",
u"[apscheduler调度异常] 异常信息:%s" % (msg),
'',
False
)
最后的代码
# -*- coding:utf-8 -*-
import redis
from datetime import datetime, timedelta
from apscheduler.schedulers.background import BackgroundScheduler, BlockingScheduler
from apscheduler.jobstores.redis import RedisJobStore
from apscheduler.jobstores.sqlalchemy import SQLAlchemyJobStore
from apscheduler.executors.pool import ThreadPoolExecutor, ProcessPoolExecutor
from apscheduler.events import EVENT_JOB_MAX_INSTANCES, EVENT_JOB_ERROR, EVENT_JOB_MISSED
from alarmkits.send_robot import RobotSender class ScheduleFactory(object):
def __init__(self):
if not hasattr(ScheduleFactory, '__scheduler'):
__scheduler = ScheduleFactory.get_instance()
self.scheduler = __scheduler @staticmethod
def get_instance():
pool = redis.ConnectionPool(
host='10.94.99.56',
port=6379,
)
r = redis.StrictRedis(connection_pool=pool)
jobstores = {
'redis': RedisJobStore(2, r),
'default': SQLAlchemyJobStore(url='sqlite:///jobs.sqlite')
}
executors = {
'default': ThreadPoolExecutor(max_workers=30),
'processpool': ProcessPoolExecutor(max_workers=30)
}
job_defaults = {
'coalesce': False,
'max_instances': 3
}
scheduler = BackgroundScheduler(jobstores=jobstores, executors=executors, job_defaults=job_defaults, daemonic=False)
# scheduler = BlockingScheduler(jobstores=jobstores, executors=executors, job_defaults=job_defaults, daemonic=False)
return scheduler def start(self):
self.scheduler.start() def shutdown(self):
self.scheduler.shutdown() def add_job(self, job_func, interval, id, job_func_params=None):
next_minute = (datetime.now() + timedelta(minutes=1)).strftime("%Y-%m-%d %H:%M")
next_run_time = datetime.strptime(next_minute, "%Y-%m-%d %H:%M")
self.scheduler.add_job(
job_func,
jobstore='default',
trigger='interval',
seconds=interval,
id=id,
kwargs=job_func_params,
executor='default',
next_run_time=next_run_time,
misfire_grace_time=30,
max_instances=3
) def remove_job(self, id):
self.scheduler.remove_job(id) def modify_job(self, id, interval):
self.scheduler.modify_job(job_id=id, seconds=interval) def add_err_listener(self):
self.scheduler.add_listener(err_listener, EVENT_JOB_MAX_INSTANCES|EVENT_JOB_MISSED|EVENT_JOB_ERROR) def err_listener(ev):
msg = ''
if ev.code == EVENT_JOB_ERROR:
msg = ev.traceback
elif ev.code == EVENT_JOB_MISSED:
msg = 'missed job, job_id:%s, schedule_run_time:%s' % (ev.job_id, ev.scheduled_run_time)
elif ev.code == EVENT_JOB_MAX_INSTANCES:
msg = 'reached maximum of running instances, job_id:%s' %(ev.job_id)
rs = RobotSender()
rs.send(
"https://oapi.dingtalk.com/robot/send?access_token=499ca69a2b45402c00503acea611a6ae6a2f1bacb0ca4d33365595d768bb2a58",
u"[apscheduler调度异常] 异常信息:%s" % (msg),
'',
False
)