def need_to_restart():
""" auto_restart=True而且长时间没有touchdb的
:return:
"""
# 超时没有touch db的
processes = Process.select().where(
Process.status == STATUS.RUNNING, Process.auto_restart == 1,
Process.timeout_timestamp < int(time.time()))
for process in processes:
if Process.update(status=STATUS.EXITED) \
.where(Process.id == process.id,
Process.status == STATUS.RUNNING,
Process.auto_restart == 1,
Process.timeout_timestamp < int(time.time())) \
.execute() == 1:
logger.info("重启进程%s" % process.id)
restart_process(program_id=process.id)
# 处于exited状态的
processes = Process.select().where(Process.status == STATUS.EXITED,
Process.auto_restart == 1)
for process in processes:
if Process.update(status=STATUS.EXITED) \
.where(Process.id == process.id,
Process.status == STATUS.EXITED,
Process.auto_restart == 1) \
.execute() == 1:
logger.info("重启进程%s" % process.id)
restart_process(program_id=process.id)
def touch_db(program_id, touch_timeout):
logger.info("进程%s运行中..." % program_id)
timeout_timestamp = int(time.time() + touch_timeout)
# 如果程序处于STARTING或STOPPING状态，无需touch db
try:
process = ProcessModel.select() \
.where(ProcessModel.id == program_id) \
.get()
except DoesNotExist:
logger.warning("touch_db失败，没有这条记录，%s可能已停止" % program_id)
return False
if process.status in (STATUS.STARTING, STATUS.STOPPING):
return True
# noinspection PyBroadException
try:
ret = ProcessModel.update(timeout_timestamp=timeout_timestamp) \
.where(ProcessModel.id == program_id,
ProcessModel.status == STATUS.RUNNING) \
.execute()
except Exception:
logger.exception("touch_db异常")
return False
if ret == 0:
logger.warning("touch_db失败，没有这条记录，%s可能已停止" % program_id)
return False
return True
def local_stop(program_id, wait_timeout=10):
try:
process = Process.select().where(Process.id == program_id).get()
except DoesNotExist:
msg = "找不到进程%s的配置数据，无法启动" % program_id
logger.warning(msg)
raise exceptions.NoConfigException(msg)
if process.status == STATUS.STOPPED:
msg = "进程%s已停止，忽略本次请求" % process.name
logger.warning(msg)
raise exceptions.AlreadyStopException()
pid = get_pid(program_id)
if pid == 0:
logger.warning("要停止的进程不存在，忽略")
return True
args = ['kill', str(pid)]
subprocess.Popen(args).wait()
if not wait_until_stop_done(program_id, wait_timeout):
logger.warning("进程%s停止失败" % process.name)
raise exceptions.StopException()
return True
def get_program(program_id=None, program_name=None, status=None):
""" 单条查询程序信息，program_id和program_name至少需要一个
:param program_id: 程序ID
:param program_name: 程序名称
:param status: 程序状态
:return: 程序对象
"""
if program_id is not None and program_name is not None:
raise exceptions.ParamValueException("program_id和program_name至少需要一个")
conditions = []
if program_id is not None:
conditions.append(Process.id == program_id)
if program_name is not None:
conditions.append(Process.name == program_name)
if status is not None:
conditions.append(Process.status == status)
try:
program = Process.select().where(*conditions).get()
except DoesNotExist:
raise exceptions.ProgramNotExistInDB()
except DatabaseError:
logger.exception("查询程序%s时，数据库发生异常" % program_id)
raise exceptions.MySQLDBException("查询程序时，数据库发生异常")
return program
def get_programs(status=None):
""" 批量查询程序信息
:param status: 程序状态
:return: 程序对象列表
"""
conditions = []
if status is not None:
conditions.append(Process.status == status)
try:
if conditions:
return Process.select().where(*conditions)
else:
return Process.select()
except DatabaseError:
logger.exception("查询程序列表时，数据库发生异常")
raise exceptions.MySQLDBException("查询程序列表时，数据库发生异常")
def update_program(program_id, **fields):
""" 更新一条进程
:param program_id:
:return:
True - 更新成功
False - 更新失败
"""
try:
Process.select().where(Process.id == program_id).get()
except DoesNotExist:
msg = "程序id=%s的配置不存在" % program_id
logger.error(msg)
raise exceptions.NoConfigException(msg)
Process.update(**fields) \
.where(Process.id == program_id) \
.execute()
return True
def get_status(program_id):
try:
process = Process.select().where(Process.id == program_id).get()
except DoesNotExist:
return False
if process.status == STATUS.STOPPED:
return False
return check_process(program_id)
def local_start(program_id, wait=3):
try:
process = Process.select().where(Process.id == program_id).get()
except DoesNotExist:
msg = "找不到进程%s的配置数据，无法启动" % program_id
logger.warning(msg)
raise exceptions.NoConfigException(msg)
if process.status == STATUS.RUNNING:
msg = "进程%s已启动，忽略本次请求" % process.name
logger.warning(msg)
raise exceptions.AlreadyStartException()
command = process.command
directory = process.directory
environment = process.environment
touch_timeout = process.touch_timeout
stdout_logfile = process.stdout_logfile
stderr_logfile = process.stderr_logfile
info = {
'program_id': process.id,
'program_name': process.name,
'directory': directory,
'environment': environment,
'touch_timeout': touch_timeout,
'stdout_logfile': stdout_logfile,
'stderr_logfile': stderr_logfile,
}
args = json.dumps(command)
info = json.dumps(info)
p = subprocess.Popen(['dswrapper', args, info], stdout=subprocess.PIPE)
p.wait()
if p.returncode != 0:
logger.warning("进程%s启动失败" % process.name)
raise exceptions.StartException()
if not check_start_status(program_id, wait):
logger.warning("进程%s启动失败" % process.name)
raise exceptions.StartException()
return True
def change_status(program_id, from_status, to_status):
""" 修改程序状态
:param program_id: 程序ID
:param from_status: 修改前的状态元组。程序状态不在from_status中时，修改失败
:param to_status: 修改后的状态
:return:
True - 修改成功
False - 修改失败
"""
if isinstance(from_status, int):
from_status = (from_status, )
fields = {"status": to_status, "update_time": tools.get_now_time()}
try:
process = Process.select().where(Process.id == program_id).get()
except DoesNotExist:
logger.error("程序%s不存在" % program_id)
return False
except DatabaseError:
logger.exception("查询进程%s时，数据库发生异常" % program_id)
return False
if process.status not in from_status:
logger.error("程序%s的状态%s不是%s" %
(program_id, process.status, from_status))
return False
try:
ret_code = Process.update(fields) \
.where(Process.id == program_id,
Process.status << from_status).execute()
except DatabaseError:
logger.exception("程序%s的状态%s=>%s更新时，数据库发生异常" %
(program_id, from_status, to_status))
return False
if ret_code == 0:
logger.warning("程序%s的状态%s=>%s更新失败，ID或状态不匹配" %
(program_id, from_status, to_status))
return False
return True