def _execute_in_fork(command_to_exec: CommandType) -> None:
pid = os.fork()
if pid:
# In parent, wait for the child
pid, ret = os.waitpid(pid, 0)
if ret == 0:
return
raise AirflowException('Celery command failed on host: ' +
get_hostname())
from airflow.sentry import Sentry
ret = 1
try:
from airflow.cli.cli_parser import get_parser
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(command_to_exec[1:])
setproctitle(f"airflow task supervisor: {command_to_exec}")
args.func(args)
ret = 0
except Exception as e: # pylint: disable=broad-except
log.error("Failed to execute task %s.", str(e))
ret = 1
finally:
Sentry.flush()
os._exit(ret) # pylint: disable=protected-access
def _execute_work_in_fork(self, command: CommandType) -> str:
pid = os.fork()
if pid:
# In parent, wait for the child
pid, ret = os.waitpid(pid, 0)
return State.SUCCESS if ret == 0 else State.FAILED
from airflow.sentry import Sentry
ret = 1
try:
import signal
from airflow.cli.cli_parser import get_parser
signal.signal(signal.SIGINT, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(command[1:])
setproctitle(f"airflow task supervisor: {command}")
args.func(args)
ret = 0
return State.SUCCESS
except Exception as e: # pylint: disable=broad-except
self.log.error("Failed to execute task %s.", str(e))
finally:
Sentry.flush()
os._exit(ret) # pylint: disable=protected-access
def _start_by_fork(self):
pid = os.fork()
if pid:
self.log.info("Started process %d to run task", pid)
return psutil.Process(pid)
else:
# Start a new process group
os.setpgid(0, 0)
import signal
signal.signal(signal.SIGINT, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
from airflow import settings
from airflow.cli.cli_parser import get_parser
from airflow.sentry import Sentry
# Force a new SQLAlchemy session. We can't share open DB handles
# between process. The cli code will re-create this as part of its
# normal startup
settings.engine.pool.dispose()
settings.engine.dispose()
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(self._command[1:])
# We prefer the job_id passed on the command-line because at this time, the
# task instance may not have been updated.
job_id = getattr(args, "job_id", self._task_instance.job_id)
self.log.info('Running: %s', self._command)
self.log.info('Job %s: Subtask %s', job_id, self._task_instance.task_id)
proc_title = "airflow task runner: {0.dag_id} {0.task_id} {0.execution_date_or_run_id}"
if job_id is not None:
proc_title += " {0.job_id}"
setproctitle(proc_title.format(args))
try:
args.func(args, dag=self.dag)
return_code = 0
except Exception as exc:
return_code = 1
self.log.error(
"Failed to execute job %s for task %s (%s; %r)",
job_id,
self._task_instance.task_id,
exc,
os.getpid(),
)
finally:
# Explicitly flush any pending exception to Sentry if enabled
Sentry.flush()
logging.shutdown()
os._exit(return_code)
def _start_by_fork(self):
pid = os.fork()
if pid:
self.log.info("Started process %d to run task", pid)
return psutil.Process(pid)
else:
import signal
from airflow import settings
from airflow.cli.cli_parser import get_parser
from airflow.sentry import Sentry
signal.signal(signal.SIGINT, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
# Start a new process group
os.setpgid(0, 0)
# Force a new SQLAlchemy session. We can't share open DB handles
# between process. The cli code will re-create this as part of its
# normal startup
settings.engine.pool.dispose()
settings.engine.dispose()
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(self._command[1:])
self.log.info('Running: %s', self._command)
self.log.info('Job %s: Subtask %s', self._task_instance.job_id,
self._task_instance.task_id)
proc_title = "airflow task runner: {0.dag_id} {0.task_id} {0.execution_date_or_run_id}"
if hasattr(args, "job_id"):
proc_title += " {0.job_id}"
setproctitle(proc_title.format(args))
try:
args.func(args, dag=self.dag)
return_code = 0
except Exception:
return_code = 1
finally:
# Explicitly flush any pending exception to Sentry if enabled
Sentry.flush()
logging.shutdown()
os._exit(return_code)
def _execute_in_fork(command_to_exec: CommandType,
celery_task_id: Optional[str] = None) -> None:
pid = os.fork()
if pid:
# In parent, wait for the child
pid, ret = os.waitpid(pid, 0)
if ret == 0:
return
msg = f'Celery command failed on host: {get_hostname()} with celery_task_id {celery_task_id}'
raise AirflowException(msg)
from airflow.sentry import Sentry
ret = 1
try:
from airflow.cli.cli_parser import get_parser
settings.engine.pool.dispose()
settings.engine.dispose()
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(command_to_exec[1:])
args.shut_down_logging = False
if celery_task_id:
args.external_executor_id = celery_task_id
setproctitle(f"airflow task supervisor: {command_to_exec}")
args.func(args)
ret = 0
except Exception as e:
log.exception("[%s] Failed to execute task %s.", celery_task_id,
str(e))
ret = 1
finally:
Sentry.flush()
logging.shutdown()
os._exit(ret)
def _start_by_fork(self): # pylint: disable=inconsistent-return-statements
pid = os.fork()
if pid:
self.log.info("Started process %d to run task", pid)
return psutil.Process(pid)
else:
from airflow.cli.cli_parser import get_parser
from airflow.sentry import Sentry
import signal
import airflow.settings as settings
signal.signal(signal.SIGINT, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
# Start a new process group
os.setpgid(0, 0)
# Force a new SQLAlchemy session. We can't share open DB handles
# between process. The cli code will re-create this as part of its
# normal startup
settings.engine.pool.dispose()
settings.engine.dispose()
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(self._command[1:])
proc_title = "airflow task runner: {0.dag_id} {0.task_id} {0.execution_date}"
if hasattr(args, "job_id"):
proc_title += " {0.job_id}"
setproctitle(proc_title.format(args))
try:
args.func(args, dag=self.dag)
return_code = 0
except Exception: # pylint: disable=broad-except
return_code = 1
finally:
# Explicitly flush any pending exception to Sentry if enabled
Sentry.flush()
os._exit(return_code) # pylint: disable=protected-access
def _execute_work_in_fork(self, command: CommandType) -> str:
pid = os.fork()
if pid:
# In parent, wait for the child
pid, ret = os.waitpid(pid, 0)
return State.SUCCESS if ret == 0 else State.FAILED
from airflow.sentry import Sentry
ret = 1
try:
import signal
from airflow.cli.cli_parser import get_parser
signal.signal(signal.SIGINT, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
signal.signal(signal.SIGUSR2, signal.SIG_DFL)
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(command[1:])
args.shut_down_logging = False
setproctitle(f"airflow task supervisor: {command}")
args.func(args)
ret = 0
return State.SUCCESS
except Exception as e:
self.log.exception("Failed to execute task %s.", e)
return State.FAILED
finally:
Sentry.flush()
logging.shutdown()
os._exit(ret)
def _start_by_fork(self):
pid = os.fork()
if pid:
self.log.info("Started process %d to run task", pid)
return psutil.Process(pid)
else:
# Start a new process group
os.setpgid(0, 0)
import signal
signal.signal(signal.SIGINT, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
from airflow import settings
from airflow.cli.cli_parser import get_parser
from airflow.sentry import Sentry
from airflow.utils.cli import get_dag
# Force a new SQLAlchemy session. We can't share open DB handles
# between process. The cli code will re-create this as part of its
# normal startup
settings.engine.pool.dispose()
settings.engine.dispose()
parser = get_parser()
# [1:] - remove "airflow" from the start of the command
args = parser.parse_args(self._command[1:])
# We prefer the job_id passed on the command-line because at this time, the
# task instance may not have been updated.
job_id = getattr(args, "job_id", self._task_instance.job_id)
self.log.info('Running: %s', self._command)
self.log.info('Job %s: Subtask %s', job_id, self._task_instance.task_id)
proc_title = "airflow task runner: {0.dag_id} {0.task_id} {0.execution_date_or_run_id}"
if job_id is not None:
proc_title += " {0.job_id}"
setproctitle(proc_title.format(args))
return_code = 0
try:
# parse dag file since `airflow tasks run --local` does not parse dag file
dag = get_dag(args.subdir, args.dag_id)
args.func(args, dag=dag)
return_code = 0
except Exception as exc:
return_code = 1
self.log.error(
"Failed to execute job %s for task %s (%s; %r)",
job_id,
self._task_instance.task_id,
exc,
os.getpid(),
)
except SystemExit as sys_ex:
# Someone called sys.exit() in the fork - mistakenly. You should not run sys.exit() in
# the fork because you can mistakenly execute atexit that were set by the parent process
# before fork happened
return_code = sys_ex.code
except BaseException:
# while we want to handle Also Base exceptions here - we do not want to log them (this
# is the default behaviour anyway. Setting the return code here to 2 to indicate that
# this had happened.
return_code = 2
finally:
try:
# Explicitly flush any pending exception to Sentry and logging if enabled
Sentry.flush()
logging.shutdown()
except BaseException:
# also make sure to silently ignore ALL POSSIBLE exceptions thrown in the flush/shutdown,
# otherwise os._exit() might never be called. We could have used `except:` but
# except BaseException is more explicit (and linters do not comply).
pass
# We run os._exit() making sure it is not run within the `finally` clause.
# We cannot run os._exit() in finally clause, because during finally clause processing, the
# Exception handled is held in memory as well as stack trace and possibly some objects that
# might need to be finalized. Running os._exit() inside the `finally` clause might cause effects
# similar to https://github.com/apache/airflow/issues/22404. There Temporary file has not been
# deleted at os._exit()
os._exit(return_code)