如何處理任務失敗並使用 on_failure 參數返回成功？

Question

我有一個prefect的Flow ，其task的輸出是dataframe 。 在下面提供的示例中，它總是失敗。 我希望該task使用@task(on_failure=handle_task_fail)返回一個狀態為SUCCESS的空dataframe 。 實現這一目標的正確語法是什么？

from pprint import pprint
import pandas as pd

from prefect import Flow, task
from prefect.engine.signals import SUCCESS


def handle_disambig_error(task, old_state, new_state):
    if new_state.is_failed():
        new_state.result["wiki_df"] = pd.DataFrame()

        # Is this needed?
        #set state to SUCCESS
    return new_state


@task(on_failure=handle_disambig_error)
def get_wiki_resource():

    wiki_df = pd.DataFrame(
        {
            "a":[1],
            "b":[1/0]
        }
    )

    return wiki_df

with Flow("Always Fail") as flow:
    wiki_df = get_wiki_resource()

state = flow.run()
task_state = state.result[wiki_df]
pprint(task_state.result)

追溯：

Traceback (most recent call last):
  File "/miniconda3/lib/python3.7/site-packages/prefect/engine/runner.py", line 161, in handle_state_change
    new_state = self.call_runner_target_handlers(old_state, new_state)
  File "/miniconda3/lib/python3.7/site-packages/prefect/engine/task_runner.py", line 120, in call_runner_target_handlers
    new_state = handler(self.task, old_state, new_state) or new_state
  File "/miniconda3/lib/python3.7/site-packages/prefect/utilities/notifications.py", line 69, in state_handler
    fn(obj, new_state)
TypeError: handle_disambig_error() missing 1 required positional argument: 'new_state'
[2020-01-28 17:39:41,759] INFO - prefect.TaskRunner | Task 'get_wiki_resource': finished task run for task with final state: 'Failed'
[2020-01-28 17:39:41,762] INFO - prefect.FlowRunner | Flow run FAILED: some reference tasks failed.

有些地方我搜索了State Handlers , Logging with a State Handler

Answer 1

這里發生了兩件事：

1.) 通用狀態處理程序：這些可以通過state_handlers kwarg 設置，並且將在每次狀態更改時調用。 狀態處理程序需要具有簽名state_handler(task: Task, old_state: State, new_state: State) -> Optional[State] （這是您正在使用的簽名）； 調用此處理程序后任務的狀態將是從處理程序返回的狀態，如果返回None則為new_state 。

2.) On Failure 回調：您在這里使用的on_failure kwarg 旨在成為狀態處理程序的便捷 API； 傳遞給這個關鍵字的函數需要有簽名fn(task: Task, state: State) -> None並且只有當這個 Task 進入Failed狀態時才會被調用。 請注意，失敗回調不能像狀態處理程序那樣改變任務的狀態。

在您的示例中，您似乎混合了兩個關鍵字參數。 我相信下面的代碼會做你所期望的：

from prefect.engine.state import Success


def handle_disambig_error(task, old_state, new_state):
    if new_state.is_failed():
        return_state = Success(result=pd.DataFrame())
    else:
        return_state = new_state
    return return_state

@task(state_handlers=[handle_disambig_error])
def get_wiki_resource():
   return df