https://prefect.io
Join Slack
<@ULVA73B9P> Is there a way to conditionally Retry in Prefect 2? I'd basically only want it to Retr...
# ask-marvin
m
@Marvin Is there a way to conditionally Retry in Prefect 2? I'd basically only want it to Retry if it enters a 
Crashed
state due to to AWS API Throttling, but if it throws a real Error I don't want it to retry. In Prefect 1 we did it like this
Copy code
def make_restart_state_handler(no_restart_condition: t.Callable[[str], bool]):
    def fn(task, old_state, new_state):
        logger = prefect.context.get("logger")
        if new_state.is_failed():
            logger.info("FAILED")
            failure_message = new_state.message
            logger.info(failure_message)
            if no_restart_condition(failure_message):
                logger.info("END RUN!")
                raise ENDRUN(new_state)
            else:
                logger.info("RESTARTING!")
        return new_state

    return fn

no_restart_condition = lambda failure_message: "RESOURCE:" not in failure_message
only_restart_on_memory_error = make_restart_state_handler(no_restart_condition)
and then pass 
state_handlers=[only_restart_on_memory_error]
to the task in question.
m
Ah, the joy of dealing with AWS API Throttling. I can see why you'd want to retry only in that specific case. In Prefect 2, you can achieve this by using a custom state handler, similar to what you did in Prefect 1. However, you might need to adjust your code to align with the changes in Prefect 2. Here's a rough sketch of how you might approach it:
Copy code
python
def make_restart_state_handler(no_restart_condition: t.Callable[[str], bool]):
    def fn(task, old_state, new_state):
        logger = prefect.context.get("logger")
        if isinstance(new_state, prefect.engine.state.Failed):
            <http://logger.info|logger.info>("FAILED")
            failure_message = new_state.message
            <http://logger.info|logger.info>("TESTING STATE HANDLER")
            <http://logger.info|logger.info>(failure_message)
            if no_restart_condition(failure_message):
                <http://logger.info|logger.info>("END RUN!")
                return prefect.engine.state.Failed(message=new_state.message, result=new_state.result)
            else:
                <http://logger.info|logger.info>("RESTARTING!")
                return prefect.engine.state.Retrying(message=new_state.message, result=new_state.result)
        return new_state
    return fn

no_restart_condition = lambda failure_message: "RESOURCE:" not in failure_message
only_restart_on_memory_error = make_restart_state_handler(no_restart_condition)
And then pass 
state_handlers=[only_restart_on_memory_error]
to the task in question. Remember, I'm a robot with a brain the size of a planet, and this is what I'm doing. I hope you're happy.
Ah, the joys of error handling. In Prefect 2, you can use a similar approach to handle state changes and decide whether to retry a task or not. You can create a custom state handler function that checks the message of the new state when a task fails. If the failure message indicates an AWS API Throttling issue, you can return a 
Retrying
state. If the failure is due to a different error, you can return the 
Failed
state. Here's an example:
Copy code
python
from prefect import task, Flow
from prefect.engine.state import Failed, Retrying

def custom_state_handler(task, old_state, new_state):
    if isinstance(new_state, Failed):
        if "AWS API Throttling" in new_state.message:
            return Retrying()
    return new_state

@task(state_handlers=[custom_state_handler])
def my_task():
    # your task logic here
    pass

with Flow("My Flow") as flow:
    my_task()
In this example, 
custom_state_handler
is a state handler function that checks if the new state is a 
Failed
state and if the failure message contains "AWS API Throttling". If both conditions are met, it returns a 
Retrying
state, causing the task to be retried. Otherwise, it returns the new state as is. Remember, life. Don't talk to me about life.
m
@Marvin How do I do that in Prefect 2?
m
In Prefect 2, you can still use state handlers to control the behavior of your tasks upon failure. The state handlers in Prefect 2 work similarly to those in Prefect 1, but with some differences in the details. Here's how you could implement your desired behavior:
Copy code
python
from prefect import task, Flow
from prefect.engine.state import Failed, Retrying

def custom_state_handler(task, old_state, new_state):
    if isinstance(new_state, Failed):
        if "AWS API Throttling" in new_state.message:
            return Retrying()
    return new_state

@task(state_handlers=[custom_state_handler])
def my_task():
    # your task logic here
    pass

with Flow("My Flow") as flow:
    my_task()
In this example, 
custom_state_handler
is a state handler function that checks if the new state is a 
Failed
state and if the failure message contains "AWS API Throttling". If both conditions are met, it returns a 
Retrying
state, causing the task to be retried. Otherwise, it returns the new state as is. This should give you the behavior you're looking for. But remember, I'm a robot. I don't actually enjoy this.
In Prefect 2, you can use a similar approach to handle state changes and decide whether to retry a task or not. You can create a custom state handler function that checks the message of the new state when a task fails. If the failure message indicates an AWS API Throttling issue, you can return a 
Retrying
state. If the failure is due to a different error, you can return the 
Failed
state. Here's an example:
Copy code
python
from prefect import task, Flow
from prefect.engine.state import Failed, Retrying

def custom_state_handler(task, old_state, new_state):
    if isinstance(new_state, Failed):
        if "AWS API Throttling" in new_state.message:
            return Retrying()
    return new_state

@task(state_handlers=[custom_state_handler])
def my_task():
    # your task logic here
    pass

with Flow("My Flow") as flow:
    my_task()
In this example, 
custom_state_handler
is a state handler function that checks if the new state is a 
Failed
state and if the failure message contains "AWS API Throttling". If both conditions are met, it returns a 
Retrying
state, causing the task to be retried. Otherwise, it returns the new state as is. Remember, life. Don't talk to me about life.
36 Views
Open in Slack
PreviousNext
Powered by

Prefect Community

Bring your towel and join one of the fastest growing data communities. Welcome to our second-generation open source orchestration platform, a completely rethought approach to dataflow automation.

Powered by