Hey everyone - is there a timeline to when the prefect will be available on python 3.8

What would be the best practice/patterns how to implement a web crawler with Prefect ? Using the LOOP signal to recurse until some kind of depth limit?

I noticed that if I perform a map on two list:

my_tak.map([1, 2, 3], ["A", "B"])

will execute 2 tasks for

(1, "A"), (2, "B")

Is there an easy way to map over the 3x2 combos and having 6 tasks for

(1, "A"), (2, "A"), (3, "A"), (1, "B"), (2, "B"), (3, "B")

?

Hi everyone, thank you for such an awesome product

Is there tutorial somewhere about how to specify "run one task per gpu" or "one task per worker" at a time? This mostly WRT running GPU jobs on a SLURM cluster

Hey all, a quick question Let's say I

map

a function A over a list L, and then want another function B to continue working on elements of L in a parallel manner after A has finished. Is my understanding correct, that in order to implement this, function A should basically return the same element it got as an input, and pass it to B? i.e. in Python/pseudocode

def func_A(list_element):
    do_stuff(list_element)
    return list_element

def func_B(list_element):
    list_element+=2 

with Flow(...):
    L = Parameter("list", default=[1,2,3])
    first_step = func_A.map(L)
    second_step = func_B.map(first_step)

This seems to work, but I wonder whether this is the most efficient way to do this. I also noticed that, if I have several consecutive `map`s set up in the above manner

A -> B -> C -> D-> ...

, it will not begin the next step until the previous step has finished executing for all mapped elements. Is there a way to go around this? Is it because I'm using LocalExecutor and not something like DaskExecutor? Thanks!

Hi All, Is it possible to have two different flows on different schedules run in the same script? It seems that it isn’t because

Flow.run()

is a blocking call?

Hi all, the following code does not work:

from prefect import Flow, task

@task
def createX():
    return []

@task
def alterX(x):
    return x.append(5)

@task
def getZ(x):
    return len(x)

with Flow("Main Flow") as flow:
    x = createX()

    alterX(x)

    z = getZ(x, upstream_tasks=[alterX])

state = flow.run()

It should create x first, and then alter X, and then get Z with the altered x. Instead, it tries to alter x before x has been created. I could add the upstream task x to alterX, but I thought it should do that automatically because it is a data dependency.

Hi folks! We use Prefect core, and I recently updated Prefect from 0.7.1 to 0.9.2. This caused some issues with a task which used a model type which can't be serialized after training:

@task
def get_fit_model(model, x, y):
    model.fit(x, y)
    return model.fit_model

I eventually realized that this was because the model.fit() operation mutates the model object, and later prefect tries to serialize the mutated object

model

, even though only a part of it was used downstream. I was able to get around this by making a deepcopy() of the untrained model, on which the training was performed. I imagine this is due to the "greedy serialization" change Chris White mentioned a month or so back. I suppose the lesson here is that all operands must be serializable at all times. I'm concerned that perhaps I would get better results if I was using result handlers. I'm also curious if this would have been easier to debug if I were running my flow in Prefect Cloud.

@here Can anyone help me with this error. thanks

Failed: DID NOT RAISE <class 'prefect.engine.signals.SUCCESS'>

^^ my code is quite simple, and it is giving the above error on line - print(result)

with raises(signals.SUCCESS) as excinfo:

	result = some_lib.some_function()
	print(result)

assert excinfo.type == signals.SUCCESS
assert excinfo.typename == 'SUCCESS'

Hi! quick question: If I'm trying to pass a secret to the credentials property of the GCSUpload task, this is how I should do it right?

my_secret = Secret('MY_SECRET') # Where the JSON credential for a GCP service account has been copied to a Secret named 'MY_SECRET' via the cloud UI

upload_task = GCSUpload(project='foo', bucket='bar', ...)
upload_task(data='This is my data', blob='path/to/blob.txt', credentials=my_secret)

I'm having trouble with upload when running via Kubernetes agent.executor and using the cloud secret, but using the same credentials via GOOGLE_APPLICATION_CREDENTIALS env variable when running locally works fine

hi guys, any way to speed up (i.e. avoid serializing the whole result and sending it back to Prefect) task mapping? I've got a ~1M item list/tuple computed with Dask Bags which I'd like to map over, and between completing computation of the bag itself and the first mapped task I'm observing a good deal of time spent waiting for the complete result list

I am getting Event loop was unresponsive in Scheduler for 18.15s. followed by distributed.scheduler - ERROR - Couldn’t gather keys for a large data set this causes a loop in the prefect flow as it reschedules the tasks it couldn’t gather and eventually hits the same error. Any ideas on what to look for of any configuration options? I am using dask-kubernetes 0.10.1 and dask 2.11.0 in AWS.

I have some questions about

FargateTaskEnvironment

. 1. Can it just run an existing ECS task definition, or does it always register a new task definition based on the specified family/taskDefinition? 2. If it can just run an existing ECS task definition, what would the minimal necessary arguments be, something like

FargateTaskEnvironment(taskDefinition='my-ecr-task-definition')

, assuming AWS credentials are accessible by boto3 via the environment? 3. If it always generates/registers a new task definition, what would the minimal set of required arguments? I could probably go through the process of figuring that out by running

boto3.client('ecs').run_task(*args)

until I get the desired task definition running and then translate that into

FargateTaskEnvironment

arguments. Curious if someone has gone through this process already.

How do you unit test flows? I didn't see anything about it in the documentation...

Hi! I have some flows that run every hour according to a cron job. My question is whether it is possible to have one of the flows repeat again after 30 minutes (so essentially running that particular flow twice in the hour). I understand that I could use the scheduler in prefect, or just create a separate cron job that runs half-hourly, but I was wondering if there was a third option as well in prefect. Thanks!

Hi - I have

task

in a

Flow

that takes a

list

. How do I set rhe

List

Trigger

to run on

any_successful

upstream tasks? https://docs.prefect.io/core/concepts/tasks.html#collections I tried a couple approaches including def_list = List(trigger=any_successful) and then calling

munged_defs_df = definitions.munge_dfs(def_list([df1, df2, df3])

Hi, I am new to prefect and just finished reading through the documentation. Question about

map

responses. When I send a list of elements to a task and one element sent to the task finishes and returns a response, is it possible to move that response to the next task before the rest of the original elements finish in the first task? Thanks in advance!

Random PSA: our team ran into issues installing

psutil

(a dependency of

distributed

) on Catalina (OSX 10.15). This thread helped us get past it: https://github.com/giampaolo/psutil/issues/1632

has renamed the channel from "prefect-core" to "prefect-community"

I'm a little confused about the

upstream_tasks

arg on

Task.map

. "a list of upstream dependencies to map over". What would mapping over the upstream dependencies do?

Hi folks! I've been poking around a bunch with Prefect for the last couple months, and over that time I've written a few utilities/extensions to help Prefect work better for my particular data and modeling workflows. In case anyone else may be able to benefit I've pulled them together into a small package — https://pypi.org/project/prefect-ds/. Please feel free to take a look 🙂

if i want to improve performance through server selection, is memory or compute optimized better for dask?

Hey everyone, I'm new to the prefect community and was hoping to get advice on how to use prefect for my particular use case. Suppose we first have some batcher task that takes a set of items and batches them into groups. We then need a task to run on each batch. (So far, this can be handled using prefect's mapping). Next, we then want to run a task for each item in the batch and we'd like to be able to run this as soon as each batch finishes, rather than waiting for all batches to finish. To visualize a simpler version:

. --------> item a1 ------------> batch a --------> item a2 batcher --------> item a3 ------------> batch b --------> item b1 --------> item b2 Basically, I'd like prefect to be aware that once the batching happens, we now have two independent, parallelizable flows happening - so if batch a finishes, item a1 can begin even if batch b is still in progress.

Hey everyone, is there a task setting to prevent a mapping task from running until all upstream tasks are complete?

Im trying to determine the best route for reading in a list of text files using Prefect. Would I use the LOOP signal to loop over the list of files or should I build my flow as pertained to one file that way my flow represents a specific file that was processed?