Hi, I'm a bit confused about how to specify images if I don't use the sandbox.config file here: https://docs.flyte.org/projects/cookbook/en/latest/auto/core/containerization/multi_images.html#sphx-glr-auto-core-containerization-multi-images-py I tried in

~/.flyte/config.yaml

but can't figure out the proper format

Subworkflow question: Is it possible to use a workflow from another project in flyte as a subworkflow or do they both need to be in the same project? Another microservice I'm working with uses a workflow that I need and don't want to reproduce. Would I be able to use that as a subworkflow without having access to the microservice's underlying code?

Hi, is it possible to, from the task code, get the hostname of the executing pod?

What do people currently do for more "stream based" workflow topologies (e.g. hyperparameter tuning, iterative optimization etc)? I'm thinking it might be possible to make a wrapper that creates a redis queue uses flyte serialization and deserialization to serialize things onto/off the queue, and then wraps python generators, but I'm wondering if anyone else has tried something like this. EDIT: in particular, it would be helpful to know if there is a callable version of the flyte serializer that could be used in something like rq

Hey guys! Is someone using Flyte Metrics from Datadog? I’m trying to make some charts to monitor the Propeller and I’m having troubles defining the metrics / aggregation / rates / etc. Metrics are arriving to Datadog already 👍 and I’m trying to follow the Monitoring docs. When trying to add a chart like in the screenshot, how can I know if the number I’m getting is correct? I’m sceptical as it changes when I change the timebound being observed, as well as the aggregation and those things, and I’m not sure which of them I should use

Hi everyone, I was able to run the blastX use case and NLP processor locally on my computer, but when I try to run them on the sandbox, it usually fails because its missing the package (gensim or blastx) needed, could someone help me out with this or guide me to the correct tutorial?

Hi everyone… who’s using tensorflow? I’m getting this

The TensorFlow library was compiled to use AVX instructions

Any specific version to install tensorflow?

Hello Community, As some of you know - my name is Ketan. I am one of the original authors of Flyte at Lyft and I am usually involved in the community. The community is growing and we are more than humbled by your love. I had a couple questions and requests from the community. 1. As we grow, we would love to help more and help you get most out of Flyte. But, often many questions are repeated. If we offer to use a hosted forum like discourse would you like it? 2. The community spreads and with large community, the entire community improves and benefits. But it does not spread without sharing and love. I would request folks to add your logos to the users list and help share the knowledge with others. I personally know so many companies that use flyte, but do not want to share the logo. This does not help the community. If you want help please dm me. Even if you do not want to share the logo, please let me know if you folks use Flyte, like it, do not like it, tried it and failed, how we can improve etc. Thank you, Love and respect, Ketan

Hi Flyte community, The JFlyte registry has not been updated since 0.3.6 (current version is 0.4.5): https://github.com/flyteorg/flytekit-java/pkgs/container/flytekit-java Are users currently building/pushing the images in their own private repository (or are the latest images stored in a different registry)?

hey quick question -- when our flyte pods start up, there is about ~1 minute between it starting to unpack the workflow to our task starting:

2023-02-28T15:52:39-05:00 tar: Removing leading `/' from member names
2023-02-28T15:53:09-05:00 
2023-02-28T15:53:31-05:00 2023-02-28 20:53:31.706 | INFO STARTED TASK

this is ubiquitous across all of our tasks and we're running a rather deep DAG locally, in serial (no --remote) it runs blazing fast (couple minutes) however, with the ~1min stall on each pod, it takes > 30minutes we'd rather not recombine tasks together if we don't need to so is there something we're missing here?

I’ve installed Flyte via the flyte-binary Single cluster simple cloud deployment path whilst testing and exploring. However, the default task resource limits are too low and I’m struggling to find where I would define these to override the defaults? Could anyone point me in the correct direction? Thanks

What do you think about making it explicit in the console whether an attempt failed due to a preemption? I get asked often by our ML engineers whether their attempt failed because of a user error or because of a preemption.

Hi everyone. I have question: I just try created a persistent volume for app: minio path /data but I got an error not permission /data

10:23:51.39

10:23:51.39 *Welcome to the Bitnami minio container*

10:23:51.39 Subscribe to project updates by watching *<https://github.com/bitnami/bitnami-docker-minio>*

10:23:51.39 Submit issues and feature requests at *<https://github.com/bitnami/bitnami-docker-minio/issues>*

10:23:51.39

10:23:51.40 INFO  ==> ** Starting MinIO setup **

/opt/bitnami/scripts/libminio.sh: line 317: /data/.access_key: Permission denied

This my spec configuration:

spec:

selector:

matchLabels:

app: minio

template:

metadata:

labels:

app: minio

spec:

containers:

- env:

- name: MINIO_ACCESS_KEY

value: xxxxxxxxxxxxxxx # I'm hiding my values

- name: MINIO_SECRET_KEY

value: xxxxxxxxxxxxxxx # I'm hiding my values

- name: MINIO_DEFAULT_BUCKETS

value: definedcrowd

image: <http://ecr.flyte.org/bitnami/minio:2021.10.13-debian-10-r0|ecr.flyte.org/bitnami/minio:2021.10.13-debian-10-r0>

name: minio

ports:

- containerPort: 9000

name: minio

- containerPort: 9001

name: minio-console

volumeMounts:

- mountPath: /data

name: minio-storage

securityContext:

fsGroup: 1001

fsGroupChangePolicy: OnRootMismatch

runAsUser: 1001

volumes:

- name: minio-storage

persistentVolumeClaim:

claimName: minio-storage-pv-claim

Hello! 👋 I have asked about problems with authentication (Azure oidc) earlier. This week I tried creating an entire new deployment using flyte-binary instead. However, this gives me the same problems during the redirect callback. In short:

Error generating encrypted accesstoken cookie [SECURE_COOKIE_ERROR] Error creating secure cookie, caused by: securecookie: the value is too long

Error setting encrypted JWT cookie [SECURE_COOKIE_ERROR] Error creating secure cookie, caused by: securecookie: the value is too long

Any advice on how I can continue is much appreciated.

our flyte-sandbox container often crashes when using multiple heavy containers. The hardware for the sandbox is a fairly powerful node with enough RAM and CPU resources. Any pointers on debuging this will be really helpful!

@lgladh I had this same error, I don’t know how I got it working again rsrsrsrs @Aswanth Krishnan I’m using cluster AKS this is not problem for me about RAM and CPU Thx guys I will be waiting for some answers about my problem, I would be very grateful

Please guys, what could be the cause of this?

RuntimeExecutionError: failed during plugin execution, caused by: failed to check existence of futures file: [User] Failed to do HEAD on futures file

I’m getting this error while creating a launch plan:

raise AssertionError("The cached values aren't the same as the current call arguments")

There isn’t much context to why it happened, what should I look for to debug it? The workflow has a datetime parameter to it, which was renamed between versions, but I don’t think it should stop it from deploying

Hi! i am currently changing from

pyflyte package

+

flytectl register

to only use

pyflyte register

when deploying flyte workflows - but now i get an authentication error. With

flytectl

i could authenticate by writing a secret to

~/.flyte/client_secret

but seems this is not working for

pyflyte

Maybe someone else had similar issues?

Does managed Flyte run on the customer's own cloud, or is it all hosted by union.ai?

@Felix Ruess didn’t finish but started looking at your pr yesterday.

Good morning@Yee @Ketan (kumare3). We recently updated to flytekit 1.3.2 and flyte 1.3.0. I’m seeing an issue where a task’s specific memory requests/limits and custom image is not being used. This is happening for a task where we set

task_config=Pod()

as we needed to customize some of the k8s spec. Usually for simpler tasks we set

requests

and

limits

for the

task

decorator It seems like the way we should be setting these values may have changed based on this? I’m reviewing this documentation now

Hello, In my organization, ETL workflows are owned by another team to generate data that my ML modeling workflows depend on. So, the ETL workflow is a daily scheduled launch plan (LP1). my modeling workflow is another daily scheduled launch plan (LP2). I need the daily execution of LP2 check for the successful completion of execution of LP1 for the same day before starting to execute. In airflow we did it via ExternalTaskSensor, is there a way to do it in Flyte? Could you share an existing pattern to avoid re-inventing the wheel? Thanks.

Hi! I'm working through the simple ML flyte demo. I am trying to figure out how far you can go with the

❯ pyflyte run --remote example.py training_workflow --hyperparameters '{"C": 0.1}'
Go to <http://localhost:30080/console/projects/flytesnacks/domains/development/executions/fa1c65afdde414c7d961> to see execution in the console.

syntax when you have custom dependencies as specified in

requirements.txt

. I got excited that the run command was somehow passing

scikit-learn

along, but now I see that

scikit-learn

is simply preinstalled in the base image. I am curious if there a way to handle custom dependencies without dropping out of the simple

pyflyte run

command and without having to have a "kitchen sink" style base image?

is there ci/cd workflows in flyte, if not did you implement workflows for ci/cd

Hello, I'm at a point with flyte now where I'm asking myself how bigger data (talking GBs) is passed efficiently between tasks. I cannot imagine that if I have a setup on e.g. AWS that those bigger output/input data is synced with S3 all the time. Or am I missing something? What would be the way to exchange big data between tasks inside a workflow without up and downloading it on every task again? I can imagine that it could be shared over a mounted volume from the k8s cluster, but that would probably interfere with the caching mechanism at some point, right?

Is there a way to configure the scheduler or flyte in some way to only allow at most N instances of a certain workflow/task to run at the same time in a namespace?

another question, I've got this task that failed and is retrying but in the taskmetadata, I've set retries=0. Was this a system retry then? Can I configure system retries on a workflow level.

I’m thinking about writing a type transformer for pydantic’s

BaseModel

(which has several benefits over data classes which is why our ML engineers asked for this).

from pydantic import BaseModel

class Conf(BaseModel):
    ...

@task
def train(conf: Conf):
   ...

Assuming there is a

class BaseModelTransformer(TypeTransformer[BaseModel]):

, would it be invoked if the user specifies

train(conf: Conf)

or only in case they specify

train(conf: BaseModel)

?

Hi everybody I'm working with FlyteFiles between two sepparate tasks: One download the file, the other works with it. In local environment, using sandbox, it work as expected. However, when running in a K8s cluster, it gives me this error:

[Errno 2] No such file or directory: '/tmp/flytep4bezuvh/local_flytekit/d8e1f470ac5ed7ac3141ab1b6d83a854/6W2MGqmQ.jpg'

Can anybody help me with this?