Hi there! My colleagues and I are wondering how Flyte 2.0 works with the removal of the DSL (i.e: replacing flytekit.conditional with if/else etc.) Do you have a link to the flyte 2.0 repository or branch so we can look at this please?

Hi I think the flytekit noop builder has a bug: • if the environment launching the flyte workflow does not have docker available, then

image_spec.exist()

will be

None

and

should_build

goes to `False`: https://github.com/flyteorg/flytekit/blob/eb5a67f76aaef96a44bde04afb72b87592cc8b7a/flytekit/image_spec/image_spec.py#L479 • except the

noop

builder relies on

should_build

being

True

in order to effectively overwrite the image name: https://github.com/flyteorg/flytekit/blob/eb5a67f76aaef96a44bde04afb72b87592cc8b7a/flytekit/image_spec/noop_builder.py#L9 Thus

noop

builder will never work as intended if the launching shell does not have

docker

available. I think that's not as intended, because the

noop

builder is meant to essentially skip docker (and so ideally would work as intended if the user environment does not have docker available eh?)

Potentially dumb bug when building/compiling for flyte that I can't figure out (mildly redacted logs below):

bfrench@LM-BFRENCH:~/Documents/Code/monorepo$ pyflyte register example/flyteify.py -p 'ml-example' -d 'staging' -v 'bf-2025-08-01'

Running pyflyte register from /Users/bfrench/Documents/Code/monorepo with images ImageConfig(default_image=Image(name='default', fqn='<http://cr.flyte.org/flyteorg/flytekit|cr.flyte.org/flyteorg/flytekit>', tag='py3.10-1.16.3', digest=None), images=[Image(name='default', fqn='<http://cr.flyte.org/flyteorg/flytekit|cr.flyte.org/flyteorg/flytekit>', tag='py3.10-1.16.3', digest=None)]) and image destination folder /root on 1 package(s) ('/Users/bfrench/Documents/Code/monorepo/example/flyteify.py',)
Registering against <http://flyte.COMPANY.net|flyte.COMPANY.net>
Detected Root /Users/bfrench/Documents/Code/monorepo/example, using this to create deployable package...
Loading packages ['flyteify'] under source root /Users/bfrench/Documents/Code/monorepo/example
No output path provided, using a temporary directory at /var/folders/8n/83r7h0kx0xbgnddnmbggtz_m0000gp/T/tmp7_mk_77l instead

AttributeError: 'str' object has no attribute 'labels'

This is code that runs perfectly normally but flyte is complaining about an error with almost no detail provided. Has anyone seen this before?

Hi, I have a question for the creators of flyte-2. I am trying out the newer verion of it and i see you guys have used asyncio for async tasks. Is there a reason why

asyncio

was choosen over

anyio

To my understanding

anyio

will provide better functionality compared to

asyncio

.

Since it is early stages of development, the change might be easier

Is flyte-2 comaptible with local Flyte cluster: config:

admin:
  endpoint: dns:///localhost:30080
  insecure: true
image:
  builder: local
task:
  domain: development
  project: flytesnacks

I tried and it isn’ table to run the task, the code is from the doc:

# hello.py

# /// script
# requires-python = "==3.13"
# dependencies = [
#    "flyte==0.2.0b23",
# ]
# ///

import flyte

# A TaskEnvironment provides a way of grouping the configuration used by tasks.
env = flyte.TaskEnvironment(name="hello_world", resources=flyte.Resources(memory="250Mi"))

# Use a TaskEnvironment to define tasks, which are regular Python functions.
@env.task
def fn(x: int) -> int: # Type annotations are recommended.
    slope, intercept = 2, 5
    return slope * x + intercept

# Tasks can call other tasks.
# All tasks defined with a given TaskEnvironment will run in their own separate containers,
# but those containers will all be configured identically.
@env.task
def main(x_list: list[int]) -> float:

    x_len = len(x_list)
    if x_len < 10:
        raise ValueError(f"x_list doesn't have a larger enough sample size, found: {x_len}")

    # flyte.map is like Python map, but runs in parallel.
    y_list = list(flyte.map(fn, x_list))
    y_mean = sum(y_list) / len(y_list)
    return y_mean

# Running this script locally will perform a flyte.run, sending your task code to your remote Union/Flyte instance.
if __name__ == "__main__":

    # Establish a remote connection from within your script.
    flyte.init_from_config("config.yaml")
    #flyte.init(project="flytesnacks", domain="development", endpoint="localhost:30080",insecure=True,insecure_skip_verify=False)

    # Run your tasks remotely inline and pass parameter data.
    run = flyte.run(main, x_list=list(range(10)))
    # x= flyte.with_runcontext(mode="local").run(main, x_list=list(range(10)))
    # print(x)

    # Print various attributes of the run.
    print(run.name)
    print(run.url)

    # Stream the logs from the remote run to the terminal.
    run.wait(run)

Hi, Is the proto files available for flyte-2

I am not able to find it in the github repo

Hi there, I’m stuck running an execution on my flyte cluster. I’m running flyte-binary on a vanilla Kubernetes Cluster in the latest version with NetApp S3 Storage. I was able to register tasks, workloads and launchplans and for that the usage of the S3 Bucket was possible. It was also possible to run an execution, but it exited with this error message:

File "/usr/local/lib/python3.12/site-packages/flytekit/core/data_persistence.py", line 614, in async_get_data
    raise FlyteDownloadDataException(
flytekit.exceptions.system.FlyteDownloadDataException: SYSTEM:DownloadDataError: error=Failed to get data from s3://*****/test-project-01/development/YR7RMXMIOOCGYZIJKBIO2N4KUI======/fast6c01ca0737d31ff994073617f3ac5dec.tar.gz to /root/ (recursive=False).

Original exception: Unable to locate credentials
.

If I get it correctly, the difference to the registration process of the workflow is that the user context changed right? But now I stuck for 2 days because I can’t figure out what is the right place and best practice for providing the credentials to the workflow. Do I have to create a k8s-Service Account and connect them to the launchplan? And what is the right way to attach the credentials to the SA? The documentation is very focused on hyperscaler usage and less on prem setups. I’m thankful for any kind of help.

👋 having some issues trying to add an init container via pod template, curious if anyone has done this before or have any insights! • I currently create a

PodTemplate

resource in our cluster, and added an item to

initContainers

within the

spec

• I use

pod_template_name

in the task decorator to route it to the above resource • I additionally specify another template via

pod_template

in the task decorator, just to add some extra tolerations • and when I launch the task, I do see its pod get most of the fields from the

PodTemplate

resource like the volumes and mounts, but it just doesn't seem to merge in the

initContainers

Hi all. I’m new to the Flyte ecosystem and had a couple of (possibly naive) questions I’d really appreciate clarification on: 1. Assuming Flyte v2 is a commercial offering (with premium features and/or a subscription model), how might that affect the long-term support, maintenance, or evolution of Flyte v1? 2. Are there any Flyte v2 features that could eventually be backported or adopted into Flyte v1? Thanks in advance for any insights - trying to get a better grasp of the roadmap and how the versions relate to each other.

🙋 I have found a pair of issues that I'm unsure about (whether either is a known bug or should be raised as a new issue on github): • My tasks were failing with this trace, which I narrowed down to being because my

PodTemplate

for this task had

/bin/bash -c

as the

command

(where

arguments

was the

pyflyte-fast-execute ...

etc)

[...]
/usr/local/lib/python3.10/dist-packages/flytekit/bin/entrypoint.py:754 in    │
│ fast_execute_task_cmd                                                        │
│                                                                              │
│ ❱ 754 │   p = subprocess.Popen(cmd, env=env)                                 │
│                                                                              │
│ /usr/lib/python3.10/subprocess.py:971 in __init__                            │
│                                                                              │
│ ❱  971 │   │   │   self._execute_child(args, executable, preexec_fn, close_f │
│                                                                              │
│ /usr/lib/python3.10/subprocess.py:1733 in _execute_child                     │
│                                                                              │
│ ❱ 1733 │   │   │   │   executable = args[0]                                  │
╰──────────────────────────────────────────────────────────────────────────────╯
IndexError: list index out of range

• I was happy to find and correct the cause, but I was also confused because another domain for this same project was using the same

PodTemplate

and didn't have the `command`: I confirmed that the configured

PodTemplate

in my

cluster_resource_manager

config did not have this

command

either. I can't figure out when I added or removed this

command

but I'm assuming I removed it at some point: I'm guessing the cluster resource manager / syncresources can't detect removed keys from a PodTemplate? Is that a known limitation? ◦ I resolved this by doing a manual

kubectl edit podtemplate...

and removing the

command

entirely but curious whether its worth reporting this

Hi I am going though the Flyte-2 code and have a suggestion. This is a only and opinion/suggestion,

# Current Implementation
run = flyte.run(main, x_list=list(range(10)))

# Suggestion
run = flyte.run(main, x_list=list(range(10),logger,log_level))

Does it sound fair if we could pass a logger to the function, so we can get a detailed trace and place the logger file on a location of our choice. We would also need a log_level [INFO,DEBUG,ERROR] for the level of verbosity that is required Flyte CLI

flyte --help

has

flyte -vvv get logs <run-name>

I don’t see anything like that for python code.

Hi Flyte team! I'm trying to understand the best way to run containerized workloads with the Flyte Slurm backend. I noticed that SlurmTask doesn't have a

container_image

parameter like regular Flyte tasks. Is there a recommended approach for running Flyte containerized tasks on Slurm?

Hi, is there a way to add decks to ContainerTasks? And what is the recommended way to capture stdout?

Hi, we are highly interested in trying out Flyte 2.0. Do I understand it correctly that for now only the Python SDK (previously flytekit) is available but the server deployment (previously flyte) hasn't been released yet?

We're adding logging links to Flyte using the task logs plugin, we're using Grafana internally, we're giving it a time range so that it can filter the region of logs that has data fast. We first tried to use

podRFC3339StartTime

and

podRFC3339FinishTime

but the problem we have is that Grafana expects datetimes in RFC3339 in millisecond precision so our links don't currently work as podRFC3339StartTime which reads the pod.CreationTimestamp, and of course the creation timestamp is in seconds precision. The way we've "solved" this is to instead use the unix timestamp and pad it with zeroes:

Copy code

&from={{ "{{" }} .podUnixStartTime {{ "}}" }}000&to={{ "{{" }} .podUnixFinishTime {{ "}}" }}000

Hopefully, that helps folks who are using Grafana for their logging.

Hi, Here is a thought from my side. I was going though the examples related to flyte-2 sdk below: [EXAMPLE](https://github.com/flyteorg/flyte-sdk/blob/main/examples) I am planning to add anyio, to it while keeping asyncio as well. This way people can choose. Please let me know : maintainers of Flyte, does it sound ok.

I'm still stuck on displaying decks from a task that is fanned-out using map_task. I've found a git issue from 2023 saying this was going to be a supported issue and another user in the community messaged me saying this should work. However, after enabling decks in both the

@task(..., enable_deck=True)

and

map_task(my_function, concurrency=some_number, enable_deck=True)

, my tasks generated using map_task still don't show a Deck button to show rendered outputs for each deck. I re-worked my code to remove other complexities like @dynamic workflows and use of functools to define constant paramenters, but even when defining a @workflow that calls map_task(task_function), the Decks don't show up in Flyte UI. Do I have to return Deck object from the mapped task in order to get it to show up in the Flyte Dashboard? Is there another way to access the Decks from each of these mapped tasks that's not through the executed workflow(e.g. some kubernetes hack, another place in the Flyte UI)? Is there something I need to consider/add in the Flyte helm configuration file?

For maintainers of this cannel, Could you please make 2 channels. For v1 and v2 as both are getting in the same place and might require different priority

Hi flyte team, reaching out to ask if flyte has webhook we can use, our use case is: there is a external job dependency, we want to trigger this job as soon as the dependent job finished, ideally by webhook or some other mechanism, thanks!

Hi everyone, we're using Flyte 1.15 and I'm struggling with grasping the caching system when dataclasses with structuredDatasets are involved. Tried with custom HashMethod but to no avail I've build the following minimal example to explain the issue I'm facing

from dataclasses import dataclass
from typing import Annotated
from flytekit import Cache, HashMethod, StructuredDataset, task, workflow
import pandas as pd
import logging


@dataclass
class Data:
    metadata: str
    df: StructuredDataset


def hash_pandas_dataframe(df: pd.DataFrame) -> str:
    return str(pd.util.hash_pandas_object(df))


def hash_data(data: Data) -> str:
    # I cannot access the pd.Dataframe in the hash function?
    return str(pd.util.hash_pandas_object(data.df.open(pd.DataFrame).all()))


@task
def generate_data_a() -> Annotated[Data, HashMethod(hash_data)]:
    data = Data(
        metadata="hello",
        df=StructuredDataset(pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})),
    )
    return data


@task(cache=Cache(version="1.3"))
def process_data_a(data: Data) -> bool:
    logging.error(f"process_data_a: {data.df.open(pd.DataFrame).all()}")
    return True


@task
def generate_data_b() -> Annotated[pd.DataFrame, HashMethod(hash_pandas_dataframe)]:
    return pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})


@task(cache=Cache(version="1.3"))
def process_data_b(data: pd.DataFrame) -> bool:
    logging.error(f"process_data_b: {data}")
    return True


@workflow
def cache_workflow() -> None:
    # With the custom hashMethod for the `Data` object it crashes but perhaps that is also not the correct way to do it
    data_a = generate_data_a()
    process_data_a(data_a)

    # This caches correctly
    data_b = generate_data_b()
    process_data_b(data_b)

    return

Hi guys, we are planning to upgrade from flyte 1.13 to the latest and it looks like it would make sense to jump straight to 1.16 but I could not find info about when is that should be released out of betas?

I'm working through our monorepo conversion right now and I'm having problems registering flyte pipelines that include cross silo dependencies. For a minimal example, my monorepo has the following structure

pipelines
- src
- - tasks
- - - mytasks.py
- - workflows
- - - myworkflow.py
- - utils
- - - pipeline_utils.py
- shared
- - src
- - - some_utils.py

mytasks.py

has the import

from shared.src.some_utils import util

which runs fine on local but when I package up using

pyflyte register pipelines/src/workflows/

from the root of the monorepo, the packed tarball only includes

pipeline/src/tasks

,

pipeline/src/utils/

and

pipeline/src/workflows

but not

shared

despite references so I get immediate crashes. Has anyone else seen/solved this problem?

Hi Flyte team! I’m new to Flyte and exploring the best way to manage configurations. I like Hydra + OmegaConf and I see there’s a

flytekitplugins-omegaconf

plugin. I’m unsure about the best pattern for loading configs into workflows. It seems like I either need: 1. A Python entry point that builds the

DictConfig

before calling the workflow 2. A Flyte task inside the workflow that builds/loads the config Is there some established best practice pattern here? With the Python entry point approach it looks like you can’t run remotely with

pyflyte run ...

and instead have to

python run.py

cc: @acoustic-oyster-33294

Hi community, we are trying to talk to a local registry and make it available for pyflyte to push there. How do I give registry credentials to pyflyte/ImageSpec?

Hi! I'm trying to understand how to best use

pyflyte package

in our deployments.. The setup: • using

flytekit==1.16.1

• tasks are using

ImageSpec

containers without specifying any of the

source_root

,

copy

or

source_copy_mode

args (i.e. using the defaults) When changing code or dependencies in one of the tasks,

pyflyte package

generates new image tags and rebuilds every

ImageSpec

container, even the ones not affected by the code/deps changes. What's the recommended strategy to ensure that the

ImageSpec

tag only depends on the relevant task code and dependencies?

Hi Team I have installed Flyte via Helm on our AKS cluster. The UI is up and accessible; however, when I click on Login, it doesn’t proceed further. I’ve checked the pods and don’t see any errors there. Could someone help me troubleshoot this issue? Slack Conversation