Hi I was wondering if anyone has run into `ValueError Empty Flyte #ask-the-community

Hi I was wondering if anyone has run into `ValueEr...

Derek Yu

05/22/2023, 4:11 AM

Hi I was wondering if anyone has run into

ValueError: Empty module name

when running a

map_task()

with flytekit and trying to load the actual

task

object. More details below ⬇️ cc: @Heidi Hurst

Derek Yu

05/22/2023, 4:16 AM

When calling

map_task

with flytekit we are seeing this

ValueError: Empty module name

Copy code

│ ❱ 546 │   _execute_map_task(                                                 │
│                                                                              │
│ /root/micromamba/envs/xxx/lib/python3.10/site-packages/flytekit/e │
│ xceptions/scopes.py:160 in system_entry_point                                │
│                                                                              │
│ ❱ 160 │   │   │   │   return wrapped(*args, **kwargs)                        │
│                                                                              │
│ /root/micromamba/envs/xxx/lib/python3.10/site-packages/flytekit/b │
│ in/entrypoint.py:394 in _execute_map_task                                    │
│                                                                              │
│ ❱ 394 │   │   map_task = mtr.load_task(loader_args=resolver_args, max_concur │
│                                                                              │
│ /root/micromamba/envs/xxx/lib/python3.10/site-packages/flytekit/c │
│ ore/utils.py:295 in wrapper                                                  │
│                                                                              │
│ ❱ 295 │   │   │   │   return func(*args, **kwargs)                           │
│                                                                              │
│ /root/micromamba/envs/xxx/lib/python3.10/site-packages/flytekit/c │
│ ore/map_task.py:368 in load_task                                             │
│                                                                              │
│ ❱ 368 │   │   resolver_obj = load_object_from_module(resolver)               │
│                                                                              │
│ /root/micromamba/envs/xxx/lib/python3.10/site-packages/flytekit/t │
│ ools/module_loader.py:43 in load_object_from_module                          │
│                                                                              │
│ ❱ 43 │   class_obj_mod = importlib.import_module(".".join(class_obj_mod))    │
│                                                                              │
│ /root/micromamba/envs/xxx/lib/python3.10/importlib/__init__.py:12 │
│ 6 in import_module                                                           │
│                                                                              │
│ ❱ 126 │   return _bootstrap._gcd_import(name[level:], package, level)        │
│ in _gcd_import:1047                                                          │
│ in _sanity_check:981                                                         │
╰──────────────────────────────────────────────────────────────────────────────╯
ValueError: Empty module name

Yee

05/22/2023, 4:26 AM

that feels like the underlying task is not accessible.

Yee

05/22/2023, 4:26 AM

where do you have that defined?

Yee

05/22/2023, 4:26 AM

can you copy paste more code? is this fast-register?

Derek Yu

05/22/2023, 4:41 AM

The workflows are registered by running

flytectl register files

map_task

is invoked as such:

Copy code

result = map_task(foo_task, concurrency=5)(
        a = x 
    )

in a workflow defined in directory

flyte/a/b/tasks

foo_task

is defined in a file in the same directory

flyte/a/b/tasks

The failing command

pyflyte-map-execute

in the logs has

--input

'task-module'

flyte.a.b.tasks

Here is the structure of the error message:

Copy code

CalledProcessError: Command '['pyflyte-map-execute', '--inputs',
'xxx', '--output-prefix',
'xxx', '--raw-output-data-prefix',
'xxx',
'--checkpoint-path',
'xxx',
'--prev-checkpoint', '""', '--dynamic-addl-distro',
'xxx', '--dynamic-dest-dir', '.',
'--resolver', 'MapTaskResolver', '--', 'vars', 'resolver',
'flytekit.core.python_auto_container.default_task_resolver', 'task-module',
'flyte.a.b.tasks', 'task-name', 'foo_task']'

Thanks Yee, let me know if there's anything else I can provide that would be helpful to troubleshoot!

Samhita Alla

05/22/2023, 5:26 AM

I think the task module should be

flyte.a.b.tasks.<module>

. Is it possible for you to share the directory structure and the serialization & registration commands you're running?

Derek Yu

05/22/2023, 6:50 AM

Thanks Samhita, here's the dir structure: workflow defined in

flyte/a/b/tasks/workflows.py

and the workflow calls a map_task():

Copy code

@workflow
def foo_workflow(): 
    result = map_task(foo_task, concurrency=5)(
        a = x 
    )

task is defined in

flyte/a/b/tasks/tasks.py

Serialization/registration commands:

Copy code

pyflyte -k a -k flyte.a serialize --image xxx workflows -f /tmp/workflows \
	&& flytectl --admin.endpoint dns:///xxx register files /tmp/workflows/* -p foo -d bar --version='xxx' \

Samhita Alla

05/22/2023, 9:29 AM

Can you try

pyflyte package

instead of

pyflyte serialize

? https://docs.flyte.org/projects/flytekit/en/latest/pyflyte.html#pyflyte-package https://docs.flyte.org/projects/cookbook/en/latest/getting_started/package_register.html#package-your-project-with-pyflyte-package

Heidi Hurst

05/22/2023, 10:20 AM

FYI dropping back down to Flytekit 1.4.2 resolved this issue, so I suspect this is something that's either broken or changed with the recent releases (the above error was encountered with flytekit 1.6.1).

Samhita Alla

05/23/2023, 5:04 AM

cc @Eduardo Apolinario (eapolinario)

Eduardo Apolinario (eapolinario)

05/23/2023, 9:59 PM

@Derek Yu, @Heidi Hurst, I'm having trouble reproing this. Can you give us more details? For example, can you show the dir structure (by running

tree

at the root of the repo)? Also, what's the type of

foo_workflow

Heidi Hurst

05/24/2023, 5:54 PM

Sure! Thanks for following up @Eduardo Apolinario (eapolinario) 🙂 I've been able to construct a small example which shows the same error: Files:

flyte/a/b/workflows.py

Copy code

@workflow
def repro_issue_workflow() -> List[int]:
    foo = foo_task()
    bar = map_task(bar_task, concurrency=3)(number=foo)
    return bar

flyte/a/b/tasks.py

Copy code

@task
def foo_task() -> List[int]:
    return [1, 2, 3]

@task
def bar_task(number: int) -> int:
    return number + 1

In this example, the type of data in the map task is

List[int]

. I've confirmed that this fails with both regular registration and fast registration. This also seems to fail when both the tasks and workflows are in the same file 🤔 In the screenshot (included) you can also see that the map task has a different naming convention than I would have expected from past flytekit versions as well, although that might be a red herring. Happy to open an issue or file a bug report if that would be helpful 🙏

165 Views

Open in Slack

Previous Next