Enables actor v2
Project description
Union.ai Actors
Developer Setup
Make sure you have a virtual environment set up and activated. Use uv.
Go into the folder this readme is in (cd fasttask/worker-v2 from base of repo). Then,
uv venv --python 3.13
source .venv/bin/activate
To build wheels containing the Rust code you'll also need this for now.
make build-builders to build the builder images
This is because the manylinux by default don't come with rust installed. Installing them into the manylinux
image every time is too time-consuming, so these are really lightweight images that just install rust/cargo/maturin.
We can get rid of these once we find better ways to build multi-arch wheels, or find maintained public images. I'm sure they exist, just haven't looked in depth.
End to end iterating
make build-wheels to build the wheels. This compiles all the Rust code
Do something like this on the Python side
actor_dist_folder = Path("/Users/ytong/go/src/github.com/unionai/flyte/fasttask/worker-v2/dist")
wheel_layer = PythonWheels(wheel_dir=actor_dist_folder, package_name="unionai-reuse")
base = flyte.Image.from_debian_base()
actor_image = base.clone(addl_layer=wheel_layer)
and also run make dist as you would normally when iterating just in the flyte-sdk.
From examples folder of flyte-sdk, using a different virtual environment. Make sure to update the wheel folder in the
example code.
flyte -c ~/.flyte/config-k3d.yaml run -d development reuse/oomer_reuse.py failure_recovery
Note when hitting Ctrl-C during the make command, the docker process doesn't stop. You'll have to docker stop/rm the container if you don't want to wait around for the build to finish.
Executor iteration
To make a change to the python side
- Just make sure that the flyte-sdk library is installed in editable mode and changes should get picked up. Test with a print statement to confirm.
To make a change to the rust side These instructions may be incomplete
Run the test server so that the executor has something to attach to.
cargo run -p unionai_actor_bridge --bin test_server
Another test service is available called ping_test - it doesn't ping, it calls devbox one.
cargo run -p unionai_actor_bridge --bin ping_test
cargo run -p unionai_actor_bridge --bin ping_test -- --test-cancel --cancel-wait 10
Run the executor (see below for building)
unionai-actor-executor --executor-registration-addr 127.0.0.1:15606 --id 0 --num-workers 1
This adds local flyte-sdk changes to the path. Also adds the examples folder to the path so that the manual override added (need to document), can be picked up.
When using this, you'll also need to pull this test branch in flyte-sdk, which helps short-circuit part of the task execution process. https://github.com/flyteorg/flyte-sdk/compare/rusty-test-abort?expand=1
PYTHONPATH=/Users/ytong/go/src/github.com/flyteorg/flyte-sdk/examples:/Users/ytong/go/src/github.com/flyteorg/flyte-sdk/src: unionai-actor-executor --executor-registration-addr 127.0.0.1:15606 --id 0 --num-workers 1
Will need to uv pip install -e . from base folder (the dir this readme is in) to pick up changes - need to investigate how to run maturin develop from the executor/ folder, which doesn't seem to work currently. Changes don't get pick up.
Bridge iteration
fill this in next time someone is working only on the bridge side.
SDK side
In addition to the test branch above, to get the flyte-sdk to pick up the whls that are build as part of the make build-wheels target,
you can use this Python snippet as the image in the TaskEnvironment
from pathlib import Path
from flyte._image import PythonWheels
actor_dist_folder = Path("/Users/yourusername/go/src/github.com/unionai/flyte/fasttask/worker-v2/dist")
wheel_layer = PythonWheels(wheel_dir=actor_dist_folder, package_name="unionai-reuse")
base = flyte.Image.from_debian_base()
actor_image = base.clone(addl_layer=wheel_layer)
If a test version has been published to test pypi, then you can use it in the task Image with
flyte.Image.from_debian_base().with_pip_packages("unionai-reuse==0.1.8b4", extra_index_urls=["https://test.pypi.org/simple/"])
Releasing
There is some CI in .github/workflows/wheels.yml at the base of this repo, but it is slow. That workflow will build and
publish wheels to pypi and test pypi based on the following:
To trigger the workflow, create and push a tag (or create on GitHub) matching unionai_reuse-v*.
- Tags on the
masterbranch publish to PyPI, while tags on feature branches publish to TestPyPI for testing. - Beta/prerelease versions (identified by
b,rc,alpha,beta, ordevin the version string, e.g.,unionai_reuse-v0.1.8b0) pushed frommasterwill also publish to TestPyPI. - The workflow builds wheels for Linux (x86_64 and aarch64) and macOS platforms.
To tag locally, run
git tag unionai_reuse-v0.1.9b0
git push origin unionai_reuse-v0.1.9b0
However the CI is super, super slow - like running the ARM build takes like 40-50 mins. Can investigate ARM machines in the future.
The below commands use the local commands and override the python version. This is the better option if you're in a hurry.
Building for amd64 on a macbook arm is much faster, at least on my M4 I'm typing this on.
The make targets are needed anyways in local development so we'll definitely keep these around.
make build-wheels SETUPTOOLS_SCM_PRETEND_VERSION=0.1.7a0b0
SETUPTOOLS_SCM_PRETEND_VERSION=0.1.7a0 python -m build --wheel
but wheels need to be renamed to this pattern
mv unionai_reuse-0.1.7a0-cp38-abi3-linux_aarch64.whl unionai_reuse-0.1.7a0-cp38-abi3-manylinux_2_28_aarch64.whl;
mv unionai_reuse-0.1.7a0-cp38-abi3-linux_x86_64.whl unionai_reuse-0.1.7a0-cp38-abi3-manylinux_2_28_x86_64.whl
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distributions
Built Distributions
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file unionai_reuse-0.1.10-cp38-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.
File metadata
- Download URL: unionai_reuse-0.1.10-cp38-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 2.1 MB
- Tags: CPython 3.8+, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
eaaec24b7d5836fd7d5cae588b519682d751c1127e115c65de4470803cb34396
|
|
| MD5 |
f92b6ef42902d8de693e6ba125aa3a88
|
|
| BLAKE2b-256 |
ba34beafa6839b79ea6a814ec1c24308e56ff3f03f442c69bd2c61dab9a36d15
|
File details
Details for the file unionai_reuse-0.1.10-cp38-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl.
File metadata
- Download URL: unionai_reuse-0.1.10-cp38-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
- Upload date:
- Size: 2.1 MB
- Tags: CPython 3.8+, manylinux: glibc 2.17+ ARM64
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
9d32a22d2903b780d8b18e119d5325a645e3483e6cc708f85cfeddc0d8ecba58
|
|
| MD5 |
23a8d0e45fe513387d1924c8f6a86ede
|
|
| BLAKE2b-256 |
7888192b6412fc7596e7e4bfbf1a347cecb84f9951afc0d2ac68482a040cff1f
|
File details
Details for the file unionai_reuse-0.1.10-cp38-abi3-macosx_11_0_arm64.whl.
File metadata
- Download URL: unionai_reuse-0.1.10-cp38-abi3-macosx_11_0_arm64.whl
- Upload date:
- Size: 1.8 MB
- Tags: CPython 3.8+, macOS 11.0+ ARM64
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
1a8b2b5012ea64b9b13ada61cc5b69557e588c7a0b49b272d897167639935989
|
|
| MD5 |
2262959a3f201b3bf37b0f6fb5387aad
|
|
| BLAKE2b-256 |
1d4f230fcc4db843072387ec39c2b3b5d89119da9739d756678702c6b91e9012
|