mirror of
https://github.com/vale981/ray
synced 2025-03-09 12:56:46 -04:00
28 lines
1.3 KiB
ReStructuredText
28 lines
1.3 KiB
ReStructuredText
Running script
|
|
--------------
|
|
There are 2 workloads. Each workerload requires a different cluster.yaml.
|
|
|
|
Make sure to copy & paste both drivers.
|
|
|
|
Run `unset RAY_ADDRESS; python workloads/streaming_shuffle.py`. Use `cluster.yaml` for this release test.
|
|
Run `unset RAY_ADDRESS; python workloads/dask_on_ray_large_scale_test.py`. Use `dask_on_ray.yaml` for this release test.
|
|
|
|
Note that when you run `dask_on_ray.yaml`, you need to follow the below procedures.
|
|
|
|
```
|
|
ray up dask_on_ray.yaml -y # Start the ray cluster.
|
|
# Wait until the cluster nodes are up. Use `watch ray status` and wait until all worker nodes are up.
|
|
ray down dask_on_ray.yaml -y # After the cluster is up, you should call ray down.
|
|
ray up dask_on_ray.yaml -y
|
|
```
|
|
|
|
This process is required because ulimit is not permitted for images that we are using. Ulimit is necessary for large cluster testing like this.
|
|
Check out https://discuss.ray.io/t/setting-ulimits-on-ec2-instances/590/2 for more details
|
|
|
|
Success Criteria
|
|
----------------
|
|
|
|
For `streaming_shuffle.py`, make sure to include the output string to the release logs.
|
|
|
|
For `dask_on_ray_large_scale_test.py`, make sure the test runs for at least for an hour. This test should succeed, otherwise, it is a release blocker.
|
|
Check out https://github.com/ray-project/ray/pull/14340#discussion_r599271079 to learn the success condition of this test.
|