Zhanghao Wu
Zhanghao Wu
This PR updates the docs for managed spot jobs with auto-translation added in #1081. Tested: - renderred locally.
Upgrade our backend ray version to 2.0.0. Backward compatibility requirement: 1. UP clusters require `sky launch` again to upgrade the packages on the cluster. Tested: - [x] `tests/run_smoke_tests.sh` - [x]...
Reproducible code: ``` sky gpunode --gpus K80:4 --region us-central1 --cloud gcp -c test-gpu sky exec --gpus K80:3 --region us-central1 test-gpu 'echo hi' ``` It raises: ``` Task from command: echo...
We are using the default disk type for all the clouds, and the price should be fixed for all the regions in each cloud. We can have a dict for...
We can check the seaweedfs to see if it has better performance and consistency guarantee than our current goofys+gcsfuse soltuion. https://github.com/seaweedfs/seaweedfs
@lhqing mentioned sharing the `~/.sky` across multiple machine is very useful, and he has successfully done that with placing it on the Google Drive. It works well for all stopped...
Currently, the buckets generated by the auto file mounts translation will be leaked if the spot job is not submitted correctly, as we rely on `persistent: False` option in the...
Currently, when we ctrl-c from `sky spot launch` the hint for canceling jobs should be `sky spot cancel`, but the current output is as following: ``` I 10-04 15:56:40 cloud_vm_ray_backend.py:2080]...
I did `sky launch -c vision-azure 'some commands'`, `sky launch -c vision-azure 'some commands'` and later `sky stop vision-azure` without error. The dashboard shows there are two `vision-azure` running.
It would be nice to show prices on different region/zones instead of per cloud based.