Zhanghao Wu

> Is there a potential issue where, say K80:4 is the only K80 config available in a region, and users do `sky launch --gpus K80:3 --region ...`. In this case,...

> However, I have a bit fundamental question: do we really need to call `accelerator_in_region_or_zone` (or any other validation check) for `sky exec`? To my understanding, comparing the user's resource...

> For now, I did the following in VM with the file mount supported by `gcsfuse`. And I can see all my files with `file_mounts` > > ```python > bucket_name...

Also requested by @lhqing .

I think the main reason for the symlink was to prevent some permission issues. Maybe @concretevitamin can say more about it. Can we use hard links instead? I think the...

> Add all the ssh config information into an external file under ~/.sky and import it in the ~/.ssh/config, instead of directly adding them in the ~/.ssh/config Hey @michaelzhiluo, do...

Hey @pounde, the problem should be fixed in #1261. Please feel free to try it out, when you get time, by installing skypilot from the master branch with ``` git...

Hey @vivekkhimani, it would be great if you want to take a look at this! I agree with @romilbhardwaj's comment that we may want to start from that the `--region...

Question: If the optimizer will yield the list of fine-grained resources to be tried, why do we still need the following lines in the `cloud_vm_ray_backend.py`? https://github.com/skypilot-org/skypilot/blob/6b53f1765839b77fcb0b39ab36b965b5bb0fad72/sky/backends/cloud_vm_ray_backend.py#L925-L936

After this PR, the failover message looks like the following (with the proposed fix for `should_be_blocked_by`): ``` I 12-08 20:41:52 optimizer.py:606] == Optimizer == I 12-08 20:41:52 optimizer.py:617] Target: minimizing...