[sky/feat/show-gpus] adding region based filtering for show-gpus command.

[vivekkhimani]

Acceptance Criteria

This PR aims to add the features proposed in #1170. Based on the feedback received from the discussion in the issue, the following changes are introduced:

• Add support for a --region flag on the show-gpus command.
  • The flag is only valid if the cloud flag is provided.
  • As we expect different cloud providers to have different naming conventions for the regions, the --region argument will be invalidated if it can't be found in the target provider.
• Add/update the relevant test cases.
• Update the show-gpus documentation to include the --region flag and the instructions to use it.

[vivekkhimani]

@romilbhardwaj can you help me approve the workflows blocked for the first-time contributors?

[romilbhardwaj]

Done, should be running now!

[vivekkhimani]

Should be ready for review contingent on the fact that the workflows pass! @romilbhardwaj

[romilbhardwaj]

Great work @vivekkhimani! Left some comments and notes on unintended behavior I observed.

[romilbhardwaj]

It would be great to have hints on regions supported by a cloud if the user enters incorrect regions.

For instance, for small typos, I noticed we have automatic suggestions which is great:

(base) romilb@romilbx1yoga:~$ sky show-gpus --region us-west1 --cloud aws
ValueError: Invalid region 'us-west1'
Did you mean one of these: 'us-west-1'?

However, for incorrectly specified regions (severe typos? :) ), the error message isn't very helpful:

(base) romilb@romilbx1yoga:~$ sky show-gpus --region uswest --cloud azure
ValueError: Invalid region 'uswest'

Can we have a line like List of supported azure regions: westus, <list of azure regions> here?

[romilbhardwaj]

Problem

I noticed that we added REGION column to sky show-gpus -a makes the output seem like SkyPilot supports only one region.

For instance, we know that V100 is available in aws us-east-1, aws us-east-2 and aws us-west-2. This is also recorded in our catalog (L4884-L4896 in aws.csv).

However, when I run sky show-gpus -a, the addition of the region column makes it seem SkyPilot supports only us-east-1 on aws:

GPU   QTY  CLOUD  INSTANCE_TYPE       vCPUs  HOST_MEMORY  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGION       
V100  1    AWS    p3.2xlarge          8      61GB         $ 3.060       $ 0.918            us-east-1    
V100  4    AWS    p3.8xlarge          32     244GB        $ 12.240      $ 3.672            us-east-1    
V100  8    AWS    p3.16xlarge         64     488GB        $ 24.480      $ 7.344            us-east-1    
V100  1    Azure  Standard_NC6s_v3    6      112GB        $ 3.060       $ 0.846            centralus    
V100  2    Azure  Standard_NC12s_v3   12     224GB        $ 6.120       $ 1.693            centralus    
V100  4    Azure  Standard_NC24rs_v3  24     448GB        $ 13.460      $ 3.724            centralus    
V100  4    Azure  Standard_NC24s_v3   24     448GB        $ 12.240      $ 3.386            centralus    
V100  1    GCP    (attachable)        -      -            $ 2.480       $ 0.740            us-central1  
V100  2    GCP    (attachable)        -      -            $ 4.960       $ 1.480            us-central1  
V100  4    GCP    (attachable)        -      -            $ 9.920       $ 2.960            us-central1
V100  8    GCP    (attachable)        -      -            $ 19.840      $ 5.920            us-central1  

Of course, we don't want to show all regions in a new line in sky show-gpus -a since that would result in a really long output.

Proposed solution

• We do not show regions when the user specifies sky show-gpus -a option.
• We show REGION column only when both a cloud and the gpu_name are specified or --region flag is used. We then print each region in new line. E.g., when I run sky show-gpus V100 --cloud aws:

(base) romilb@romilbx1yoga:~$ sky show-gpus V100 --cloud aws
*NOTE*: for most GCP accelerators, INSTANCE_TYPE == (attachable) means the host VM's cost is not included.

GPU   QTY  CLOUD  INSTANCE_TYPE  vCPUs  HOST_MEMORY  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGIONS     
V100  1    AWS    p3.2xlarge     8      61GB         $ 3.060       $ 0.918            us-east-1
V100  1    AWS    p3.2xlarge     8      61GB         $ 3.060       $ 0.918            us-west-2
V100  4    AWS    p3.8xlarge     32     244GB        $ 12.240      $ 3.672            us-east-1
V100  4    AWS    p3.8xlarge     32     244GB        $ 12.240      $ 3.672            us-west-2
V100  8    AWS    p3.16xlarge    64     488GB        $ 24.480      $ 7.344            us-east-1
V100  8    AWS    p3.16xlarge    64     488GB        $ 24.480      $ 7.344            us-west-2
<show more regions, currently we don't>

GPU        QTY  CLOUD  INSTANCE_TYPE  vCPUs  HOST_MEMORY  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGION     
V100-32GB  8    AWS    p3dn.24xlarge  96     768GB        $ 31.212      $ 9.364            us-east-1  
<show more regions, currently we don't>

Only problem is this will likely be a long output with lots of similar lines, but it does solve the pain point in #1170. @Michaelvll please feel free to suggest any other way you'd like to see per region pricing information.

[vivekkhimani]

@romilbhardwaj thanks for the review! All of these comments make sense and I have a pretty good idea to fix them. I have been caught up w some school work but I will get to this by end of this week.

[romilbhardwaj]

Hi @vivekkhimani - any updates on this?

[vivekkhimani]

It would be great to have hints on regions supported by a cloud if the user enters incorrect regions.

For instance, for small typos, I noticed we have automatic suggestions which is great:

(base) romilb@romilbx1yoga:~$ sky show-gpus --region us-west1 --cloud aws
ValueError: Invalid region 'us-west1'
Did you mean one of these: 'us-west-1'?

However, for incorrectly specified regions (severe typos? :) ), the error message isn't very helpful:

(base) romilb@romilbx1yoga:~$ sky show-gpus --region uswest --cloud azure
ValueError: Invalid region 'uswest'

Can we have a line like List of supported azure regions: westus, <list of azure regions> here?

Added a fix for this @romilbhardwaj

---

Problem

I noticed that we added REGION column to sky show-gpus -a makes the output seem like SkyPilot supports only one region.

For instance, we know that V100 is available in aws us-east-1, aws us-east-2 and aws us-west-2. This is also recorded in our catalog (L4884-L4896 in aws.csv).

However, when I run sky show-gpus -a, the addition of the region column makes it seem SkyPilot supports only us-east-1 on aws:

GPU   QTY  CLOUD  INSTANCE_TYPE       vCPUs  HOST_MEMORY  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGION       
V100  1    AWS    p3.2xlarge          8      61GB         $ 3.060       $ 0.918            us-east-1    
V100  4    AWS    p3.8xlarge          32     244GB        $ 12.240      $ 3.672            us-east-1    
V100  8    AWS    p3.16xlarge         64     488GB        $ 24.480      $ 7.344            us-east-1    
V100  1    Azure  Standard_NC6s_v3    6      112GB        $ 3.060       $ 0.846            centralus    
V100  2    Azure  Standard_NC12s_v3   12     224GB        $ 6.120       $ 1.693            centralus    
V100  4    Azure  Standard_NC24rs_v3  24     448GB        $ 13.460      $ 3.724            centralus    
V100  4    Azure  Standard_NC24s_v3   24     448GB        $ 12.240      $ 3.386            centralus    
V100  1    GCP    (attachable)        -      -            $ 2.480       $ 0.740            us-central1  
V100  2    GCP    (attachable)        -      -            $ 4.960       $ 1.480            us-central1  
V100  4    GCP    (attachable)        -      -            $ 9.920       $ 2.960            us-central1
V100  8    GCP    (attachable)        -      -            $ 19.840      $ 5.920            us-central1  

Of course, we don't want to show all regions in a new line in sky show-gpus -a since that would result in a really long output.

Proposed solution

• We do not show regions when the user specifies sky show-gpus -a option.
• We show REGION column only when both a cloud and the gpu_name are specified or --region flag is used. We then print each region in new line. E.g., when I run sky show-gpus V100 --cloud aws:

(base) romilb@romilbx1yoga:~$ sky show-gpus V100 --cloud aws
*NOTE*: for most GCP accelerators, INSTANCE_TYPE == (attachable) means the host VM's cost is not included.

GPU   QTY  CLOUD  INSTANCE_TYPE  vCPUs  HOST_MEMORY  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGIONS     
V100  1    AWS    p3.2xlarge     8      61GB         $ 3.060       $ 0.918            us-east-1
V100  1    AWS    p3.2xlarge     8      61GB         $ 3.060       $ 0.918            us-west-2
V100  4    AWS    p3.8xlarge     32     244GB        $ 12.240      $ 3.672            us-east-1
V100  4    AWS    p3.8xlarge     32     244GB        $ 12.240      $ 3.672            us-west-2
V100  8    AWS    p3.16xlarge    64     488GB        $ 24.480      $ 7.344            us-east-1
V100  8    AWS    p3.16xlarge    64     488GB        $ 24.480      $ 7.344            us-west-2
<show more regions, currently we don't>

GPU        QTY  CLOUD  INSTANCE_TYPE  vCPUs  HOST_MEMORY  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGION     
V100-32GB  8    AWS    p3dn.24xlarge  96     768GB        $ 31.212      $ 9.364            us-east-1  
<show more regions, currently we don't>

Only problem is this will likely be a long output with lots of similar lines, but it does solve the pain point in #1170. @Michaelvll please feel free to suggest any other way you'd like to see per region pricing information.

@romilbhardwaj very true! fixed this by disabling the region column on -a option.

[vivekkhimani]

Hi @vivekkhimani - any updates on this?

@romilbhardwaj extremely sorry for the huge delay! I got busy with school and my own project because of a conference deadline and then was traveling during the winter break. I fixed/resolved all the comments and most of the pylint/yapf issues. Can you give me a maintainer approval to run the pipelines again?

[vivekkhimani]

@romilbhardwaj supporting screenshots for the fix.

• hints on supported regions:

- regions column disabled on ```-a``` flag:

[romilbhardwaj]

Thanks @vivekkhimani! I'll get to this soon.

[romilbhardwaj]

Thanks @vivekkhimani - I tested this out and it works nicely (including for gcp TPUs)! Left some minor nit comments - should be ready to ship once resolved!

[romilbhardwaj]

Awesome work @vivekkhimani! This looks good to go!

[vivekkhimani]

Thanks for reviewing it all along, @romilbhardwaj ! Don't think I can merge this as I don't have the write access to repository so do you mind doing it for me?