[k8s] Update the instance type format to support number only accelerator type

[DanielZhangQD]

Update the instance type format for Kubernetes to support number-only accelerator type, which is reported in #4608.

Tested (run the relevant ones):

• Code formatting: bash format.sh
• Any manual or new tests for this PR (please specify below)
  • Number-only accelerator type test

✗ python -m sky.cli launch -y -c mygpucluster --cloud k8s --gpus 4090:1 --cpus 1 --memory 6 --dryrun -- "nvidia-smi"

Command to run: nvidia-smi
Running on cluster: mygpucluster
D 02-20 10:15:11 optimizer.py:301] #### Task<name=sky-cmd>(run='nvidia-smi')
D 02-20 10:15:11 optimizer.py:301]   resources: Kubernetes(cpus=1, mem=6, {'4090': 1}) ####
D 02-20 10:15:11 optimizer.py:316] Defaulting the task's estimated time to 1 hour.
D 02-20 10:15:11 optimizer.py:338] resources: Kubernetes(1CPU--6GB--4090:1, cpus=1, mem=6, {'4090': 1})
D 02-20 10:15:11 optimizer.py:357]   estimated_runtime: 3600 s (1.0 hr)
D 02-20 10:15:11 optimizer.py:361]   estimated_cost (not incl. egress): $0.0
I 02-20 10:15:11 optimizer.py:884] Considered resources (1 node):
I 02-20 10:15:11 optimizer.py:952] -------------------------------------------------------------------------------------------------------
I 02-20 10:15:11 optimizer.py:952]  CLOUD        INSTANCE            vCPUs   Mem(GB)   ACCELERATORS   REGION/ZONE     COST ($)   CHOSEN
I 02-20 10:15:11 optimizer.py:952] -------------------------------------------------------------------------------------------------------
I 02-20 10:15:11 optimizer.py:952]  Kubernetes   1CPU--6GB--4090:1   1       6         4090:1         kind-skypilot   0.00          ✔
I 02-20 10:15:11 optimizer.py:952] -------------------------------------------------------------------------------------------------------
D 02-20 10:15:11 cloud_vm_ray_backend.py:4596] cluster_ever_up: False
D 02-20 10:15:11 cloud_vm_ray_backend.py:4597] record: None
D 02-20 10:15:11 utils.py:54] more than 1 credential file found
D 02-20 10:15:11 backend_utils.py:667] Using ssh_proxy_command: None
I 02-20 10:15:11 execution.py:323] Dryrun finished.

• Letter prefixed accelerator type (for backward compatibility)

✗ python -m sky.cli launch -y -c mygpucluster --cloud k8s --gpus K80:1 --cpus 1 --memory 6 --dryrun -- "nvidia-smi"

Command to run: nvidia-smi
Running on cluster: mygpucluster
D 02-20 10:14:38 optimizer.py:301] #### Task<name=sky-cmd>(run='nvidia-smi')
D 02-20 10:14:38 optimizer.py:301]   resources: Kubernetes(cpus=1, mem=6, {'K80': 1}) ####
D 02-20 10:14:38 optimizer.py:316] Defaulting the task's estimated time to 1 hour.
D 02-20 10:14:38 optimizer.py:338] resources: Kubernetes(1CPU--6GB--K80:1, cpus=1, mem=6, {'K80': 1})
D 02-20 10:14:38 optimizer.py:357]   estimated_runtime: 3600 s (1.0 hr)
D 02-20 10:14:38 optimizer.py:361]   estimated_cost (not incl. egress): $0.0
I 02-20 10:14:38 optimizer.py:884] Considered resources (1 node):
I 02-20 10:14:38 optimizer.py:952] ------------------------------------------------------------------------------------------------------
I 02-20 10:14:38 optimizer.py:952]  CLOUD        INSTANCE           vCPUs   Mem(GB)   ACCELERATORS   REGION/ZONE     COST ($)   CHOSEN
I 02-20 10:14:38 optimizer.py:952] ------------------------------------------------------------------------------------------------------
I 02-20 10:14:38 optimizer.py:952]  Kubernetes   1CPU--6GB--K80:1   1       6         K80:1          kind-skypilot   0.00          ✔
I 02-20 10:14:38 optimizer.py:952] ------------------------------------------------------------------------------------------------------
D 02-20 10:14:38 cloud_vm_ray_backend.py:4596] cluster_ever_up: False
D 02-20 10:14:38 cloud_vm_ray_backend.py:4597] record: None
D 02-20 10:14:38 utils.py:54] more than 1 credential file found
D 02-20 10:14:38 backend_utils.py:667] Using ssh_proxy_command: None
I 02-20 10:14:38 execution.py:323] Dryrun finished.

• All smoke tests: pytest tests/test_smoke.py
• Relevant individual smoke tests: pytest tests/test_smoke.py::test_fill_in_the_name
• Backward compatibility tests: conda deactivate; bash -i tests/backward_compatibility_tests.sh

[DanielZhangQD]

/smoke-test --kubernetes

[DanielZhangQD]

Hi @Michaelvll, please help review this PR, thanks!

[Michaelvll]

/smoke-test --kubernetes

[Michaelvll]

Thanks for fixing this @DanielZhangQD ! Could we add a unit test for this newly added code as well?

[DanielZhangQD]

Hi @Michaelvll, thanks for the reminder!
Sorry that I wrote UT cases for the regex patterns in a private file but did not migrate to the repo, and I have added UT cases with the following result:

✗ pytest -v --kubernetes tests/unit_tests/kubernetes/test_instance_type.py
========================================================================================== test session starts ==========================================================================================
platform darwin -- Python 3.11.11, pytest-8.3.4, pluggy-1.5.0
rootdir: /Users/matrixorigin/dan/go/src/github.com/skypilot-org/skypilot
configfile: pyproject.toml
plugins: env-1.1.5, time-machine-2.16.0, anyio-4.7.0, xdist-3.6.1
16 workers [1 item]
.
=========================================================================================== 1 passed in 5.32s ===========================================================================================

PTAL again, thanks!

[Michaelvll]

Thanks @DanielZhangQD! This is awesome! LGTM.