[caroline-db]

Changes

• Fetch workspace compute options from the AI Training endpoint for databricks air run -h config.compute.
• Display the options returned by the backend, including per-node accelerator counts and preview stages. The CLI does not apply its own accelerator filter.
• Show an explicit message when the backend returns no options.
• Bound the lookup to three seconds and fall back to the built-in list when the request is unavailable or fails. Unexpected failures also produce a warning.

Why

Static help can advertise accelerators that cannot be used in the selected workspace. The AI Training backend determines which options are supported by both the workspace and the Jobs AI Runtime task API, keeping that compatibility logic out of the CLI.

The fallback preserves useful help while the endpoint is unavailable or still being rolled out.

Tests

• ./task test
• ./task lint
• ./task checks
• Verified successful, empty, unavailable, and failed endpoint responses.
• Verified backend filtering end to end.

[eng-dev-ecosystem-bot]

Integration test report

Commit: efefea9

Run: 39650353516

	Env	🔄​flaky	✅​pass	🙈​skip	Time
✅​	aws linux-2core-8gb		281	16	4:13
🔄​	aws-windows-latest-4core-16gb	2	281	14	3:51
✅​	azure linux-2core-8gb		280	16	4:06
✅​	azure-windows-latest-4core-16gb		282	14	4:48
✅​	gcp linux-2core-8gb		281	16	4:19
🔄​	gcp-windows-latest-4core-16gb	2	281	14	3:52

	Test Name	aws-windows-latest-4core-16gb	gcp-windows-latest-4core-16gb
🔄​	TestFsCpDir	✅​p	🔄​f
🔄​	TestFsCpDir/dbfs_to_uc-volumes	✅​p	🔄​f
🔄​	TestFilerRecursiveDelete	🔄​f	✅​p
🔄​	TestFilerRecursiveDelete/files	🔄​f	✅​p

Top 3 slowest tests (at least 2 minutes):

duration	env	testname
4:45	azure-windows-latest-4core-16gb	TestAccept
3:31	aws-windows-latest-4core-16gb	TestAccept
3:27	gcp-windows-latest-4core-16gb	TestAccept

[pardis-beikzadeh-db]

make sure if any of these are RPC calls that they have retries

[caroline-db]

yup, this uses the databricks sdk client.Do which retries SDK-classified transient failures, and there is a 3-second context timeout here

[pardis-beikzadeh-db]

ditto

[github-actions]

Commit: efefea990 | Workflow run

Regression test report

Selected 1 changed tests (limit=20, 0 not selected)

1 tests: regression=1 unreleased=0 coverage=0 failing=0

test	branch	main (bef6321)	latest (v1.21.0)
TestAccept/air/config-help	✅	❌	❌

TestAccept/air/config-help: branch ✅ | main (bef6321) ❌ | latest (v1.21.0) ❌

main (bef6321):

acceptance_test.go:1307: Diff:
        --- air/config-help/output.txt
        +++ /tmp/TestAcceptairconfig-help1628838081/002/output.txt
        @@ -74,10 +74,6 @@
         config.compute
           Which accelerators to run on and how many.
         
        -  Supported accelerator types:
        -    GPU_1xH100 - 1 accelerator per node [Private Preview]
        -    GPU_8xH100 - 8 accelerators per node
        -
           Fields:
             num_accelerators  Total number of GPUs to allocate.
             accelerator_type  Which accelerator to run on, e.g. GPU_1xA10.
        
FAIL

latest (v1.21.0): same output as main (bef6321)