[artemy-kolesnikov]

Title: air: Use the run index for a self user filter and bound the index fallback in air list

Changes

• --all-status --filter user=<own email> now uses the AiTrainingService index. air list resolved the current user only when no user= filter was set. So filtering on your own email never matched currentUser, and the command scanned the workspace's full Jobs run history. With --all-status, the current user is now resolved even when an explicit user= filter is given, and a case-insensitive match counts as self. If that lookup fails, the command keeps the old scan rather than failing. The default active view still makes no extra call.
• The index fallback is visible and bounded. If the index load fails, the CLI now logs a Warn line instead of a Debug one (Warn is the CLI's default log level, so users see it on stderr). The line names the cause and says the CLI is falling back to a slower scan. That replacement scan can't use active_only, so it walks the full history. It is now capped at maxFallbackListScan = 250 runs instead of maxListScan = 2000. Hitting either cap warns "results may be incomplete. Narrow the search with --filter or a smaller --limit". The interactive table now gives the same warning when it exits; before, only one-shot output did.
• Index requests fail fast. The index client is built from w.Config with HTTPTimeout = 15s and RetryTimeout = 15s, following the dedicated-client pattern in validateconfig.go. Before, they used the SDK defaults: a 60 s inactivity timeout and a 5 min retry window. The global SDK config is unchanged.
• -o json stdout is unchanged. All new messages go to stderr.

Why

A report against a workspace with a large run history had air list take minutes to return a small number of rows. The traces showed each call walking every full-history runs/list page (several seconds per page), with no ai-training/workflows index request and no scim/Me call from the list path: the index was never attempted, because the user had passed --filter user=... or --all-users. A self filter is the case this PR fixes. Manually verified against a workspace: the same self-filtered --all-status list drops from minutes to a few seconds, with identical rows.

Tests

• cmd/air/list_index_test.go:
  • TestListSelfUserFilterUsesIndex runs the full command (--all-status --filter user=me@…, JSON) against a stub server. It asserts one index call, zero runs/list calls, and the run in stdout. It fails on main: 0 index calls, 1 scan, empty result.
  • TestFallbackScanWarnsAndIsBounded: the index returns 500 and runs/list serves endless non-AIR pages. It asserts exactly 250/25 = 10 pages, no active_only, both warnings logged, and empty stdout.
  • TestRegularScanKeepsFullCap: --all-users still scans up to 2000 runs, with no fallback warning.
  • TestNewListStrategyGate: two new cases, self filter in a different case → index, and unknown current user → scan.
  • TestNewListStrategyFallsBackWhenIndexFails now uses a cmdio context. Its assertions are unchanged.
• New acceptance fixture acceptance/air/list-index-fallback: the index returns 500. The text output shows the Warn line, and the JSON output (stderr dropped) is unchanged. No existing acceptance/air/list* golden changed: the new line appears only when the index fails, and no existing fixture fails it.
• go test ./cmd/air/... passes. go test ./acceptance -run TestAccept/air/ passes (21 tests). gofmt and go vet are clean.

Follow-ups (not in this PR)

• AICM / ai-training: --all-status walks the caller's whole index without a bound (63 pages for a heavy user, about 4 s). The index has no server-side ordering or limit. handleListTrainingWorkflows also ignores active_only, so the default active view can't use the index. Both need a server-side state filter plus an ordered, limited index.
• --all-users --all-status is still a full-history scan, capped at 2000 runs. That is inherent to Jobs runs/list (several seconds per page on a busy workspace).
• Python air CLI parity was skipped: the legacy CLI is frozen.

This pull request and its description were written by Isaac.

[github-actions]

Commit: ba33c5ce2 | 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 (c4559ed)	latest (v1.26.0)
TestAccept/air/list-index-fallback	✅	❌	❌

TestAccept/air/list-index-fallback: branch ✅ | main (c4559ed) ❌ | latest (v1.26.0) ❌

main (c4559ed):

acceptance_test.go:1374: Diff:
        --- air/list-index-fallback/output.txt
        +++ /tmp/TestAcceptairlist-index-fallback2319011935/002/output.txt
        @@ -2,7 +2,6 @@
         === list --all-status falls back to a bounded Jobs scan when the index fails (text)
         
         >>> [CLI] air list --all-status
        -Warn: air list: the AI Runtime run index is unavailable (failed to list training workflows: index unavailable); falling back to a slower scan of up to 250 recent job runs
            Run ID           Experiment  Status     Started              Duration  Progress  MLflow                  User                   Accelerators
            334747067049496  qwen-train  ● SUCCESS  [TIMESTAMP]       12s         -  qwen-train-001          [USERNAME]  8x H100
         
        
FAIL

latest (v1.26.0):

acceptance_test.go:1374: Diff:
        --- air/list-index-fallback/output.txt
        +++ /tmp/TestAcceptairlist-index-fallback361912709/002/output.txt
        @@ -2,7 +2,6 @@
         === list --all-status falls back to a bounded Jobs scan when the index fails (text)
         
         >>> [CLI] air list --all-status
        -Warn: air list: the AI Runtime run index is unavailable (failed to list training workflows: index unavailable); falling back to a slower scan of up to 250 recent job runs
            Run ID           Experiment  Status     Started              Duration  Progress  MLflow                  User                   Accelerators
            334747067049496  qwen-train  ● SUCCESS  [TIMESTAMP]       12s         -  qwen-train-001          [USERNAME]  8x H100
         
        @@ -18,12 +17,7 @@
                 "user": "[USERNAME]",
                 "status": "SUCCESS",
                 "started_at": "[TIMESTAMP]",
        -        "is_sweep": false,
        -        "experiment_name": "qwen-train",
        -        "duration_seconds": 12,
        -        "accelerator_type": "GPU_8xH100",
        -        "num_accelerators": 8,
        -        "dashboard_url": "[DATABRICKS_URL]/jobs/runs/334747067049496?o=[WORKSPACE_ID]"
        +        "is_sweep": false
               }
             ]
           }
        
FAIL

[eng-dev-ecosystem-bot]

✅ Integration test report

Commit: ba33c5c

Run: 43529936109

	Env	💚​RECOVERED	✅​pass	🙈​skip	Time
💚​	aws linux-2core-8gb	1	270	16	6:15
💚​	aws-windows-latest-4core-16gb	1	272	14	5:19
💚​	azure linux-2core-8gb	1	269	16	5:21
💚​	azure-windows-latest-4core-16gb	1	271	14	3:31
💚​	gcp linux-2core-8gb	1	270	16	4:52
💚​	gcp-windows-latest-4core-16gb	1	272	14	3:53

	Test Name	aws linux-2core-8gb	aws-windows-latest-4core-16gb	azure linux-2core-8gb	azure-windows-latest-4core-16gb	gcp linux-2core-8gb	gcp-windows-latest-4core-16gb
💚​	TestAccept	💚​R	💚​R	💚​R	💚​R	💚​R	💚​R

Top 3 slowest tests (at least 2 minutes):

duration	env	testname
3:27	azure-windows-latest-4core-16gb	TestAccept
3:25	gcp-windows-latest-4core-16gb	TestAccept
3:10	aws-windows-latest-4core-16gb	TestAccept