Skip to content

[https://nvbugs/6600098][test] Waive flaky TestKVCacheV2Llama scheduler tests on DGX_B200 - #17577

Merged
brnguyen2 merged 1 commit into
NVIDIA:mainfrom
brnguyen2:waive-kvcachev2-b200-flaky
Aug 12, 2026
Merged

[https://nvbugs/6600098][test] Waive flaky TestKVCacheV2Llama scheduler tests on DGX_B200#17577
brnguyen2 merged 1 commit into
NVIDIA:mainfrom
brnguyen2:waive-kvcachev2-b200-flaky

Conversation

@brnguyen2

@brnguyen2 brnguyen2 commented Aug 12, 2026

Copy link
Copy Markdown
Collaborator

Description

SKIP-waive four chronically-flaky TestKVCacheV2Llama KV-cache-v2 scheduler
integration tests in the pre-merge L0 DGX_B200 stage. They fail
intermittently independent of the PR under test, so they redden unrelated PRs
and — via fail-fast — cost a full GPU-matrix rerun each.

Flake rates over the last 30 days, measured across all of main CI (not one
PR) via the infra stability report:

Test flake rate distinct MRs distinct failing users
test_overlap_scheduler[overlap] 5.3% (104/1954) 593 13
test_token_budget_limited 5.2% (102/1978) 597 11
test_overlap_scheduler[non_overlap] 4.9% (94/1913) 592 10
test_eviction[cuda_graph] 2.1% (39/1845) 589 8

Failing users span the team; all failures land on DGX_B200 with a mix of
Test terminated unexpectedly / CUDA OOM / assertion — a shared B200-side
instability, not a single bug. Two siblings in the same class are already waived
(test_chunked_prefill → nvbugs/6428002, test_eviction_with_block_reuse
nvbugs/6462303).

The waived tests keep running post-merge, so regression coverage is retained
— this only takes them off the pre-merge critical path until the underlying
instability is fixed. test_eviction[no_cuda_graph] (1.7%) is left pre-merge for now.

Tracked under umbrella bug nvbugs/6600098 (links the two existing P0s).

Test Coverage

Test-list change only. pre-commit passed the waives.txt duplicate check and the
AST check that every waived node-id exists in the test DB.

PR Checklist

  • Commit messages follow [NVBUG][type] description, DCO signed-off.
  • Tests remain covered post-merge; only pre-merge gating changes.

Dev Engineer Review

  • Added four waiver entries for TestKVCacheV2Llama under nvbugs/6600098.
  • Waived:
    • test_overlap_scheduler[overlap]
    • test_token_budget_limited
    • test_overlap_scheduler[non_overlap]
    • test_eviction[cuda_graph]
  • The change is limited to tests/integration/test_lists/waives.txt.
  • The waiver entries preserve post-merge execution.
  • test_eviction[no_cuda_graph] remains in pre-merge testing.
  • The waiver format and bug reference are consistent with the file.
  • No public API, performance, or error-handling changes were made.
  • No duplicate waiver or unintended scope change was reported.

QA Engineer Review

  • No test-db/ or qa/ files were modified.
  • The change adds four pre-merge waivers and does not modify test code.
  • The listed tests remain covered by post-merge execution.
  • Verdict: sufficient.

…er tests on DGX_B200

The KV-cache-v2 scheduler integration tests test_overlap_scheduler[overlap]
(5.3%), test_token_budget_limited (5.2%), test_overlap_scheduler[non_overlap]
(4.9%), and test_eviction[cuda_graph] (2.1%) fail intermittently in the
pre-merge L0 DGX_B200 stage, independent of the PR under test — flake rates
measured over 30 days across ~590 distinct MRs and 8-13 distinct users each
(infra stability report). Error modes are a mix of terminated / CUDA OOM /
assertion, i.e. a shared B200-side instability. Two siblings in the same class
are already waived (test_chunked_prefill nvbugs/6428002,
test_eviction_with_block_reuse nvbugs/6462303).

SKIP-waive the high-flake variants pre-merge under umbrella bug nvbugs/6600098;
they keep running post-merge, so coverage is retained. test_eviction[no_cuda_graph]
(1.7%) is left pre-merge for now.

Signed-off-by: Brian Nguyen <brnguyen@nvidia.com>
@brnguyen2
brnguyen2 marked this pull request as ready for review August 12, 2026 20:00
@brnguyen2

Copy link
Copy Markdown
Collaborator Author

/bot skip --comment "waive only"

@brnguyen2
brnguyen2 enabled auto-merge (squash) August 12, 2026 20:01
@coderabbitai

coderabbitai Bot commented Aug 12, 2026

Copy link
Copy Markdown
Contributor

Review Change Stack

Walkthrough

The integration test waiver list adds skip entries for five KV-cache v2 scheduler tests linked to NV bug 6600098.

Changes

KV-cache v2 scheduler waivers

Layer / File(s) Summary
Add scheduler test waivers
tests/integration/test_lists/waives.txt
Adds skip records for CUDA-graph eviction, non-overlap and overlap scheduling, and token-budget-limited execution.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Possibly related PRs

Suggested reviewers: bowenfu, qijune

🚥 Pre-merge checks | ✅ 5
✅ Passed checks (5 passed)
Check name Status Explanation
Description check ✅ Passed The description explains the four waivers, their cause, scope, retained post-merge coverage, test validation, and checklist status.
Title check ✅ Passed The title clearly identifies the NVBugs ticket, test-related change, flaky scheduler tests, and DGX_B200 scope.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests

Comment @coderabbitai help to get the list of available commands.

@coderabbitai coderabbitai Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/integration/test_lists/waives.txt`:
- Line 286: Update the waiver entries for TestKVCacheV2Llama eviction tests in
waives.txt to use stage-aware scoping, or move them into a pre-merge-only waiver
file, so they do not suppress DGX_B200 post-merge execution. Apply the same
correction to all four newly added entries.
🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

  • Push a commit to this branch (recommended)
  • Create a new PR with the fixes

ℹ️ Review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 6603b2d4-124b-427c-a162-75f5c0fc729c

📥 Commits

Reviewing files that changed from the base of the PR and between 3612c80 and 3e9df88.

📒 Files selected for processing (1)
  • tests/integration/test_lists/waives.txt

Comment thread tests/integration/test_lists/waives.txt
@brnguyen2

Copy link
Copy Markdown
Collaborator Author

/bot run

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65669 [ skip ] triggered by Bot. Commit: 3e9df88 Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65672 [ run ] triggered by Bot. Commit: 3e9df88 Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65669 [ skip ] completed with state ABORTED. Commit: 3e9df88

Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65672 [ run ] completed with state FAILURE. Commit: 3e9df88
/LLM/main/L0_MergeRequest_PR pipeline #53392 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

  • Please check the failed tests and fix your PR
  • If you cannot view the failures, ask the CI triggerer to share details
  • Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

@brnguyen2

Copy link
Copy Markdown
Collaborator Author

/bot run

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65682 [ run ] triggered by Bot. Commit: 3e9df88 Link to invocation

@brnguyen2

Copy link
Copy Markdown
Collaborator Author

/bot skip --comment "waive only"

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65690 [ skip ] triggered by Bot. Commit: 3e9df88 Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65682 [ run ] completed with state ABORTED. Commit: 3e9df88

Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #65690 [ skip ] completed with state SUCCESS. Commit: 3e9df88
Skipping testing for commit 3e9df88

Link to invocation

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants