[CI] Shard hybrid SSM disaggregated accuracy tests - #52354
Open
khluu wants to merge 3 commits into
Open
Conversation
khluu
changed the title
…roups Split hybrid-ssm-nixlconnector-pd-accuracy-tests-4-gpus (~45.7m wall in build 83851) into parallelism: 3 via a new opt-in CONFIG_INDICES env in config_sweep_accuracy_test.sh that selects configs from the chosen array by zero-based index. Unset means run everything, so the other jobs sharing the sweep script (tp/dp_ep/sw_attn variants and the AMD mirrors) are unchanged. CONFIG_EXPECTED_COUNT=7 makes each shard fail loudly if the hybrid config array changes without rebalancing the index lists. Buckets are balanced on measured per-config runtime: shard 0 = configs 0,4 (~14.5m), shard 1 = configs 5,6 (~13.2m), shard 2 = configs 1,2,3 (~16.3m); expected wall ~18m incl. ~1.7m install/canary per shard. Timeout 60 -> 30. The AMD mirror keeps its own unchanged command and is pinned to parallelism: 1 (stays a single unsharded job; requires the pipeline generator change in vllm-project/ci-infra#473). Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
Build 83924 passed 3/3 but the worst shard landed at 19.42m with only
0.58m under the 20m target while measured config-time variance is
~1.4m. Rebalance to parallelism: 4 with index buckets {0} / {4,5} /
{6,1} / {2,3} (per-config walls from 83924: 8.22 / 6.17+6.80 /
6.63+6.30 / 5.58+5.80m -> projected max ~14.7m incl. setup). Also
validate BUILDKITE_PARALLEL_JOB explicitly so an unexpected index fails
instead of passing empty. Union stays {0..6} = 7/7 exactly once with
CONFIG_EXPECTED_COUNT=7 on every shard; AMD mirror still pinned to one
unsharded copy; the shared sweep script is unchanged in this commit.
Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
khluu
force-pushed
the
ci-shard-hybrid-ssm-pd
branch
from
September 1, 2026 10:45
27064f3 to
c8bc2ee
Compare
khluu
changed the title
khluu
marked this pull request as ready for review
September 1, 2026 22:27
khluu
requested review from
ApostaC,
Harry-Chen,
ivanium and
orozery
as code owners
September 1, 2026 22:27
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Claude Code Review
This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.
Tip: disable this comment in your organization's Code Review settings.
Derive each shard from the Buildkite shard id and count so configuration additions are covered automatically and the YAML does not need manually balanced index lists. Co-authored-by: OpenAI Codex <codex@openai.com> Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
Member
Author
|
/ci run |
khluu
enabled auto-merge (squash)
September 2, 2026 21:41
|
✅ Triggered Buildkite CI #86947 for commit |
github-actions
Bot
added
the
ready
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
NVIDIA Hybrid SSM NixlConnector PD accuracywas 2,796s P90 across 6 runs in the 24-hour dashboard window ending 2026-09-01 10:31 UTC.What changed
Split the configuration sweep into four static groups selected by the Buildkite shard index, with fail-closed dispatch for unexpected indices.
No merged or active non-draft PR covers this job.
Validation
ci-infraBuildkite step schemabash -nand ShellCheckbash -ngit diff --checkTargeted CI
Build #86575 passed all four shards:
9:14, 13:39, 13:51, 15:25.This is a draft pending human review. AI assistance was used.