TL/CUDA: fix NVLS on fabric cluster UUID#1308
Open
ikryukov wants to merge 1 commit into
Open
Conversation
NVML cliqueId/partitionId are unique only within a fabric cluster, so unrelated standalone DGX nodes can report the same cliqueId/partitionId with different fabrics and silently pass ucc_topo_is_single_nvlink_domain. Multinode NVLS init then crashes late at cuMulticastAddDevice and wedges team creation. Read clusterUuid (nvmlGpuFabricInfo_v2_t, the globally-unique fabric identifier) into ucc_gpu_info_t.fabric_cluster_uuid and require it to be non-all-zero and identical across ranks before approving a single NVLink domain. Verified on gaia DGX H100 where NVML reports all-zero clusterUuid with state=COMPLETED: the topo guard now correctly disables multinode NVLS instead of failing late. Signed-off-by: Ilya Kryukov <ikryukov@nvidia.com>
ikryukov
requested review from
Sergei-Lebedev,
janjust and
nsarka
and removed request for
Sergei-Lebedev
Collaborator
Author
|
/build |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Fix silent multinode NVLS crash on standalone DGX nodes.
Why
NVML
cliqueId/partitionIdare unique only within a fabric cluster, so unrelated DGX nodes can collide on these values, falsely passucc_topo_is_single_nvlink_domain, and crash later incuMulticastAddDevice.How
Read
clusterUuid(globally unique) fromnvmlGpuFabricInfo_v2_tintoucc_gpu_info_t.fabric_cluster_uuidand require it to be non-all-zero and matching across ranks.