Conversation
Documents pl2/milvus-fixes/sriov-gpu-design.md's user-facing surface: spec.gpu (Standalone top-level and per Distributed role) and spec.network.sriov (per Distributed role, with a Standalone-scoped subset). - concepts/milvus.md: adds a spec.gpu and spec.network field reference section, alongside the existing spec.podTemplate section. - guides/milvus/gpu-sriov/: new guide covering the MilvusVersion GPU capability gate, why the SR-IOV advertise-IP init container is Distributed-only, the topology-coupling requirement (all five roles, not just the GPU-bearing ones), a full worked example, and known limitations -- carried over honestly from the design doc rather than overclaiming validation this hasn't had yet. Companion to kubedb/apimachinery#1898 and kubedb/milvus#69. Signed-off-by: Tamal Saha <tamal@appscode.com>
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
Visit the preview URL for this PR (updated for commit f37583c): https://kubedb-v2-hugo--pr1061-milvus-gpu-sriov-jiljn0by.web.app (expires Tue, 22 Sep 2026 16:57:56 GMT) 🔥 via Firebase Hosting GitHub Action 🌎 Sign: 0f29ae8ae0bd54a99bf2b223b6833be47acd5943 |
…ition The GPU/SR-IOV guide and Milvus CRD field reference predated all four follow-ups closed in kubedb/apimachinery#1898 and kubedb/milvus#69 after the initial pass: the env-var (MILVUS_CONF_<SECTION>_IP) advertise-IP mechanism, the optional GDS second SR-IOV attachment, heterogeneous GPU pools (spec.topology.distributed.<role>.groups) including per-group OpsRequest scaling and per-group StreamingNode storage, and the new Milvus.status.conditions[SRIOVNetworkAttached] observability condition. Signed-off-by: Tamal Saha <tamal@appscode.com>
What
Documents the user-facing surface of
pl2/milvus-fixes/sriov-gpu-design.md:spec.gpu(Standalone top-level and per Distributed role) andspec.network.sriov(per Distributed role, with a Standalone-scoped subset), including the follow-ups added after the initialkubedb/apimachinery#1898/kubedb/milvus#69pass.concepts/milvus.md: addsspec.gpu/spec.networkfield reference sections, including the optionalnetwork.sriov.gdssecond attachment andspec.topology.distributed.<role>.groups(heterogeneous GPU pools, with per-groupstorageType/storageforstreamingnode).guides/milvus/gpu-sriov/: guide covering theMilvusVersionGPU capability gate, why the SR-IOV advertise-IP init container is Distributed-only (now documenting both theuser.yamlfile overlay and theMILVUS_CONF_<SECTION>_IPenv-var mechanism it runs belt-and-braces), the topology-coupling requirement, GPU Direct Storage vianetwork.sriov.gds,groups(including scoping aMilvusOpsRequestto a single group), theMilvus.status.conditions[SRIOVNetworkAttached]condition, a full worked example, and known limitations — carried over honestly from the design doc rather than overclaiming validation this hasn't had.Depends on
Stacked on
milvus-etcd(#1060). Companion tokubedb/apimachinery#1898andkubedb/milvus#69.Testing
No live-cluster verification — this guide is drafted from the CRD schema and the design doc, consistent with how the rest of this branch's docs were produced (per #1060's own testing note).