feat(uat): enroll azure-h100 in the nightly batch with [training] - #1726
Conversation
The Azure UAT acceptance run (29125390442) went green end-to-end — provision, prep, install, validate all phases, TrainJob, verify, signed-evidence ingest to the corroboration bucket, and clean teardown — which was the stated gate for flipping azure-h100 off its bring-up opt-out. Enrolls [training] only; inference joins after a green manual intent=inference dispatch (the serve CUJ step remains disabled pending #1644 on all clouds, matching AWS/GCP). Updates the uat.md enrollment table/prose and the committed-registry lock-in test. Signed-off-by: Mark Chmarny <mark@chmarny.com>
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Enterprise Run ID: 📒 Files selected for processing (3)
📝 WalkthroughWalkthroughAzure H100 is enrolled in nightly training through the reservation configuration, with registry validation updated to expect Estimated code review effort: 2 (Simple) | ~10 minutes Possibly related PRs
Suggested labels: Suggested reviewers: ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
|
🌿 Preview your docs: https://nvidia-preview-feat-azure-nightly-enrollment.docs.buildwithfern.com/aicr |
Coverage Report ✅
Coverage BadgeNo Go source files changed in this PR. |
…IDIA#1726) Signed-off-by: Mark Chmarny <mark@chmarny.com>
Summary
Enrolls
azure-h100in the nightly UAT batch withnightly-intents: [training], ending its bring-up manual-dispatch-only phase. Updates theuat.mdenrollment table/prose and the committed-registry lock-in test.Motivation / Context
The stated gate for flipping
azure-h100offnightly-intents: []was a green manual run. The acceptance run (29125390442) completed green end-to-end: provision → prep → install → validate (all phases) → TrainJob → verify → signed-evidence ingest to the corroboration bucket (now rendered on validation.aicr.run) → clean teardown.Training-only for now: inference joins after a green manual
intent=inferencedispatch, matching how the intent set is data-driven per reservation. (The serve CUJ step remains disabled on all clouds pending #1644, so an Azure inference cell would provision + validate the inference stack, same as AWS/GCP.)Fixes: N/A
Related: #1709, #1722, #1275, #1276
Type of Change
Component(s) Affected
docs/)infra/uat/reservations.yaml),pkg/uatbrokerlock-in testImplementation Notes
inferenceto the list rather than removing the key (a barenightly-intents:decodes as absent and defaults to[training]).TestCommittedRegistryValidupdated — it deliberately pins the committed intent set, so this data change fails the test without the matching expectation change.Testing
uat-nightly-batch.yamlcron (0 4 UTC) — the azure-h100 training cell dispatches through the sameuat-runpath the acceptance run used.Risk Assessment
Checklist
make testwith-race)make lint)docs/contributor/uat.md)