Add decommission as a new Flow operation type. A single DecommissionRack call creates one independent task per rack and drives the decommission sequence in dependency order via Temporal: Stage 1: Compute — DecommissionControl + WaitDecommissioned Stage 2: NVSwitch — DecommissionControl + WaitDecommissioned Stage 3: PowerShelf — DecommissionControl + WaitDecommissioned Each stage polls Core's controller_state every 30 s until every component reports "Decommissioned". A failure in any stage prevents later stages from starting; earlier stages are not rolled back. The Decommission/Switch/PowerShelf/Machine Core RPCs are stubbed with TODO comments pending Eric's work (issues #3819, #3824, #3825). FindMachineControllerStates is implemented now via FindMachinesByIds. Closes #3826 --------- Signed-off-by: Krishna Dhulipala <kdhulipala@nvidia.com> Signed-off-by: Kun Zhao <kunzhao@nvidia.com> Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Kun Zhao <kunzhao@nvidia.com>

Uh oh!

Managed switch decommissioning #3824

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions