-
Bug
-
Resolution: Unresolved
-
Normal
-
None
-
4.17.z
-
Quality / Stability / Reliability
-
False
-
-
3
-
None
-
None
-
None
-
None
-
Rejected
-
Metal Platform 277, Metal Platform 278
-
2
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Description of problem:
When attempting to provision NVIDIA DGX B200 nodes through Advanced Cluster Management (ACM), the BareMetalHost (BMH) status hangs at Provisioning.
The Bare Metal Operator (BMO) and Ironic logs show Redfish VirtualMedia mounting failures. The BMC rejects the VirtualMedia.InsertMedia request with an HTTP 400 error, requiring UserName and Password parameters in the request body.
This behavior prevents automated provisioning of DGX B200 nodes.
Version-Release number of selected component (if applicable):
- OpenShift: 4.18.3 (hub cluster)
- ACM: 2.13
- Ironic/BMO components from OCP 4.18.3 release
How reproducible:
Always – occurs consistently when attempting to provision NVIDIA DGX B200 nodes.
Steps to Reproduce:
- Register NVIDIA DGX B200 node in ACM using BareMetalHost (via BMC/Redfish).
- Observe provisioning workflow.
- Check Ironic/BMO logs when provisioning fails.
Actual results:
- BMH stuck in Provisioning state.
- ISO not mounted via Redfish VirtualMedia.
- BMO/Ironic logs show:
HTTP request: POST https://<bmc>/redfish/v1/Managers/BMC/VirtualMedia/CD1/Actions/VirtualMedia.InsertMedia body: {'Image': 'https://<iso-server>/boot-<uuid>.iso', 'TransferProtocolType': 'HTTPS'}
Response:
status code: 400 error: Base.1.12.GeneralError: The action ... requires the parameter UserName to be present in the request body. MessageId: Base.1.12.ActionParameterMissing
Expected results:
- Redfish VirtualMedia ISO should mount successfully.
- BMH should proceed past Provisioning and complete host registration.
Additional info:
- The fix for handling UserName in Redfish InsertMedia was merged upstream (sushy) and included in OCP 4.18.1 (Provide vmedia username and password if required).
- However, in this scenario, the firmware’s response differs: the error lacks a structured code field, causing credential detection logic to fail.
- Engineering suspects a firmware bug, but BMO/Ironic may require additional handling (e.g., checking MessageId=ActionParameterMissing or the error field directly) as a workaround.