Troubleshoot BMM provisioning in Azure Operator Nexus cluster

As part of cluster deploy action, bare metal machines (BMM) are provisioned with required roles to participate in the cluster. This document supports troubleshooting for common provisioning issues using Azure CLI, Azure portal, and the server baseboard management controller (BMC). For the Azure Operator Nexus platform, the underlying server hardware uses integrated Dell remote access controller (iDRAC) as the BMC. Provisioning uses the Preboot eXecution Environment (PXE) interface to load the Operating System (OS) on the BMM.

Prerequisites

  1. Install the latest version of the appropriate CLI extensions
  2. Collect the following information:
    • Subscription ID (SUBSCRIPTION)
    • Cluster name (CLUSTER)
    • Resource group (CLUSTER_RG)
    • Managed resource group (CLUSTER_MRG)
  3. Request subscription access to run Azure Operator Nexus network fabric (NF) and network cloud (NC) CLI extension commands.
  4. Log in to Azure CLI and select the subscription where the cluster is deployed.

BMM roles

For a given SKU, there are required roles to manage and operate the underlying kubernetes cluster.

The following roles are assigned to BMM resources (see BMM roles reference):

  • Control plane: BMM responsible for running the kubernetes control plane agents for cluster.
  • Management plane: BMM responsible for running the platform agents including controllers and extensions.
  • Compute plane: BMM responsible for running actual tenant workloads including kubernetes clusters and virtual machines.

Listing BMM status

This command will list all bareMetalMachineName resources in the managed resource group with simple status:

az networkcloud baremetalmachine list -g $CLUSTER_MRG -o table

Name          ResourceGroup                  DetailedStatus    DetailedStatusMessage
------------  -----------------------------  ----------------  ---------------------------------------
BMM_NAME      CLUSTER_MRG                    STATUS            STATUS_MSG

Where STATUS goes through the following phases through the BMM provisioning process (see BMM Status in Azure Operator Nexus Compute Concepts):

Registering -> Preparing -> Inspecting -> Available -> Provisioning -> Provisioned

These phases are defined as follows:

Phase Actions
Registering Verifying BMC connectivity/BMC credentials and adding BMM to provisioning service.
Preparing Rebooting BMM, resetting BMC, and verifying power state.
Inspecting Updating firmware, applying BIOS settings, and configuring storage.
Available BMM is ready to install OS.
Provisioning OS image installing on the BMM. After OS is installed, BMM attempts to join cluster.
Provisioned BMM successfully provisioned and joined to cluster.
Deprovisioning BMM provisioning failed. Provisioning service is cleaning up resource for retry.
Failed BMM provisioning failed and manual recovery is required. All retries exhausted.

During any phase, the BMM detailed status is set to failed and the phase is blocked if any of the following occurs:

  • BMC is unavailable
  • Network port is down
  • Hardware component fails

To get a more detailed status of the BMM:

az networkcloud baremetalmachine list -g $CLUSTER_MRG --query "sort_by([].{name:name,readyState:readyState,provisioningState:provisioningState,detailedStatus:detailedStatus,detailedStatusMessage:detailedStatusMessage,powerState:powerState,machineRoles:machineRoles| join(', ', @),createdAt:systemData.createdAt}, &name)" --output table

Name            ReadyState    ProvisioningState    DetailedStatus    DetailedStatusMessage                      PowerState    MachineRoles                                      CreatedAt
------------    ----------    -----------------    --------------    -----------------------------------------  ----------    ------------------------------------------------  -----------
BMM_NAME        RSTATE        PROV_STATE           STATUS            STATUS_MSG                                 POWER_STATE   BMM_ROLE                                          CREATE_DATE

Where the output is defined as follows:

Output Definition
BMM_NAME BMM name
RSTATE Cluster participation status (True,False).
PROV_STATE Provisioning state (Succeeded,Failed).
STATUS Provisioning detailed status (Registering,Preparing,Inspecting,Available,Provisioning,Provisioned,Deprovisioning,Failed).
STATUS_MSG Detailed provisioning status message.
POWER_STATE Power state of BMM (On,Off).
BMM_ROLE BMM cluster role (control-plane,management-plane,compute-plane).
CREATE_DATE BMM creation date.

For example:

x01dev01c01w01  True          Succeeded            Provisioned       The OS is provisioned to the machine       On            platform.afo-nc.microsoft.com/compute-plane=true  2024-05-03T15:12:48.0934793Z
x01dev01c01w01  False         Failed               Preparing         Preparing for provisioning of the machine  Off           platform.afo-nc.microsoft.com/compute-plane=true  2024-05-03T15:12:48.0934793Z

BMM details

To show details and status of a single BMM:

az networkcloud baremetalmachine show -g $CLUSTER_MRG -n $BMM_NAME

For BMM details specific to troubleshooting:

az networkcloud baremetalmachine show -g $CLUSTER_MRG -n $BMM_NAME --query "{name:name,BootMAC:bootMacAddress,BMCMAC:bmcMacAddress,Connect:bmcConnectionString,SN:serialNumber,rackId:rackId,RackSlot:rackSlot}" -o table

Troubleshooting failed provisioning state

The following conditions can cause provisioning failures:

Error Type Resolution
BMC shows Backplane Comm critical error. 1) Execute BMM remote flea drain. 2) Perform BMM physical flea drain. 3) Execute BMM replace action.
Boot (PXE) network data response empty from BMC. 1) Reset port on fabric device. 2) Execute BMM remote flea drain. 3) Perform BMM physical flea drain. 4) Execute BMM replace action.
Boot (PXE) MAC address mismatch. 1) Validate BMM MAC address data against BMC data. 2) Execute BMM remote flea drain. 3) Perform BMM physical flea drain. 4) Execute BMM replace action.
BMC MAC address mismatch 1) Validate BMM MAC address data against BMC data. 2) Execute BMM remote flea drain. 3) Perform BMM physical flea drain. 4) Execute BMM replace action.
Disk data response empty from BMC. 1) Remove/replace disk. 2) Remove/replace storage controller. 3) Execute BMM remote flea drain. 4) Perform BMM physical flea drain. 5) Execute BMM replace action.
BMC unreachable. 1) Reset port on fabric device. 2) Remove/replace cable. 3) Execute BMM remote flea drain. 4) Perform BMM physical flea drain. 5) Execute BMM replace action.
BMC fails log in. 1) Update credentials on BMC. 2) Execute BMM replace action.
Memory, CPU, OEM critical errors on BMC. 1) Resolve hardware issue with remove/replace. 2) Execute BMM remote flea drain. 3) Perform BMM physical flea drain. 4) Execute BMM replace action.
Console stuck at boot loader (GRUB) menu. 1) Execute NVRAM reset. 2) Execute BMM replace action.

Azure BMM activity log

  1. Log in to Azure portal.
  2. Search on the BMM name in the top Search box.
  3. Select the Bare Metal Machine (Operator Nexus) from the search results.
  4. Select Activity log on the left side menu.
  5. Make sure the Timespan encompasses the provisioning period.
  6. Expand the BareMetalMachines_Update operation and select any that show Failed status.
  7. Select JSON tab to get the detailed status message.

Look for failures related to invalid credentials or BMC unavailable.

Determine BMC IPv4 address

The IPv4 address of the BMC (BMC_IP) is in the Connect value returned from the previous BMM Details section.

Validate MAC address of BMM against BMC data

To get the MAC address information from the BMM:

az networkcloud baremetalmachine show -g $CLUSTER_MRG -n $BMM_NAME --query "{name:name,BootMAC:bootMacAddress,BMCMAC:bmcMacAddress,SN:serialNumber,rackId:rackId,RackSlot:rackSlot}" -o table

Verify the MAC address data against the BMC through the WEB UI: BMC -> Dashboard - Shows BMC MAC address BMC -> System Info -> Network -> Embedded.1-1-1 - Shows Boot MAC address

Verify the MAC address using racadm from a jumpbox that has access to the BMC network:

racadm --nocertwarn -r $IP -u $BMC_USR -p $BMC_PWD getsysinfo | grep "MAC Address "        #BMC MAC
racadm --nocertwarn -r $IP -u $BMC_USR -p $BMC_PWD getsysinfo | grep "NIC.Embedded.1-1-1"  #Boot MAC

If the MAC address supplied to the cluster is incorrect, use the BMM replace action at BMM actions to correct the addresses.

Ping test BMC connectivity

Attempt to run ping against the BMC IPv4 address:

  1. Obtain the IPv4 address (BMC_IP) from the previous Determine BMC IPv4 address.

  2. Test ping to the BMC:

    To test from a jumpbox that has access to the BMC network:

    ping $BMC_IP -c 3
    

    To test from a BMM control-plane host using Azure CLI:

    az networkcloud baremetalmachine run-read-command -g $CLUSTER_MRG -n $BMM_NAME --limit-time-seconds 60 --commands "[{command:'ping',arguments:['$BMC_IP',-c,3]}]"
    

Reset port on fabric device

If the BMC_IP isn't responsive, a reset of the fabric device port retriggers autonegotiation on the port and may bring it back online.

To find the Network Fabric port from Azure:

  1. Obtain the RackID and RackSlot from the previous BMM Details section.

  2. In Azure portal, drill down to the Network Rack RackID for the BMM.

  3. Select Network Devices tab and the management (Mgmt) switch for the rack.

  4. Under Resources, select Network Interfaces and then the BMC (iDRAC) or boot (PXE) interface for the port that requires reset.

    Collect the following information:

    • Network fabric resource group (NF_RG)
    • Device name (NF_DEVICE_NAME)
    • Interface name (NF_DEVICE_INTERFACE_NAME)
  5. Reset the port:

    To reset the port using Azure CLI:

    az networkfabric interface update-admin-state -g $NF_RG --network-device-name $NF_DEVICE_NAME --resource-name $NF_DEVICE_INTERFACE_NAME --state Disable
    az networkfabric interface update-admin-state -g $NF_RG --network-device-name $NF_DEVICE_NAME --resource-name $NF_DEVICE_INERFACE_NAME --state Enable
    

BMM remote power drain (flea drain)

Perform a remote flea drain against the BMM through the BMC UI: BMC -> Configuration -> BIOS Settings -> Miscellaneous Settings -> Select "Full Power Cycle" under Power Cycle Request -> Apply and reboot

Perform a remote flea drain using racadm from a jumpbox that has access to the BMC network:

racadm set bios.miscsettings.powercyclerequest FullPowerCycle
racadm jobqueue create BIOS.Setup.1-1
racadm serveraction powercycle

BMM physical power drain (flea drain)

For a physical flea drain, the local site hands physically disconnect the power cables from both power adapters for 5 minutes and then restore power. This process ensures the server, capacitors, and all components have complete power removal and all cached data is cleared.

Reset NVRAM

If provisioning failed due to an OEM or hardware error, the boot sequence may be locked in NVRAM to PXE boot instead of showing hdd or hard drive listed first in the boot order.

This condition typically shows the BMM at the bootloader stage on the console and is blocked without manual keystroke intervention.

To reset the NVRAM, use the following sequence in the BMC UI: Maintenance -> Diagnostics -> Reset iDrac to Factory Defaults -> Discard All Settings, but preserve user and network settings -> Apply and reboot

Reset BMC password

If the activity log indicates invalid credentials on the BMC, run the following command from a jumpbox that has access to the BMC network:

racadm -r $BMC_IP -u $BMC_USER -p $CURRENT_PASSWORD  set iDRAC.Users.2.Password $BMC_PWD

Adding servers back into the cluster after a repair

After hardware is fixed, run BMM replace action following instructions from the following page BMM actions.

If you still have questions, contact support. For more information about support plans, see Azure Support plans.