SambaRack Manager Power Operations - SambaNova Documentation

The snctl power command group controls and monitors DC power for SN40L nodes. Power operations live in their own top-level command group, separate from snctl node.

Power operations affect actual hardware. They perform DC cycle operations through management interfaces (IPMI, Redfish, or SSH), and do not control AC power (PDUs), network switches, serial console servers, or other infrastructure components.

Key concepts

Read this section once. Everything after it depends on these five things.

Prerequisites

Every power command authenticates to the target devices, so device credentials must be registered first. If they are missing or wrong, the command fails to authenticate.

snctl secret set --device-type bmc
snctl secret set --device-type card
snctl secret set --device-type host
snctl secret set --device-type xrdu

See Device credentials for details.

What each `--type` targets

`--type`	Targets	Notes
`node`	The complete physical unit — host CPU and all XRDUs	Default. This is the ordering firmware procedures expect before a BIOS or XRDU upgrade.
`host`	The host system only
`xrdu`	The SambaNova execution units only

snctl power cycle does not accept --type — it always targets the whole node.

Choosing a target

Either --node or --group must be specified. They are mutually exclusive.

Flag	Accepts
`--node <name_or_rack_id>`	A node name (`sn40l-16-001-n1`) or a rack ID (`001`)
`--group <name>`	A group name — targets every node in that group

The SambaNova Daemon check

snctl power on and snctl power cycle wait for snd.service (the SambaNova Daemon) to become active before returning, so a successful exit means the node is ready for work rather than merely powered. Pass --skip-snd-check to return as soon as power is applied, without waiting for the daemon. Use it when powering on hardware that is not expected to boot into a working SambaStack environment — for example, immediately before a firmware upgrade.

snctl power on --node sn40l-16-001-n1 --type node --skip-snd-check

snctl power off and snctl power state do not wait for the daemon and have no --skip-snd-check flag.

Flags

Flag	Description	Applies to
`--node <name_or_rack_id>`	Target a single node by name or rack ID	All
`--group <name>`	Target all nodes in a named group	All
`--type <device_type>`	`node` (default), `host`, or `xrdu`	`state`, `on`, `off`
`-o, --output <format>`	`text` (default), `wide`, or `json`	All
`--protocol <protocol>`	Override the default management protocol: `ipmi`, `ssh`, or `redfish`	All
`--skip-snd-check`	Return without waiting for `snd.service`	`on`, `cycle`
`-p, --parallelism-racks <n>`	Max racks processed simultaneously (`0` = uncapped)	All
`--parallelism-devices <n>`	Global max nodes in-flight across all racks (`0` = uncapped)	All

Concurrency defaults

--parallelism-racks caps how many racks are processed at once, and its default differs per subcommand:

Subcommand	Default `--parallelism-racks`
`power state`	`0` (uncapped — all racks queried in parallel)
`power on`	`4`
`power off`	`1`
`power cycle`	`1`

# Process up to 4 racks at a time (power on default)
snctl power on --group production --type node --parallelism-racks 4

# Process racks one at a time (safer for power off / cycle)
snctl power off --group production --type node --parallelism-racks 1

# No limit — start all racks at once
snctl power state --group production --type node --parallelism-racks 0

--parallelism-devices is a separate, global cap on nodes in-flight across all racks at once, regardless of how many racks are being processed. Use it to limit the total number of simultaneous IPMI, Redfish, and SSH sessions against your management network:

# Allow up to 8 racks in parallel, but cap total nodes at 5
snctl power on --group large-cluster --parallelism-racks 8 --parallelism-devices 5

Check power state

Query the current power state of a node or its components. This is read-only and safe to run at any time.

snctl power state \
  [--node <node_name_or_rack_id>] \
  [--group <group_name>] \
  [--type <device_type>]

Each cell reports On, Off, or Unknown. A component that could not be reached shows Error, with the reason listed as a note below the table. Single node:

snctl power state --node sn40l-16-001-n1 --type node

Node (Rack)            Host  XRDU/0  XRDU/1  XRDU/2  XRDU/3  XRDU/4  XRDU/5  XRDU/6  XRDU/7
---------------------  ----  ------  ------  ------  ------  ------  ------  ------  ------
sn40l-16-001-n1 (001)  On    On      On      On      On      On      On      On      On

Every node in a group:

snctl power state --group production --type node

Node (Rack)            Host  XRDU/0  XRDU/1  XRDU/2  XRDU/3  XRDU/4  XRDU/5  XRDU/6  XRDU/7
---------------------  ----  ------  ------  ------  ------  ------  ------  ------  ------
sn40l-16-001-n1 (001)  On    On      On      On      On      On      On      On      On
sn40l-16-002-n1 (002)  On    On      On      On      On      On      On      On      On

Host only:

snctl power state --node sn40l-16-001-n1 --type host

Node (Rack)            Host
---------------------  ----
sn40l-16-001-n1 (001)  On

Including the protocol that answered each query:

snctl power state --node sn40l-16-001-n1 --type node -o wide

With -o wide, the protocol is appended inline to each state rather than added as a separate column, because the table is pivoted with one column per component:

Node (Rack)  Host     XRDU/0   XRDU/1   XRDU/2   XRDU/3   XRDU/4   XRDU/5   XRDU/6   XRDU/7
-----------  -------  -------  -------  -------  -------  -------  -------  -------  -------
sn40l-16-001-n1 (001)  On (ipmi)  On (redfish)  On (redfish)  On (redfish)  On (redfish)  On (redfish)  On (redfish)  On (redfish)  On (redfish)

Power on

Turn on a node, its host, or its XRDUs.

snctl power on \
  [--node <node_name_or_rack_id>] \
  [--group <group_name>] \
  [--type <device_type>] \
  [--skip-snd-check]

# Power on an entire node
snctl power on --node sn40l-16-001-n1 --type node

#Expected Output
sn40l-16-001-n1 (001)
  Powering on XRDUs  OK
  Powering on Host  OK
  Waiting for Host to Boot  OK
  Waiting for snd.service to start  OK

# Power on the host only
snctl power on --node sn40l-16-001-n1 --type host

# Power on every node in a group
snctl power on --group production --type node

Power off

Turn off a node, its host, or its XRDUs.

snctl power off \
  [--node <node_name_or_rack_id>] \
  [--group <group_name>] \
  [--type <device_type>]

# Power off an entire node (host and XRDUs)
snctl power off --node sn40l-16-001-n1 --type node

#Expected Output
sn40l-16-001-n1 (001)
  Powering off Host  OK
  Powering off Node  OK

# Power off the host only
snctl power off --node sn40l-16-001-n1 --type host

# Power off every node in a group
snctl power off --group production --type node

Power cycle

Power-cycle the entire node. Unlike the other subcommands, power cycle always targets the whole node — it does not accept a component --type.

snctl power cycle \
  [--node <node_name_or_rack_id>] \
  [--group <group_name>] \
  [--skip-snd-check]

# Power cycle a single node
snctl power cycle --node sn40l-16-001-n1

#Expected Output
sn40l-16-001-n1 (001)
  Powering off Host  OK
  Powering off XRDUs  OK
  Waiting before power-on  OK
  Powering on XRDUs  OK
  Powering on Host  OK
  Waiting for Host to Boot  OK
  Waiting for snd.service to start  OK

# Power cycle every node in a group
snctl power cycle --group production

Next step

To check or upgrade firmware on these nodes, see Firmware Updates.

Get started

Models

Features

Build

Resources

Power Operations

Key concepts

Prerequisites

What each `--type` targets

Choosing a target

The SambaNova Daemon check

Flags

Concurrency defaults

Check power state

Power on

Power off

Power cycle

Next step

​Key concepts

​Prerequisites

​What each --type targets

​Choosing a target

​The SambaNova Daemon check

​Flags

​Concurrency defaults

​Check power state

​Power on

​Power off

​Power cycle

​Next step

Key concepts

Prerequisites

What each `--type` targets

Choosing a target

The SambaNova Daemon check

Flags

Concurrency defaults

Check power state

Power on

Power off

Power cycle

Next step