Stop System Components

Remove, stop, or power down components.

When a system administrator removes, stops, or powers down components, any applications and compute processes that are running on those components are lost.

Reserve a Component

To allow applications and compute processes to complete before stopping components, use the HSS xtcli set_reserve idlist command to prevent the selected nodes from accepting new jobs.

A node running CNL and using ALPS is considered to be down by ALPS after it is reserved using the xtcli set_reserve command. The output from apstat will show the node as down (DN), even though there may be an application running on that node. This DN designation indicates that no other work will be placed on the node after the currently running application has terminated.

For more information, see the xtcli_set(8) man page.

Reserve a component

crayadm@smw> xtcli set_reserve idlist

Power Down Blades or Cabinets

Warning: Power down the cabinets with software commands. Tripping the circuit breakers may result in damage to system components.
Warning: Before powering down a blade or a cabinet, ensure the operating system is not running.

The xtcli power down command powers down the specified cabinet and/or blades within the specified partition, chassis or list of blades. Cabinets must be in the READY state to receive power commands.

When a request is made to power down a blade consisting of Intel® Xeon® processor Scalable Family nodes or a Cabinet containing processor blades of this type, the nodes are powered off into the G3 state (full power off) prior to the Cabinet controller removing power from the blade. See System Component States.

The xtcli power down command has the following form, where physIDlist is a comma-separated list of cabinets, blades, or nodes present on the system.
xtcli power down physIDlist

The xtcli power force_down and xtcli power down_slot commands are aliases for the xtcli power down command. For information about disabling and enabling components, see Disabling Hardware Components, and Enabling Hardware Components, respectively.

Warning: Although a blade is powered off, the HSS in the cabinet is live and has power.

For information about powering down a component, see the xtcli_power(8) man page.

Power down a specified blade

For this example, power down a blade with the ID c0-0c0s7:
crayadm@smw> xtcli power down c0-0c0s7

Power Down a Specific Node

The xtcli power down_node command powers down the specified node and/or nodes within a specified partition, chassis, list of blades, or list of nodes. When specifying a specific node or list of nodes, all node types are powered down to the G3 state except for Intel® Xeon® processor Scalable Family nodes, which are powered down to the S5 state (soft off). These nodes can be powered down to the G3 state using one of the following methods:
  • Issue the xtcli power down_node command with the --with-si flag.
  • Power down the blade that the Intel® Xeon® processor Scalable Family nodes reside on. Blades must be in the READY state to receive power commands. See System Component States.
The xtcli power down_node command has the following form, where physIDlist is a comma-separated list of cabinets, blades, or nodes present on the system.
xtcli power down_node physIDlist

Power down specified nodes

In these example commands, c0-0c0s7n0 is a Haswell node and c0-1c1s8n2 is a Intel® Xeon® processor Scalable Family node. The following down_node power command does not include the --with-si flag.
crayadm@smw> xtcli power down_node c0-0c0s7n0,c0-1c1s8n2

HSS reports both nodes as being in the off state. The state of c0-0c0s7n0 is G3, and the state of c0-1c1s8n2 is S5.

The next example uses the --with-si flag to power down the same two nodes.

crayadm@smw> xtcli power down_node --with-si c0-0c0s7n0,c0-1c1s8n2

HSS reports both nodes as being in the off state. Both nodes are in the G3 state. See the xtcli_power(8) man page for more information.

Halt Selected Nodes

Use the HSS xtcli halt command to halt selected nodes. For more information, see the xtcli(8) man page.

Halt a node

For this example, halt node 157:

crayadm@smw> xtcli halt 157