Kubernetes Identifier Description

**Table 1** Node labels used in cluster scheduling components
Node Label	Description	Value	Required Component
accelerator	Processor of a node.	huawei-Ascend910, huawei-Ascend310, and huawei-Ascend310P	Ascend Device Plugin
host-arch	CPU architecture of a node.	huawei-x86 and huawei-arm	Volcano
masterselector	Master node of MindX DL.	dls-master-node	Volcano, HCCL-Controller, and Resilience-Controller
nodeDEnable	Whether to enable NodeD.	on	Volcano and Resilience-Controller
workerselector	Worker node of MindX DL.	dls-worker-node	Ascend Device Plugin, NodeD, and NPU-Exporter
accelerator-type	Type of the Atlas training server.	card, module, and half	Ascend Device Plugin and Volcano
servertype	Atlas 200I SoC A1 core board.	soc	Volcano
huawei.com/Ascend910-Recover	Faulty Ascend 910 AI Processor.	ID of the faulty processor	Ascend Device Plugin
huawei.com/Ascend910-NetworkUnhealthyRecover	Network of an Ascend 910 AI Processor recovers.	ID of the faulty processor	Ascend Device Plugin

**Table 2** Node annotations used in cluster scheduling components
Node Annotation	Description	Value	Required Component
noded/heartbeat	NodeD heartbeat that indicates whether a node is healthy.	string	Volcano, NodeD, and Resilience-Controller
noded/heartbeat-interval	NodeD heartbeat interval.	string	Volcano, NodeD, and Resilience-Controller

**Table 3** Pod labels used in cluster scheduling components
Package	Function	Value	Required Component
ring-controller.atlas	An Atlas PoD.	ascend-910	Ascend Device Plugin and HCCL-Controller
fault-scheduling	Whether to enable fault rescheduling.	grace, force, and off	Volcano and Resilience-Controller
elastic-scheduling	Whether to enable job elastic scheduling.	on	Resilience-controller

**Table 4** Pod annotations used in cluster scheduling components
Name	Function	Value	Required Component
ascend.kubectl.kubernetes.io/ascend-910-configuration	Data source of hccl.json generated by HCCL-Controller	String in MAP format	Ascend Device Plugin and HCCL-Controller
hccl/rankIndex	Basis for retaining the original rank ID during resumable training.	[0,1000]	Volcano and HCCL-Controller
huawei.com/Ascend910	Basis for Ascend Device Plugin to allocate processors to PoDs	String	Volcano and Ascend Device Plugin
huawei.com/AscendReal	Records of the processors allocated to pods by Ascend Device Plugin.	String	Volcano and Ascend Device Plugin
huawei.com/kltDev	Records of the processors allocated to PoDs by kubelet.	String	Ascend Device Plugin
predicate-time	Sequence for Ascend Device Plugin to allocate processors to PoDs.	String	Volcano and Ascend Device Plugin

**Table 5** ConfigMaps used in cluster scheduling components
ConfigMap	Namespace	Description	Required Component
vcjob-fault-npu-cm	volcano-system	Fixes fault rescheduling content.	Volcano
volcano-scheduler-configmap	volcano-system	Volcano-Scheduler configuration file (native)	Volcano
mindx-dl-deviceinfo-node name	kube-system	Processor information on a node reported by Ascend Device Plugin	Volcano, Ascend Device Plugin, and Resilience-Controller
Fault-config-job name	job space	Information about faulty rank IDs required for resumable training	Volcano and Elastic-Agent
Rings-config-job name	job space	hccl.json content	Ascend Device Plugin and HCCL-Controller

Parent topic: APIs