rescheduling

const (
	// RePropertyName name specifying re-scheduler cm
	RePropertyName = "re-scheduling"
	// CmName Name of ReSchedulerConfigmap
	CmName = "vcjob-fault-npu-cm"
	// CmNameSpace Namespace of ReSchedulerConfigmap
	CmNameSpace = "volcano-system"

	// JobRescheduleLabelKey key word of re-scheduling configuration
	JobRescheduleLabelKey = "fault-scheduling"
	// JobGraceRescheduleLabelValue Grace delete reschedule job, possible value of re-scheduling configuration
	JobGraceRescheduleLabelValue = "grace"
	// JobForceRescheduleLabelValue Force delete reschedule job, possible value of re-scheduling configuration
	JobForceRescheduleLabelValue = "force"
	// JobOffRescheduleLabelValue not delete reschedule job, possible value of re-scheduling configuration
	JobOffRescheduleLabelValue = "off"
	// GraceOverTimeKey for GraceOverTime config by user
	GraceOverTimeKey = "grace-over-time"
	// ElasticSchedulingKey for distinguishing whether a job is enabled with elastic scheduling
	ElasticSchedulingKey = "elastic-scheduling"
	// JobOnElasticScheduling job enabled with elastic scheduling
	JobOnElasticScheduling = "on"
	// JobOffElasticScheduling job not enabled with elastic scheduling
	JobOffElasticScheduling = "off"

	// CmFaultNodeKind key in configmap which saves the FaultNode cache
	CmFaultNodeKind = "fault-node"
	// CmFaultJob910bx2Kind key in configmap which saves the 910bx2 FaultJob cache
	CmFaultJob910bx2Kind = "fault-job-910bx2"
	// CmFaultJob910bx2InferKind key in configmap which saves the 910bx2-infer FaultJob cache
	CmFaultJob910bx2InferKind = "fault-job-910bx2-infer"
	// CmFaultJob910bx8Kind key in configmap which saves the 910bx8 FaultJob cache
	CmFaultJob910bx8Kind = "fault-job-910bx8"
	// CmFaultJob910bx16Kind key in configmap which saves the 910bx16 FaultJob cache
	CmFaultJob910bx16Kind = "fault-job-910bx16"
	// CmFaultJob910x8Kind key in configmap which saves the 910x8 FaultJob cache
	CmFaultJob910x8Kind = "fault-job-910x8"
	// CmFaultJob910x4Kind key in configmap which saves the 910x8 FaultJob cache
	CmFaultJob910x4Kind = "fault-job-910x4"
	// CmFaultJob910x2Kind key in configmap which saves the 910x8 FaultJob cache
	CmFaultJob910x2Kind = "fault-job-910x2"
	// CmFaultJob310x4Kind key in configmap which saves the 310x4 FaultJob cache
	CmFaultJob310x4Kind = "fault-job-310x4"
	// CmFaultJob310PKind key in configmap which saves the 310P FaultJob cache
	CmFaultJob310PKind = "fault-job-310P"
	// CmNodeHeartbeatKind judging node fault needs heartbeat info from former session, so should be recorded
	CmNodeHeartbeatKind = "node-heartbeat"
	// CmJobRemainRetryTimes judging node fault needs heartbeat info from former session, so should be recorded
	CmJobRemainRetryTimes = "remain-retry-times"
	// CmNodeRankTimeMapKind record map jobUID rankIndex node and times of occurrence
	CmNodeRankTimeMapKind = "node-rankIndex-Occurrence"
	// CmCheckCode Check code key
	CmCheckCode = "checkCode"

	// CmFaultJob key in configmap which saves the FaultJob cache
	CmFaultJob = "fault-job"

	// DefaultGraceOverTime time interval for grace delete
	DefaultGraceOverTime = 900

	// CardHealthy represents a healthy card
	CardHealthy = "Healthy"
	// CardUnhealthy represents an unhealthy card
	CardUnhealthy = "Unhealthy"
	// CardNetworkUnhealthy represents a network unhealthy card
	CardNetworkUnhealthy = "NetworkUnhealthy"
	// NodeHealthy represents node is available for scheduling
	NodeHealthy = "Healthy"
	// NodeUnhealthy represents node is unhealthy by judging heartbeat
	NodeUnhealthy = "NodeUnhealthy"
	// NodeCardUnhealthy represents node is unhealthy because of the card is unhealthy
	NodeCardUnhealthy = "CardUnhealthy"
	// NodeCardNetworkUnhealthy represents node is unhealthy because of card is network unhealthy
	NodeCardNetworkUnhealthy = "CardNetworkUnhealthy"
	// NoFaultJobsErr none fault jobs
	NoFaultJobsErr = "none fault jobs to be restarted in cache"
	// JobFaultRankIDCMPre the job cm name prefix, for retraining
	JobFaultRankIDCMPre = "fault-config-"
	// JobFaultRankIDCMDataKey the job cm value key.
	JobFaultRankIDCMDataKey = "fault-npus"
	// JobRecovery Name of cm for recovery
	JobRecovery = "job-recovery"
	// DeviceFaultCmKey the key of DeviceFault info
	DeviceFaultCmKey = "huawei.com/Ascend910-Fault"
	// PodFailed the state of failed pod
	PodFailed = "pod-failed"
	// PodHealthy the state of healthy pod
	PodHealthy = "pod-healthy"

	// FaultRetryTimesKey key of fault-retry-times label
	FaultRetryTimesKey = "fault-retry-times"
)

const (
	// PreSeparateNPU fault type waiting user check
	PreSeparateNPU = "PreSeparateNPU"
	// NotHandleFault fault type not handle
	NotHandleFault = "NotHandleFault"
	// NodeFaultCode fault type nodeUnhealthy
	NodeFaultCode = "heartbeatTimeOut"
	// AcJobTag the tag of AcJob
	AcJobTag = "group-name"
	// AcJobVersion the api version of AcJob
	AcJobVersion = "mindxdl.gitee.com"
	// SubHealthFault subHealth code
	SubHealthFault = "SubHealthFault"
)

const (

	// SuperPodAnnoKey annotation key of super pod
	SuperPodAnnoKey = "sp-block"
)