Skip to main content

Kubernetes Training Jobs

Monitor Kubernetes training jobs

To view the dashboard, go to the Training Jobs dashboard.

Info

For accessing CoreWeave Grafana Dashboards instructions, see Access CoreWeave Grafana Dashboards.

The Kubernetes Training Job dashboard helps monitor hardware resources. It shows metrics for GPU utilization, network bandwidth (like InfiniBand), and storage I/O (both local and NFS), helping to diagnose performance bottlenecks and ensure compute-intensive tasks are running efficiently.

Panel TitleDescription
KindShows the Kubernetes object kind.
NameShows the name of the resource.
NodesShows the total number of Nodes.
PodsShows the total number of Pods.
UptimeShows the overall uptime.
Pod Readiness TimelineShows the readiness state of the Pods.
Active GPUsShows the number of active GPUs.
Job EfficiencyShows the efficiency of running jobs.
Current FP8 FLOPSDisplays the current floating-point operations per second in FP8 precision.
Node conditionsDisplays the current conditions of the Nodes.
AlertsDisplays active alerts related to this resource.
Nodes (Range)Shows the individual Pods, their running status, and uptime on each Node.
GPU Temperatures Running JobsShows the GPU temperatures for running jobs.
GPU Core UtilizationShows GPU core usage.
SM UtilizationShows the utilization of streaming multiprocessors on the GPU.
GPU Mem CopyShows GPU memory copy operations.
Tensor Core UtilShows the utilization of Tensor Cores.
Current FP8Shows the current FP8 performance.
VRAM UsageDisplays the video RAM usage.
GPUs TemperatureDisplays the temperature of the GPUs.
InfiniBand Aggregate BandwidthShows the total network bandwidth over the InfiniBand interconnect.
GPUs Power UsageDisplays the power consumption of the GPUs.
Local Max Disk I/O Utilization (1m)Shows the maximum disk I/O utilization on the local disk over 1 minute.
Local Avg Bytes Read / Written Per NodShows the average bytes read/written per Node on the local disk.
Local Total Bytes Read / Written (2m)Shows the total bytes read/written on the local disk over 2 minutes.
Local Total Read / Write Rate (2m)Shows the total read/write rate on the local disk over 2 minutes.
NFS Average Request Time by OperationShows duration requests took from when a request was enqueued to when it was completely handled for a given operation, in seconds.
NFS Avg Bytes Read / Written Per NodeShows the average bytes read/written per Node on the NFS.
NFS Total Bytes Read / Written (2m)Shows the total bytes read/written on the NFS over 2 minutes.
NFS Total Read / Write Rate (2m)Shows the total read/write rate on the NFS over 2 minutes.
NFS Average Response Time by OperationShows duration requests took to get a reply back after a request for a given operation was transmitted, in seconds.
NFS Avg Write Rate Per Active Node (2m)Shows the average NFS write rate per active Node. Only includes Nodes reading/writing over 10KB/s.
NFS Avg Read Rate Per Active Node (2m)Shows the average NFS read rate per active Node. Only includes Nodes reading/writing over 10KB/s.
NFS Nodes with RetransmissionsShows the count of NFS Nodes experiencing network retransmissions.

Node alerts

Alert NameDescription
DCGMSRAMThresholdExceededThis alert indicates that the SRAM threshold has been exceeded on a GPU. This indicates a memory issue and requires investigation by reliability teams.
DPUContainerdThreadExhaustionThe DPUContainerdThreadExhaustion alert indicates that the containerd process has run out of threads on the DPU. This requires an update to the dpu-health container to patch.
DPUContainerdThreadExhaustionCPXThe DPUContainerdThreadExhaustion alert indicates that the containerd process has run out of threads on the DPU. This requires an update to the dpu-health container to patch.
DPULinkFlappingCPXThe DPULinkFlapping alert indicates that a DPU (Data Processing Unit) link has become unstable. It specifically triggers when a link on a DPU flaps (goes up and down) multiple times within a monitoring period.
DPUNetworkFrameErrsThe DPUNetworkFrameErrs alert indicates frame errors occurring on DPU (Data Processing Unit) network interfaces. These errors typically indicate a problem with the physical network link.
DPURouteCountMismatchThe DPURouteCountMismatch alert indicates an inconsistency in routes between what the DPU learns and has installed. A software component on the DPU will need to be restarted.
DPURouteLoopThe DPURouteLoop alert indicates that a route loop has been detected on the DPU. This can be caused by a miscabling issue in the data center.
DPURouteLoopCPXThe DPURouteLoop alert indicates that a route loop has been detected on the DPU. This can be caused by a miscabling issue in the data center.
DPUUnexpectedPuntedRoutesThe DPUUnexpectedPuntedRoutes alert indicates a failure in offloading which can cause connectivity issues for the host. Node will be automatically reset to restore proper connectivity.
DPUUnexpectedPuntedRoutesCPXThe DPUUnexpectedPuntedRoutes alert indicates a failure in offloading which can cause connectivity issues for the host. The issue typically occurs after a power reset (when the host reboots without the DPU rebooting).
ECCDoubleVolatileErrorsECCDoubleVolatileErrors is an alert that indicates when DCGM double-bit volatile ECC (Error Correction Code) errors are increasing over a 5-minute period on a GPU.
GPUContainedECCErrorGPU Contained ECC Error (Xid 94) indicates a uncorrectable memory error was encountered and contained. Workload has been impacted but the node is generally healthy. No action needed.
GPUECCUncorrectableErrorUncontainedGPU Uncorrectable Error Uncontained (Xid 95) indicates a uncorrectable memory error was encountered but not successfully contained. Workload has been impacted and the node will be restarted.
GPUFallenOffBusGPU Fallen Off The Bus (Xid 79) indicates a fatal hardware error where the GPU shuts down and is completely inaccessible from the system. The node will immediately and automatically be taken out of service.
GPUFallenOffBusHGXGPU Fallen Off The Bus (Xid 79) indicates a fatal hardware error where the GPU shuts down and is completely inaccessible from the system. The node will immediately and automatically be taken out of service.
GPUNVLinkSWDefinedErrorNVLink SW Defined Error (Xid 155) indicates link down events which are flagged as intentional will trigger this Xid. Node will be reset.
GPUPGraphicsEngineErrorGPU Graphics Enginer Error (Xid 69) has impacted the workload but the node is generally healthy. No action needed.
GPUPRowRemapFailureGPU Row Remap Failure (Xid 64) is caused by a uncorrectable error resulting in a GPU memory remapping event that failed. The node will immediately and automatically be taken out of service.
GPUTimeoutErrorGPU Timeout Error (Xid 46) indicates GPU stopped processing and the node will be restarted.
GPUUncorrectableDRAMErrorGPU Uncorrectable DRAM Error (Xid 171) provides complementary information to Xid 48. No action is needed.
GPUUncorrectableSRAMErrorGPU Uncorrectable SRAM Error (Xid 172) provides complementary information to Xid 48. No action is needed.
GPUVeryHotThe GPUVeryHot alert triggers when a GPU's temperature exceeds 90°C.
KubeNodeNotReadyThe KubeNodeNotReady alert indicates when a node's status condition is not Ready in a Kubernetes cluster. This alert can be an indicator of critical system health issues.
KubeNodeNotReadyHGXThe KubeNodeNotReadyHGX alert indicates that a node has been unready or offline for more than 15 minutes.
ManyUCESingleBankH100The ManyUCESingleBankH100 alert triggers when there are two or more DRAM Uncorrectable Errors (UCEs) on the same row remapper bank of an H100 GPU.
MetalDevRedfishErrorThe MetalDevRedfishError alert indicates an out-of-band action against a BMC failed.
NVL72GPUHighFECCKSThe NVL72GPUHighFECCKS alert indicates that a GPU is observing a high rate of forward error correction indicating signal integrity issues.
NVL72SwitchHighFECCKSThe NVL72SwitchHighFECCKS alert indicates that a NVSwitch is observing a high rate of forward error correction indicating signal integrity issues.
NVLinkDomainFullyTriagedNVLinkDomainFullyTriaged indicates rack is entirely triaged. This rack should either be investigated for an unexpected rack level event or returned to fleet.
NVLinkDomainProductionNodeCountLowNVLinkDomainDegraded indicates rack has less nodes in a production state than expected. This rack will need manual intervention to either restore capacity or reclaim for further triage.
NodeBackendLinkFaultThe NodeBackendLinkFault alert indicates that the backend bandwidth is degraded and the interface may be potentially lost.
NodeBackendMisconnectedNode-to-leaf ports are either missing or incorrectly connected.
NodeCPUHZThrottleLongAn extended period of CPU frequency throttling has occured. CPU throttling most often occurs due to power delivery or thermal node level problems. The node will immediately and automatically be taken out of service and the job interrupted.
NodeGPUNVLinkDownThe node is experiencing NVLink issues and will be automatically triaged.
NodeGPUXID149NVSwitchA GPU has experienced a fatal NVLink error. The node will be restarted to recover the GPU.
NodeGPUXID149s4aLinkIssueFordPintoRepeatedA GPU has experienced a fatal NVLink error. This is a frequent offender and automation will remove the node from the cluster.
NodeGPUXID149s4aLinkIssueLamboRepeatedA GPU has experienced a fatal NVLink error. This is a frequent offender and automation will remove the node from the cluster.
NodeGPUXID149s4aLinkIssueNeedsUpgradeRepeatedA GPU has experienced a fatal NVLink error. This is a frequent offender and automation will remove the node from the cluster.
NodeLoadAverageHighThe NodeLoadAverageHigh alert triggers when a node's load average exceeds 1000 for more than 15 minutes.
NodeMemoryErrorThe NodeMemoryError alert indicates that a node has one or more bad DIMM (memory) modules.
NodeNetworkReceiveErrsNodeNetworkReceiveErrs alert indicates that a network interface has encountered receive errors exceeding a 1% threshold over a 2-minute period for 1 hour.
NodePCIErrorH100GPUThe NodePCIErrorH100GPU alert indicates when a GPU is experiencing PCI bus communication errors.
NodePCIErrorH100PLXThe NodePCIErrorH100PLX alert indicates a high rate of PCIe bus errors occurring on the PLX switch that connects H100 GPUs.
NodeRepeatUCEThe NodeRepeatUCE alert indicates that a node has experienced frequent GPU Uncorrectable ECC (UCE) errors.
NodeVerificationFailureNVFabricThe node is experiencing NVLink issues and will be automatically triaged.
NodeVerificationMegatronDeadlockHPC-Perftest failed megatron_lm test due to possible deadlock. Node should be triaged.
PendingStateExtendedTimeThe PendingStateExtendedTime alert indicates that a node has been in a pending state for an extended period of time. This alert helps identify nodes that need to be removed from their current state but are stuck for an extended time.
PendingStateExtendedTimeLowGpuUtilThe PendingStateExtendedTimeLowGpuUtil alert triggers when a node has been in a pending state for more than 10 days and has had less than 1% GPU utilization in the last hour. This alert helps indicate if a node needs to be removed from its current state but has been stuck for an extended time.
UnknownNVMLErrorOnContainerStartThe UnknownNVMLErrorOnContainerStart alert typically indicates that a GPU has fallen off the bus or is experiencing hardware issues.