{ "ServerPort": 5000, "CommonConfig": { "MetricsFieldPrefix": "amd_", "HealthService": { "Enable" : true, "PollingRate": "30s" }, "Logging": { "Level": "info", "MaxFileSizeMB": 10, "MaxBackups": 3, "MaxAgeDays": 7 } }, "GPUConfig": { "Fields": [ "GPU_NODES_TOTAL", "GPU_PACKAGE_POWER", "GPU_AVERAGE_PACKAGE_POWER", "GPU_EDGE_TEMPERATURE", "GPU_JUNCTION_TEMPERATURE", "GPU_MEMORY_TEMPERATURE", "GPU_HBM_TEMPERATURE", "GPU_GFX_ACTIVITY", "GPU_UMC_ACTIVITY", "GPU_MMA_ACTIVITY", "GPU_VCN_ACTIVITY", "GPU_JPEG_ACTIVITY", "GPU_VOLTAGE", "GPU_GFX_VOLTAGE", "GPU_MEMORY_VOLTAGE", "PCIE_SPEED", "PCIE_MAX_SPEED", "PCIE_BANDWIDTH", "GPU_ENERGY_CONSUMED", "PCIE_REPLAY_COUNT", "PCIE_RECOVERY_COUNT", "PCIE_REPLAY_ROLLOVER_COUNT", "PCIE_NACK_SENT_COUNT", "PCIE_NACK_RECEIVED_COUNT", "GPU_CLOCK", "GPU_POWER_USAGE", "GPU_TOTAL_VRAM", "GPU_ECC_CORRECT_TOTAL", "GPU_ECC_UNCORRECT_TOTAL", "GPU_ECC_CORRECT_SDMA", "GPU_ECC_UNCORRECT_SDMA", "GPU_ECC_CORRECT_GFX", "GPU_ECC_UNCORRECT_GFX", "GPU_ECC_CORRECT_MMHUB", "GPU_ECC_UNCORRECT_MMHUB", "GPU_ECC_CORRECT_ATHUB", "GPU_ECC_UNCORRECT_ATHUB", "GPU_ECC_CORRECT_BIF", "GPU_ECC_UNCORRECT_BIF", "GPU_ECC_CORRECT_HDP", "GPU_ECC_UNCORRECT_HDP", "GPU_ECC_CORRECT_XGMI_WAFL", "GPU_ECC_UNCORRECT_XGMI_WAFL", "GPU_ECC_CORRECT_DF", "GPU_ECC_UNCORRECT_DF", "GPU_ECC_CORRECT_SMN", "GPU_ECC_UNCORRECT_SMN", "GPU_ECC_CORRECT_SEM", "GPU_ECC_UNCORRECT_SEM", "GPU_ECC_CORRECT_MP0", "GPU_ECC_UNCORRECT_MP0", "GPU_ECC_CORRECT_MP1", "GPU_ECC_UNCORRECT_MP1", "GPU_ECC_CORRECT_FUSE", "GPU_ECC_UNCORRECT_FUSE", "GPU_ECC_CORRECT_UMC", "GPU_ECC_UNCORRECT_UMC", "GPU_XGMI_NBR_0_NOP_TX", "GPU_XGMI_NBR_0_REQ_TX", "GPU_XGMI_NBR_0_RESP_TX", "GPU_XGMI_NBR_0_BEATS_TX", "GPU_XGMI_NBR_1_NOP_TX", "GPU_XGMI_NBR_1_REQ_TX", "GPU_XGMI_NBR_1_RESP_TX", "GPU_XGMI_NBR_1_BEATS_TX", "GPU_XGMI_NBR_0_TX_THRPUT", "GPU_XGMI_NBR_1_TX_THRPUT", "GPU_XGMI_NBR_2_TX_THRPUT", "GPU_XGMI_NBR_3_TX_THRPUT", "GPU_XGMI_NBR_4_TX_THRPUT", "GPU_XGMI_NBR_5_TX_THRPUT", "GPU_USED_VRAM", "GPU_FREE_VRAM", "GPU_TOTAL_VISIBLE_VRAM", "GPU_USED_VISIBLE_VRAM", "GPU_FREE_VISIBLE_VRAM", "GPU_TOTAL_GTT", "GPU_USED_GTT", "GPU_FREE_GTT", "GPU_ECC_CORRECT_MCA", "GPU_ECC_UNCORRECT_MCA", "GPU_ECC_CORRECT_VCN", "GPU_ECC_UNCORRECT_VCN", "GPU_ECC_CORRECT_JPEG", "GPU_ECC_UNCORRECT_JPEG", "GPU_ECC_CORRECT_IH", "GPU_ECC_UNCORRECT_IH", "GPU_ECC_CORRECT_MPIO", "GPU_ECC_UNCORRECT_MPIO", "GPU_HEALTH", "GPU_XGMI_LINK_RX", "GPU_XGMI_LINK_TX", "GPU_VIOLATION_CURRENT_ACCUMULATED_COUNTER", "GPU_VIOLATION_PROCESSOR_HOT_RESIDENCY_ACCUMULATED", "GPU_VIOLATION_PPT_RESIDENCY_ACCUMULATED", "GPU_VIOLATION_SOCKET_THERMAL_RESIDENCY_ACCUMULATED", "GPU_VIOLATION_VR_THERMAL_RESIDENCY_ACCUMULATED", "GPU_VIOLATION_HBM_THERMAL_RESIDENCY_ACCUMULATED", "GPU_GFX_BUSY_INSTANTANEOUS", "GPU_VCN_BUSY_INSTANTANEOUS", "GPU_JPEG_BUSY_INSTANTANEOUS", "GPU_PROF_GRBM_GUI_ACTIVE", "GPU_PROF_SQ_WAVES", "GPU_PROF_GRBM_COUNT", "GPU_PROF_CPC_CPC_STAT_BUSY", "GPU_PROF_CPC_CPC_STAT_IDLE", "GPU_PROF_CPC_CPC_STAT_STALL", "GPU_PROF_CPC_CPC_TCIU_BUSY", "GPU_PROF_CPC_CPC_TCIU_IDLE", "GPU_PROF_CPC_CPC_UTCL2IU_BUSY", "GPU_PROF_CPC_CPC_UTCL2IU_IDLE", "GPU_PROF_CPC_CPC_UTCL2IU_STALL", "GPU_PROF_CPC_ME1_BUSY_FOR_PACKET_DECODE", "GPU_PROF_CPC_ME1_DC0_SPI_BUSY", "GPU_PROF_CPC_UTCL1_STALL_ON_TRANSLATION", "GPU_PROF_CPC_ALWAYS_COUNT", "GPU_PROF_CPC_ADC_VALID_CHUNK_NOT_AVAIL", "GPU_PROF_CPC_ADC_DISPATCH_ALLOC_DONE", "GPU_PROF_CPC_ADC_VALID_CHUNK_END", "GPU_PROF_CPC_SYNC_FIFO_FULL_LEVEL", "GPU_PROF_CPC_SYNC_FIFO_FULL", "GPU_PROF_CPC_GD_BUSY", "GPU_PROF_CPC_TG_SEND", "GPU_PROF_CPC_WALK_NEXT_CHUNK", "GPU_PROF_CPC_STALLED_BY_SE0_SPI", "GPU_PROF_CPC_STALLED_BY_SE1_SPI", "GPU_PROF_CPC_STALLED_BY_SE2_SPI", "GPU_PROF_CPC_STALLED_BY_SE3_SPI", "GPU_PROF_CPC_LTE_ALL", "GPU_PROF_CPC_SYNC_WRREQ_FIFO_BUSY", "GPU_PROF_CPC_CANE_BUSY", "GPU_PROF_CPC_CANE_STALL", "GPU_PROF_CPF_CMP_UTCL1_STALL_ON_TRANSLATION", "GPU_PROF_CPF_CPF_STAT_BUSY", "GPU_PROF_CPF_CPF_STAT_IDLE", "GPU_PROF_CPF_CPF_STAT_STALL", "GPU_PROF_CPF_CPF_TCIU_BUSY", "GPU_PROF_CPF_CPF_TCIU_IDLE", "GPU_PROF_CPF_CPF_TCIU_STALL", "GPU_PROF_FETCH_SIZE", "GPU_PROF_WRITE_SIZE", "GPU_PROF_TOTAL_16_OPS", "GPU_PROF_TOTAL_32_OPS", "GPU_PROF_TOTAL_64_OPS", "GPU_PROF_GUI_UTIL_PERCENT", "GPU_PROF_OCCUPANCY_PERCENT", "GPU_PROF_TENSOR_ACTIVE_PERCENT", "GPU_PROF_VALU_PIPE_ISSUE_UTIL", "GPU_PROF_SM_ACTIVE", "GPU_PROF_OCCUPANCY_ELAPSED", "GPU_PROF_OCCUPANCY_PER_ACTIVE_CU", "GPU_PROF_OCCUPANCY_PER_CU", "GPU_PROF_SIMD_UTILIZATION", "PCIE_RX", "PCIE_TX", "PCIE_BIDIRECTIONAL_BANDWIDTH", "GPU_AFID_ERRORS", "GPU_MIN_CLOCK", "GPU_MAX_CLOCK", "GPU_VIOLATION_PROCESSOR_HOT_RESIDENCY_PERCENTAGE", "GPU_VIOLATION_PPT_RESIDENCY_PERCENTAGE", "GPU_VIOLATION_SOCKET_THERMAL_RESIDENCY_PERCENTAGE", "GPU_VIOLATION_VR_THERMAL_RESIDENCY_PERCENTAGE", "GPU_VIOLATION_HBM_THERMAL_RESIDENCY_PERCENTAGE", "GPU_VIOLATION_GFX_CLOCK_BELOW_HOST_LIMIT_POWER_ACCUMULATED", "GPU_VIOLATION_GFX_CLOCK_BELOW_HOST_LIMIT_THERMAL_ACCUMULATED", "GPU_VIOLATION_GFX_CLOCK_LOW_UTILIZATION_ACCUMULATED", "GPU_VIOLATION_GFX_CLOCK_BELOW_HOST_LIMIT_TOTAL_ACCUMULATED", "GPU_VIOLATION_GFX_CLOCK_BELOW_HOST_LIMIT_POWER_PERCENTAGE", "GPU_VIOLATION_GFX_CLOCK_BELOW_HOST_LIMIT_THERMAL_PERCENTAGE", "GPU_VIOLATION_GFX_CLOCK_LOW_UTILIZATION_PERCENTAGE", "GPU_VIOLATION_GFX_CLOCK_BELOW_HOST_LIMIT_TOTAL_PERCENTAGE" ], "Labels": [ "GPU_UUID", "SERIAL_NUMBER", "GPU_ID", "POD", "NAMESPACE", "CONTAINER", "JOB_ID", "JOB_USER", "JOB_PARTITION", "CLUSTER_NAME", "CARD_SERIES", "CARD_MODEL", "CARD_VENDOR", "DRIVER_VERSION", "VBIOS_VERSION", "HOSTNAME", "GPU_PARTITION_ID", "GPU_COMPUTE_PARTITION_TYPE", "GPU_MEMORY_PARTITION_TYPE", "KFD_PROCESS_ID", "DEPLOYMENT_MODE" ], "HealthThresholds" : { "GPU_ECC_UNCORRECT_SDMA" : 0, "GPU_ECC_UNCORRECT_GFX" : 0, "GPU_ECC_UNCORRECT_MMHUB" : 0, "GPU_ECC_UNCORRECT_ATHUB" : 0, "GPU_ECC_UNCORRECT_BIF" : 0, "GPU_ECC_UNCORRECT_HDP" : 0, "GPU_ECC_UNCORRECT_XGMI_WAFL" : 0, "GPU_ECC_UNCORRECT_DF" : 0, "GPU_ECC_UNCORRECT_SMN" : 0, "GPU_ECC_UNCORRECT_SEM" : 0, "GPU_ECC_UNCORRECT_MP0" : 0, "GPU_ECC_UNCORRECT_MP1" : 0, "GPU_ECC_UNCORRECT_FUSE" : 0, "GPU_ECC_UNCORRECT_UMC" : 0, "GPU_ECC_UNCORRECT_MCA" : 0, "GPU_ECC_UNCORRECT_VCN" : 0, "GPU_ECC_UNCORRECT_JPEG" : 0, "GPU_ECC_UNCORRECT_IH" : 0, "GPU_ECC_UNCORRECT_MPIO" : 0 }, "CustomLabels" : { "CLUSTER_NAME" : "amdgpu-k8s-metrics-exporter" }, "ExtraPodLabels" : { "WORKLOAD_ID" : "amd-workload-id", "USERGROUP_ID" : "amd-usergroup-id" }, "ProfilerMetrics" : { "all" : false, "hostname1" : false }, "ProfilerConfig": { "SamplingInterval": 1000, "PtlDelay" : 0 } }, "NICConfig": { "Fields": [ "NIC_PORT_STATS_FRAMES_RX_OK", "NIC_PORT_STATS_FRAMES_RX_ALL", "NIC_PORT_STATS_FRAMES_RX_BAD_FCS", "NIC_PORT_STATS_FRAMES_RX_BAD_ALL", "NIC_PORT_STATS_FRAMES_RX_PAUSE", "NIC_PORT_STATS_FRAMES_RX_BAD_LENGTH", "NIC_PORT_STATS_FRAMES_RX_UNDERSIZED", "NIC_PORT_STATS_FRAMES_RX_OVERSIZED", "NIC_PORT_STATS_FRAMES_RX_FRAGMENTS", "NIC_PORT_STATS_FRAMES_RX_JABBER", "NIC_PORT_STATS_FRAMES_RX_PRIPAUSE", "NIC_PORT_STATS_FRAMES_RX_STOMPED_CRC", "NIC_PORT_STATS_FRAMES_RX_TOO_LONG", "NIC_PORT_STATS_FRAMES_RX_DROPPED", "NIC_PORT_STATS_FRAMES_RX_UNICAST", "NIC_PORT_STATS_FRAMES_RX_MULTICAST", "NIC_PORT_STATS_FRAMES_RX_BROADCAST", "NIC_PORT_STATS_FRAMES_RX_PRI_0", "NIC_PORT_STATS_FRAMES_RX_PRI_1", "NIC_PORT_STATS_FRAMES_RX_PRI_2", "NIC_PORT_STATS_FRAMES_RX_PRI_3", "NIC_PORT_STATS_FRAMES_RX_PRI_4", "NIC_PORT_STATS_FRAMES_RX_PRI_5", "NIC_PORT_STATS_FRAMES_RX_PRI_6", "NIC_PORT_STATS_FRAMES_RX_PRI_7", "NIC_PORT_STATS_FRAMES_TX_OK", "NIC_PORT_STATS_FRAMES_TX_ALL", "NIC_PORT_STATS_FRAMES_TX_BAD", "NIC_PORT_STATS_FRAMES_TX_PAUSE", "NIC_PORT_STATS_FRAMES_TX_PRIPAUSE", "NIC_PORT_STATS_FRAMES_TX_LESS_THAN_64B", "NIC_PORT_STATS_FRAMES_TX_TRUNCATED", "NIC_PORT_STATS_FRAMES_TX_UNICAST", "NIC_PORT_STATS_FRAMES_TX_MULTICAST", "NIC_PORT_STATS_FRAMES_TX_BROADCAST", "NIC_PORT_STATS_FRAMES_TX_PRI_0", "NIC_PORT_STATS_FRAMES_TX_PRI_1", "NIC_PORT_STATS_FRAMES_TX_PRI_2", "NIC_PORT_STATS_FRAMES_TX_PRI_3", "NIC_PORT_STATS_FRAMES_TX_PRI_4", "NIC_PORT_STATS_FRAMES_TX_PRI_5", "NIC_PORT_STATS_FRAMES_TX_PRI_6", "NIC_PORT_STATS_FRAMES_TX_PRI_7", "NIC_PORT_STATS_OCTETS_RX_OK", "NIC_PORT_STATS_OCTETS_RX_ALL", "NIC_PORT_STATS_OCTETS_TX_OK", "NIC_PORT_STATS_OCTETS_TX_ALL", "NIC_PORT_STATS_RSFEC_CORRECTABLE_WORD", "NIC_PORT_STATS_RSFEC_CH_SYMBOL_ERR_CNT", "NIC_LIF_STATS_RX_UNICAST_PACKETS", "NIC_LIF_STATS_RX_UNICAST_DROP_PACKETS", "NIC_LIF_STATS_RX_MULTICAST_DROP_PACKETS", "NIC_LIF_STATS_RX_BROADCAST_DROP_PACKETS", "NIC_LIF_STATS_RX_DMA_ERRORS", "NIC_LIF_STATS_TX_UNICAST_PACKETS", "NIC_LIF_STATS_TX_UNICAST_DROP_PACKETS", "NIC_LIF_STATS_TX_MULTICAST_DROP_PACKETS", "NIC_LIF_STATS_TX_BROADCAST_DROP_PACKETS", "NIC_LIF_STATS_TX_DMA_ERRORS", "RDMA_TX_UCAST_PKTS", "RDMA_TX_CNP_PKTS", "RDMA_RX_UCAST_PKTS", "RDMA_RX_CNP_PKTS", "RDMA_RX_ECN_PKTS", "RDMA_REQ_RX_PKT_SEQ_ERR", "RDMA_REQ_RX_RNR_RETRY_ERR", "RDMA_REQ_RX_RMT_ACC_ERR", "RDMA_REQ_RX_RMT_REQ_ERR", "RDMA_REQ_RX_OPER_ERR", "RDMA_REQ_RX_IMPL_NAK_SEQ_ERR", "RDMA_REQ_RX_CQE_ERR", "RDMA_REQ_RX_CQE_FLUSH", "RDMA_REQ_RX_DUP_RESP", "RDMA_REQ_RX_INVALID_PKTS", "RDMA_REQ_TX_LOC_ERR", "RDMA_REQ_TX_LOC_OPER_ERR", "RDMA_REQ_TX_MEM_MGMT_ERR", "RDMA_REQ_TX_RETRY_EXCD_ERR", "RDMA_REQ_TX_LOC_SGL_INV_ERR", "RDMA_RESP_RX_DUP_REQUEST", "RDMA_RESP_RX_OUTOF_BUF", "RDMA_RESP_RX_OUTOUF_SEQ", "RDMA_RESP_RX_CQE_ERR", "RDMA_RESP_RX_CQE_FLUSH", "RDMA_RESP_RX_LOC_LEN_ERR", "RDMA_RESP_RX_INVALID_REQUEST", "RDMA_RESP_RX_LOC_OPER_ERR", "RDMA_RESP_RX_OUTOF_ATOMIC", "RDMA_RESP_TX_PKT_SEQ_ERR", "RDMA_RESP_TX_RMT_INVAL_REQ_ERR", "RDMA_RESP_TX_RMT_ACC_ERR", "RDMA_RESP_TX_RMT_OPER_ERR", "RDMA_RESP_TX_RNR_RETRY_ERR", "RDMA_RESP_TX_LOC_SGL_INV_ERR", "RDMA_RESP_RX_S0_TABLE_ERR", "QP_SQ_REQ_TX_NUM_PACKET", "QP_SQ_REQ_TX_NUM_SEND_MSGS_WITH_RKE", "QP_SQ_REQ_TX_NUM_LOCAL_ACK_TIMEOUTS", "QP_SQ_REQ_TX_RNR_TIMEOUT", "QP_SQ_REQ_TX_TIMES_SQ_DRAINED", "QP_SQ_REQ_TX_NUM_CNP_SENT", "QP_SQ_REQ_RX_NUM_PACKET", "QP_SQ_REQ_RX_NUM_PKTS_WITH_ECN_MARKING", "QP_SQ_QCN_CURR_BYTE_COUNTER", "QP_SQ_QCN_NUM_BYTE_COUNTER_EXPIRED", "QP_SQ_QCN_NUM_TIMER_EXPIRED", "QP_SQ_QCN_NUM_ALPHA_TIMER_EXPIRED", "QP_SQ_QCN_NUM_CNP_RCVD", "QP_SQ_QCN_NUM_CNP_PROCESSED", "QP_RQ_RSP_TX_NUM_PACKET", "QP_RQ_RSP_TX_RNR_ERROR", "QP_RQ_RSP_TX_NUM_SEQUENCE_ERROR", "QP_RQ_RSP_TX_NUM_RP_BYTE_THRES_HIT", "QP_RQ_RSP_TX_NUM_RP_MAX_RATE_HIT", "QP_RQ_RSP_RX_NUM_PACKET", "QP_RQ_RSP_RX_NUM_SEND_MSGS_WITH_RKE", "QP_RQ_RSP_RX_NUM_PKTS_WITH_ECN_MARKING", "QP_RQ_RSP_RX_NUM_CNPS_RECEIVED", "QP_RQ_RSP_RX_MAX_RECIRC_EXCEEDED_DROP", "QP_RQ_RSP_RX_NUM_MEM_WINDOW_INVALID", "QP_RQ_RSP_RX_NUM_DUPL_WITH_WR_SEND_OPC", "QP_RQ_RSP_RX_NUM_DUPL_READ_BACKTRACK", "QP_RQ_RSP_RX_NUM_DUPL_READ_ATOMIC_DROP", "QP_RQ_QCN_CURR_BYTE_COUNTER", "QP_RQ_QCN_NUM_BYTE_COUNTER_EXPIRED", "QP_RQ_QCN_NUM_TIMER_EXPIRED", "QP_RQ_QCN_NUM_ALPHA_TIMER_EXPIRED", "QP_RQ_QCN_NUM_CNP_RCVD", "QP_RQ_QCN_NUM_CNP_PROCESSED", "ETH_TX_PACKETS", "ETH_TX_BYTES", "ETH_RX_PACKETS", "ETH_RX_BYTES", "ETH_FRAMES_RX_BROADCAST", "ETH_FRAMES_RX_MULTICAST", "ETH_FRAMES_TX_BROADCAST", "ETH_FRAMES_TX_MULTICAST", "ETH_FRAMES_RX_PAUSE", "ETH_FRAMES_TX_PAUSE", "ETH_FRAMES_RX_64B", "ETH_FRAMES_RX_65B_127B", "ETH_FRAMES_RX_128B_255B", "ETH_FRAMES_RX_256B_511B", "ETH_FRAMES_RX_512B_1023B", "ETH_FRAMES_RX_1024B_1518B", "ETH_FRAMES_RX_1519B_2047B", "ETH_FRAMES_RX_2048B_4095B", "ETH_FRAMES_RX_4096B_8191B", "ETH_FRAMES_RX_BAD_FCS", "ETH_FRAMES_RX_PRI0", "ETH_FRAMES_RX_PRI1", "ETH_FRAMES_RX_PRI2", "ETH_FRAMES_RX_PRI3", "ETH_FRAMES_RX_PRI4", "ETH_FRAMES_RX_PRI5", "ETH_FRAMES_RX_PRI6", "ETH_FRAMES_RX_PRI7", "ETH_FRAMES_TX_PRI0", "ETH_FRAMES_TX_PRI1", "ETH_FRAMES_TX_PRI2", "ETH_FRAMES_TX_PRI3", "ETH_FRAMES_TX_PRI4", "ETH_FRAMES_TX_PRI5", "ETH_FRAMES_TX_PRI6", "ETH_FRAMES_TX_PRI7", "ETH_FRAMES_RX_DROPPED", "ETH_FRAMES_RX_ALL", "ETH_FRAMES_RX_BAD_ALL", "ETH_FRAMES_TX_ALL", "ETH_FRAMES_TX_BAD", "ETH_HW_TX_DROPPED", "ETH_HW_RX_DROPPED", "ETH_RX_0_DROPPED", "ETH_RX_1_DROPPED", "ETH_RX_2_DROPPED", "ETH_RX_3_DROPPED", "ETH_RX_4_DROPPED", "ETH_RX_5_DROPPED", "ETH_RX_6_DROPPED", "ETH_RX_7_DROPPED", "ETH_RX_8_DROPPED", "ETH_RX_9_DROPPED", "ETH_RX_10_DROPPED", "ETH_RX_11_DROPPED", "ETH_RX_12_DROPPED", "ETH_RX_13_DROPPED", "ETH_RX_14_DROPPED", "ETH_RX_15_DROPPED", "ETH_FRAMES_RX_OK", "ETH_FRAMES_TX_OK", "ETH_OCTETS_RX_OK", "ETH_OCTETS_TX_OK", "ETH_OCTETS_TX_TOTAL", "ETH_FRAMES_RX_UNICAST", "ETH_FRAMES_TX_UNICAST", "ETH_FRAMES_RX_8192B_9215B", "ETH_FRAMES_TX_8192B_9215B", "ETH_FRAMES_TX_64B", "ETH_FRAMES_TX_65B_127B", "ETH_FRAMES_TX_128B_255B", "ETH_FRAMES_TX_256B_511B", "ETH_FRAMES_TX_512B_1023B", "ETH_FRAMES_TX_1024B_1518B", "ETH_FRAMES_TX_1519B_2047B", "ETH_FRAMES_TX_2048B_4095B", "ETH_FRAMES_TX_4096B_8191B" ], "Labels":[ "NIC_UUID", "NIC_ID", "HOSTNAME", "SERIAL_NUMBER", "POD", "NAMESPACE", "CONTAINER", "FIRMWARE_VERSION" ], "CustomLabels": { "CLUSTER_NAME" : "amdnetwork-k8s-metrics-exporter" }, "ExtraPodLabels" : { "WORKLOAD_ID" : "amd-workload-id", "USERGROUP_ID" : "amd-usergroup-id" }, "HealthCheckConfig": { "InterfaceAdminDownAsUnhealthy": false } }, "IFOEConfig": { "Labels": [ "GPU_UUID", "HOSTNAME" ], "Fields": [ "IFOE_TOTAL_DEVICES", "IFOE_TOTAL_STATIONS", "IFOE_TOTAL_PORTS", "IFOE_NUMBER_FAILEDOVER_STREAMS", "IFOE_NUMBER_PAUSED_STREAMS", "IFOE_BIT_ERROR_RATE", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS0", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS1", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS2", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS3", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS4", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS5", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS6", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS7", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS8", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS9", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS10", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS11", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS12", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS13", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS14", "IFOE_FEC_CODEWORD_SYMBOL_ERRORS15" ], "CustomLabels": { "CLUSTER_NAME": "amdifoe-k8s-metrics-exporter" }, "ExtraPodLabels": { "WORKLOAD_ID": "amd-workload-id", "USERGROUP_ID": "amd-usergroup-id" } } }