Add two missing metrics

Signed-off-by: Xiaodong Ye <[email protected]>
prometheus · Nov 19, 2024 · 087adf3 · 087adf3
1 parent e35eb3c
commit 087adf3
Showing 1 changed file with 32 additions and 30 deletions.
diff --git a/collector/rdma_linux.go b/collector/rdma_linux.go
@@ -40,36 +40,38 @@ var (
 	rdmaIncludedMetrics = kingpin.Flag("collector.rdma.metrics-include", "Regexp of rdma stats to include.").Default(".*").String()
 
 	rdmaHwCounters = map[string]string{
-		"roce_slow_restart_cnps":     "RDMA RoCE slow restart CNPS",
-		"rp_cnp_ignored":             "RDMA RP CNP ignored",
-		"roce_adp_retrans_to":        "RDMA RoCE adaptive retransmission timeout",
-		"rx_icrc_encapsulated":       "RDMA RX ICRC encapsulated",
-		"resp_local_length_error":    "RDMA response local length error",
-		"np_ecn_marked_roce_packets": "RDMA NP ECN marked RoCE packets",
-		"roce_slow_restart_trans":    "RDMA RoCE slow restart transactions",
-		"req_remote_invalid_request": "RDMA request remote invalid request",
-		"local_ack_timeout_err":      "RDMA local ACK timeout error",
-		"lifespan":                   "RDMA lifespan",
-		"req_cqe_error":              "RDMA request CQE error",
-		"rnr_nak_retry_err":          "RDMA RNR NAK retry error",
-		"np_cnp_sent":                "RDMA NP CNP sent",
-		"rx_dct_connect":             "RDMA RX DCT connect",
-		"rp_cnp_handled":             "RDMA RP CNP handled",
-		"implied_nak_seq_err":        "RDMA implied NAK sequence error",
-		"roce_slow_restart":          "RDMA RoCE slow restart",
-		"req_cqe_flush_error":        "RDMA request CQE flush error",
-		"packet_seq_err":             "RDMA packet sequence error",
-		"duplicate_request":          "RDMA duplicate request",
-		"roce_adp_retrans":           "RDMA RoCE adaptive retransmission",
-		"out_of_buffer":              "RDMA out of buffer",
-		"resp_cqe_error":             "RDMA response CQE error",
-		"resp_cqe_flush_error":       "RDMA response CQE flush error",
-		"out_of_sequence":            "RDMA out of sequence",
-		"rx_read_requests":           "RDMA RX read requests",
-		"rx_atomic_requests":         "RDMA RX atomic requests",
-		"req_remote_access_errors":   "RDMA request remote access errors",
-		"rx_write_requests":          "RDMA RX write requests",
-		"resp_remote_access_errors":  "RDMA response remote access errors",
+		"roce_slow_restart_cnps":         "RDMA RoCE slow restart CNPS",
+		"rp_cnp_ignored":                 "RDMA RP CNP ignored",
+		"roce_adp_retrans_to":            "RDMA RoCE adaptive retransmission timeout",
+		"rx_icrc_encapsulated":           "RDMA RX ICRC encapsulated",
+		"resp_local_length_error":        "RDMA response local length error",
+		"np_ecn_marked_roce_packets":     "RDMA NP ECN marked RoCE packets",
+		"roce_slow_restart_trans":        "RDMA RoCE slow restart transactions",
+		"req_remote_invalid_request":     "RDMA request remote invalid request",
+		"local_ack_timeout_err":          "RDMA local ACK timeout error",
+		"lifespan":                       "RDMA lifespan",
+		"req_cqe_error":                  "RDMA request CQE error",
+		"rnr_nak_retry_err":              "RDMA RNR NAK retry error",
+		"np_cnp_sent":                    "RDMA NP CNP sent",
+		"rx_dct_connect":                 "RDMA RX DCT connect",
+		"rp_cnp_handled":                 "RDMA RP CNP handled",
+		"implied_nak_seq_err":            "RDMA implied NAK sequence error",
+		"roce_slow_restart":              "RDMA RoCE slow restart",
+		"req_cqe_flush_error":            "RDMA request CQE flush error",
+		"packet_seq_err":                 "RDMA packet sequence error",
+		"duplicate_request":              "RDMA duplicate request",
+		"roce_adp_retrans":               "RDMA RoCE adaptive retransmission",
+		"out_of_buffer":                  "RDMA out of buffer",
+		"resp_cqe_error":                 "RDMA response CQE error",
+		"resp_cqe_flush_error":           "RDMA response CQE flush error",
+		"out_of_sequence":                "RDMA out of sequence",
+		"rx_read_requests":               "RDMA RX read requests",
+		"rx_atomic_requests":             "RDMA RX atomic requests",
+		"req_remote_access_errors":       "RDMA request remote access errors",
+		"rx_write_requests":              "RDMA RX write requests",
+		"resp_remote_access_errors":      "RDMA response remote access errors",
+		"req_transport_retries_exceeded": "RDMA request transport retries exceeded",
+		"req_rnr_retries_exceeded":       "RDMA request RNR retries exceeded",
 	}
 	rdmaCounters = map[string]string{
 		"unicast_rcv_packets":             "RDMA unicast received packets",