Collect InfiniBand port state and physical state (prometheus#1357)

Collect the InfiniBand port state, the physical state, and the maximum signal transfer rate. Signed-off-by: Benjamin Drung <[email protected]>
shatteredsilicon · Apr 9, 2024 · b99aec3 · b99aec3
1 parent f2756fc
commit b99aec3
Show file tree

Hide file tree

Showing 3 changed files with 37 additions and 0 deletions.
diff --git a/collector/fixtures/e2e-64k-page-output.txt b/collector/fixtures/e2e-64k-page-output.txt
@@ -840,6 +840,11 @@ node_infiniband_multicast_packets_received_total{device="mlx4_0",port="2"} 0
 # TYPE node_infiniband_multicast_packets_transmitted_total counter
 node_infiniband_multicast_packets_transmitted_total{device="mlx4_0",port="1"} 16
 node_infiniband_multicast_packets_transmitted_total{device="mlx4_0",port="2"} 0
+# HELP node_infiniband_physical_state_id Physical state of the InfiniBand port (0: no change, 1: sleep, 2: polling, 3: disable, 4: shift, 5: link up, 6: link error recover, 7: phytest)
+# TYPE node_infiniband_physical_state_id gauge
+node_infiniband_physical_state_id{device="i40iw0",port="1"} 5
+node_infiniband_physical_state_id{device="mlx4_0",port="1"} 5
+node_infiniband_physical_state_id{device="mlx4_0",port="2"} 5
 # HELP node_infiniband_port_constraint_errors_received_total Number of packets received on the switch physical port that are discarded
 # TYPE node_infiniband_port_constraint_errors_received_total counter
 node_infiniband_port_constraint_errors_received_total{device="mlx4_0",port="1"} 0
@@ -872,6 +877,16 @@ node_infiniband_port_packets_transmitted_total{device="mlx4_0",port="1"} 6.23586
 # HELP node_infiniband_port_transmit_wait_total Number of ticks during which the port had data to transmit but no data was sent during the entire tick
 # TYPE node_infiniband_port_transmit_wait_total counter
 node_infiniband_port_transmit_wait_total{device="mlx4_0",port="1"} 4.294967295e+09
+# HELP node_infiniband_rate_bytes_per_second Maximum signal transfer rate
+# TYPE node_infiniband_rate_bytes_per_second gauge
+node_infiniband_rate_bytes_per_second{device="i40iw0",port="1"} 1.25e+09
+node_infiniband_rate_bytes_per_second{device="mlx4_0",port="1"} 5e+09
+node_infiniband_rate_bytes_per_second{device="mlx4_0",port="2"} 5e+09
+# HELP node_infiniband_state_id State of the InfiniBand port (0: no change, 1: down, 2: init, 3: armed, 4: active, 5: act defer)
+# TYPE node_infiniband_state_id gauge
+node_infiniband_state_id{device="i40iw0",port="1"} 4
+node_infiniband_state_id{device="mlx4_0",port="1"} 4
+node_infiniband_state_id{device="mlx4_0",port="2"} 4
 # HELP node_infiniband_unicast_packets_received_total Number of unicast packets received (including errors)
 # TYPE node_infiniband_unicast_packets_received_total counter
 node_infiniband_unicast_packets_received_total{device="mlx4_0",port="1"} 61148

diff --git a/collector/fixtures/e2e-output.txt b/collector/fixtures/e2e-output.txt
@@ -1112,6 +1112,11 @@ node_infiniband_multicast_packets_received_total{device="mlx4_0",port="2"} 0
 # TYPE node_infiniband_multicast_packets_transmitted_total counter
 node_infiniband_multicast_packets_transmitted_total{device="mlx4_0",port="1"} 16
 node_infiniband_multicast_packets_transmitted_total{device="mlx4_0",port="2"} 0
+# HELP node_infiniband_physical_state_id Physical state of the InfiniBand port (0: no change, 1: sleep, 2: polling, 3: disable, 4: shift, 5: link up, 6: link error recover, 7: phytest)
+# TYPE node_infiniband_physical_state_id gauge
+node_infiniband_physical_state_id{device="i40iw0",port="1"} 5
+node_infiniband_physical_state_id{device="mlx4_0",port="1"} 5
+node_infiniband_physical_state_id{device="mlx4_0",port="2"} 5
 # HELP node_infiniband_port_constraint_errors_received_total Number of packets received on the switch physical port that are discarded
 # TYPE node_infiniband_port_constraint_errors_received_total counter
 node_infiniband_port_constraint_errors_received_total{device="mlx4_0",port="1"} 0
@@ -1144,6 +1149,16 @@ node_infiniband_port_packets_transmitted_total{device="mlx4_0",port="1"} 6.23586
 # HELP node_infiniband_port_transmit_wait_total Number of ticks during which the port had data to transmit but no data was sent during the entire tick
 # TYPE node_infiniband_port_transmit_wait_total counter
 node_infiniband_port_transmit_wait_total{device="mlx4_0",port="1"} 4.294967295e+09
+# HELP node_infiniband_rate_bytes_per_second Maximum signal transfer rate
+# TYPE node_infiniband_rate_bytes_per_second gauge
+node_infiniband_rate_bytes_per_second{device="i40iw0",port="1"} 1.25e+09
+node_infiniband_rate_bytes_per_second{device="mlx4_0",port="1"} 5e+09
+node_infiniband_rate_bytes_per_second{device="mlx4_0",port="2"} 5e+09
+# HELP node_infiniband_state_id State of the InfiniBand port (0: no change, 1: down, 2: init, 3: armed, 4: active, 5: act defer)
+# TYPE node_infiniband_state_id gauge
+node_infiniband_state_id{device="i40iw0",port="1"} 4
+node_infiniband_state_id{device="mlx4_0",port="1"} 4
+node_infiniband_state_id{device="mlx4_0",port="2"} 4
 # HELP node_infiniband_unicast_packets_received_total Number of unicast packets received (including errors)
 # TYPE node_infiniband_unicast_packets_received_total counter
 node_infiniband_unicast_packets_received_total{device="mlx4_0",port="1"} 61148

diff --git a/collector/infiniband_linux.go b/collector/infiniband_linux.go
@@ -57,6 +57,7 @@ func NewInfiniBandCollector() (Collector, error) {
 		"link_error_recovery_total":                  "Number of times the link successfully recovered from an error state",
 		"multicast_packets_received_total":           "Number of multicast packets received (including errors)",
 		"multicast_packets_transmitted_total":        "Number of multicast packets transmitted (including errors)",
+		"physical_state_id":                          "Physical state of the InfiniBand port (0: no change, 1: sleep, 2: polling, 3: disable, 4: shift, 5: link up, 6: link error recover, 7: phytest)",
 		"port_constraint_errors_received_total":      "Number of packets received on the switch physical port that are discarded",
 		"port_constraint_errors_transmitted_total":   "Number of packets not transmitted from the switch physical port",
 		"port_data_received_bytes_total":             "Number of data octets received on all links",
@@ -67,6 +68,8 @@ func NewInfiniBandCollector() (Collector, error) {
 		"port_packets_received_total":                "Number of packets received on all VLs by this port (including errors)",
 		"port_packets_transmitted_total":             "Number of packets transmitted on all VLs from this port (including errors)",
 		"port_transmit_wait_total":                   "Number of ticks during which the port had data to transmit but no data was sent during the entire tick",
+		"rate_bytes_per_second":                      "Maximum signal transfer rate",
+		"state_id":                                   "State of the InfiniBand port (0: no change, 1: down, 2: init, 3: armed, 4: active, 5: act defer)",
 		"unicast_packets_received_total":             "Number of unicast packets received (including errors)",
 		"unicast_packets_transmitted_total":          "Number of unicast packets transmitted (including errors)",
 	}
@@ -105,6 +108,10 @@ func (c *infinibandCollector) Update(ch chan<- prometheus.Metric) error {
 		for _, port := range device.Ports {
 			portStr := strconv.FormatUint(uint64(port.Port), 10)
 
+			c.pushMetric(ch, "state_id", uint64(port.StateID), port.Name, portStr, prometheus.GaugeValue)
+			c.pushMetric(ch, "physical_state_id", uint64(port.PhysStateID), port.Name, portStr, prometheus.GaugeValue)
+			c.pushMetric(ch, "rate_bytes_per_second", port.Rate, port.Name, portStr, prometheus.GaugeValue)
+
 			c.pushCounter(ch, "legacy_multicast_packets_received_total", port.Counters.LegacyPortMulticastRcvPackets, port.Name, portStr)
 			c.pushCounter(ch, "legacy_multicast_packets_transmitted_total", port.Counters.LegacyPortMulticastXmitPackets, port.Name, portStr)
 			c.pushCounter(ch, "legacy_data_received_bytes_total", port.Counters.LegacyPortRcvData64, port.Name, portStr)