Collect non-numeric data from /sys/class/infiniband (prometheus#1563)

Let the node exporter collect the non-numeric data from /sys/class/infiniband: board ID, firmware version, and HCA type. Signed-off-by: Benjamin Drung <[email protected]> Co-authored-by: Ben Kochie <[email protected]>
shatteredsilicon · Apr 9, 2024 · 4dc0171 · 4dc0171
1 parent eec00ca
commit 4dc0171
Show file tree

Hide file tree

Showing 3 changed files with 20 additions and 1 deletion.
diff --git a/collector/fixtures/e2e-64k-page-output.txt b/collector/fixtures/e2e-64k-page-output.txt
@@ -792,6 +792,10 @@ node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp2"} 84
 node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp3"} 84
 node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp4"} 84
 node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp5"} 84
+# HELP node_infiniband_info Non-numeric data from /sys/class/infiniband/<device>, value is always 1.
+# TYPE node_infiniband_info gauge
+node_infiniband_info{board_id="I40IW Board ID",device="i40iw0",firmware_version="0.2",hca_type="I40IW"} 1
+node_infiniband_info{board_id="SM_1141000001000",device="mlx4_0",firmware_version="2.31.5050",hca_type="MT4099"} 1
 # HELP node_infiniband_legacy_data_received_bytes_total Number of data octets received on all links
 # TYPE node_infiniband_legacy_data_received_bytes_total counter
 node_infiniband_legacy_data_received_bytes_total{device="mlx4_0",port="1"} 1.8527668e+07

diff --git a/collector/fixtures/e2e-output.txt b/collector/fixtures/e2e-output.txt
@@ -1085,6 +1085,10 @@ node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp2"} 84
 node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp3"} 84
 node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp4"} 84
 node_hwmon_temp_max_celsius{chip="platform_coretemp_1",sensor="temp5"} 84
+# HELP node_infiniband_info Non-numeric data from /sys/class/infiniband/<device>, value is always 1.
+# TYPE node_infiniband_info gauge
+node_infiniband_info{board_id="I40IW Board ID",device="i40iw0",firmware_version="0.2",hca_type="I40IW"} 1
+node_infiniband_info{board_id="SM_1141000001000",device="mlx4_0",firmware_version="2.31.5050",hca_type="MT4099"} 1
 # HELP node_infiniband_legacy_data_received_bytes_total Number of data octets received on all links
 # TYPE node_infiniband_legacy_data_received_bytes_total counter
 node_infiniband_legacy_data_received_bytes_total{device="mlx4_0",port="1"} 1.8527668e+07

diff --git a/collector/infiniband_linux.go b/collector/infiniband_linux.go
@@ -31,6 +31,7 @@ type infinibandCollector struct {
 	fs          sysfs.FS
 	metricDescs map[string]*prometheus.Desc
 	logger      log.Logger
+	subsystem   string
 }
 
 func init() {
@@ -80,10 +81,11 @@ func NewInfiniBandCollector(logger log.Logger) (Collector, error) {
 	}
 
 	i.metricDescs = make(map[string]*prometheus.Desc)
+	i.subsystem = "infiniband"
 
 	for metricName, description := range descriptions {
 		i.metricDescs[metricName] = prometheus.NewDesc(
-			prometheus.BuildFQName(namespace, "infiniband", metricName),
+			prometheus.BuildFQName(namespace, i.subsystem, metricName),
 			description,
 			[]string{"device", "port"},
 			nil,
@@ -114,6 +116,15 @@ func (c *infinibandCollector) Update(ch chan<- prometheus.Metric) error {
 	}
 
 	for _, device := range devices {
+		infoDesc := prometheus.NewDesc(
+			prometheus.BuildFQName(namespace, c.subsystem, "info"),
+			"Non-numeric data from /sys/class/infiniband/<device>, value is always 1.",
+			[]string{"device", "board_id", "firmware_version", "hca_type"},
+			nil,
+		)
+		infoValue := 1.0
+		ch <- prometheus.MustNewConstMetric(infoDesc, prometheus.GaugeValue, infoValue, device.Name, device.BoardID, device.FirmwareVersion, device.HCAType)
+
 		for _, port := range device.Ports {
 			portStr := strconv.FormatUint(uint64(port.Port), 10)