cmd/ocprometheus: Allow decoding enumerated values

cmd/ocprometheus/README.md

@@ -2,8 +2,12 @@
 
 This is a client for the OpenConfig gRPC interface that pushes telemetry to
 Prometheus. Numerical and boolean (converted to 1 for true and 0 for false) are
-supported. Non-numerical data isn't supported by Prometheus and is silently
-dropped. Arrays (even with numeric values) are not yet supported.
+supported. String values can be inserted as a label using the `valuelabel` option
+with the numeric value of the metric specified using the `defaultvalue` option.
+For enumerated string values, the option `decodeenum` can be used to transform
+each string value into a numeric value. Other non-numerical data isn't supported
+by Prometheus and is silently dropped. Arrays (even with numeric values) are not
+yet supported.
 
 This tool requires a config file to specify how to map the path of the
 notificatons coming out of the OpenConfig gRPC interface onto Prometheus

cmd/ocprometheus/collector.go

@@ -33,6 +33,7 @@ type labelledMetric struct {
 	labels       []string
 	defaultValue float64
 	stringMetric bool
+	decodeEnum   map[string]float64
 }
 
 type collector struct {
@@ -113,6 +114,18 @@ func (c *collector) update(addr string, message proto.Message) {
 		c.m.Lock()
 		// Use the cached labels and descriptor if available
 		if m, ok := c.metrics[src]; ok {
+			if m.decodeEnum != nil {
+				// decode the string value according to the enum lookup
+				decodedVal, enumExists := m.decodeEnum[strVal]
+				if enumExists {
+					strUpdate = false
+					floatVal = decodedVal
+				} else {
+					glog.V(8).Infof("Failed to decode enum value %+v of update %v at %s:%s",
+						value, update, device, path)
+				}
+			}
+
 			if strUpdate {
 				// Skip string updates for non string metrics
 				if !m.stringMetric {
@@ -139,7 +152,20 @@ func (c *collector) update(addr string, message proto.Message) {
 			continue
 		}
 
-		if metric.stringMetric {
+		enumDecoded := false
+		if metric.decodeEnum != nil {
+			// Decode the string value according to the enum lookup
+			decodedVal, enumExists := metric.decodeEnum[strVal]
+			if enumExists {
+				floatVal = decodedVal
+				enumDecoded = true
+			} else {
+				glog.V(8).Infof("Failed to decode enum value %+v of update %v at %s:%s",
+					value, update, device, path)
+			}
+		}
+
+		if metric.stringMetric && !enumDecoded {
 			if !strUpdate {
 				// A float was parsed from the update, yet metric expects a string.
 				// Store the float as a string.
@@ -159,6 +185,7 @@ func (c *collector) update(addr string, message proto.Message) {
 			labels:       metric.labels,
 			defaultValue: metric.defaultValue,
 			stringMetric: metric.stringMetric,
+			decodeEnum:   metric.decodeEnum,
 		}
 		c.m.Unlock()
 	}

cmd/ocprometheus/config.go

@@ -48,6 +48,9 @@ type MetricDef struct {
 	// Label to store string values
 	ValueLabel string
 
+	// Map decoding enumeration values to numeric values
+	DecodeEnum map[string]float64
+
 	// Default value to display for string values
 	DefaultValue float64
 
@@ -67,6 +70,7 @@ type metricValues struct {
 	labels       []string
 	defaultValue float64
 	stringMetric bool
+	decodeEnum   map[string]float64
 }
 
 // Parses the config and creates the descriptors for each path and device.
@@ -128,7 +132,7 @@ func (c *Config) getMetricValues(s source) *metricValues {
 				desc = def.desc
 			}
 			return &metricValues{desc: desc, labels: groups[1:], defaultValue: def.DefaultValue,
-				stringMetric: def.stringMetric}
+				stringMetric: def.stringMetric, decodeEnum: def.DecodeEnum}
 		}
 	}
 

cmd/ocprometheus/sample_configs/enum_decoding-7280.yml

@@ -0,0 +1,139 @@
+# Per-device labels. Optional
+# Exactly the same set of labels must be specified for each device.
+# If device address is *, the labels apply to all devices not listed explicitly.
+# If any explicit device if listed below, then you need to specify all devices you're subscribed to,
+# or have a wildcard entry. Otherwise, updates from non-listed devices will be ignored.
+#deviceLabels:
+#        10.1.1.1:
+#                lab1: val1
+#                lab2: val2
+#        '*':
+#                lab1: val3
+#                lab2: val4
+
+# Subscriptions to OpenConfig paths.
+subscriptions:
+        - eos_native:/Smash/counters/ethIntf
+        - eos_native:/Smash/interface/counter/lag/current/counter
+        - eos_native:/Sysdb/environment/archer/cooling/status
+        - eos_native:/Sysdb/environment/archer/power/status
+        - eos_native:/Sysdb/environment/archer/temperature/status
+        - eos_native:/Sysdb/hardware/archer/xcvr/status
+        - eos_native:/Sysdb/interface/config/eth
+        - eos_native:/Kernel/proc
+        - eos_native:/Sysdb/connectivityMonitor
+        - /interfaces/interface
+
+# Prometheus metrics configuration.
+# If you use named capture groups in the path, they will be extracted into labels with the same name.
+# All fields are mandatory.
+metrics:
+        - name: interfaceDescription
+          path: /Sysdb/interface/config/eth/phy/slice/1/intfConfig/(?P<interface>Ethernet.+)/description
+          help: Description
+          valuelabel: description
+          defaultvalue: 15
+        - name: intfCounter
+          path: /Smash/counters/ethIntf/SandCounters/current/(counter)/(?P<intf>.+)/statistics/(?P<direction>(?:in|out))(?P<type>(?:Octets|Errors|Discards))
+          help: Per-Interface Bytes/Errors/Discards Counters
+        - name: intfLagCounter
+          path: /Smash/interface/counter/lag/current/(counter)/(?P<intf>.+)/statistics/(?P<direction>(?:in|out))(Octets|Errors|Discards)
+          help: Per-Lag Bytes/Errors/Discards Counters
+        - name: intfPktCounter
+          path: /Smash/counters/ethIntf/SandCounters/current/counter/(?P<intf>.+)/statistics/(?P<direction>(?:in|out))(?P<type>(?:Ucast|Multicast|Broadcast))Pkt
+          help: Per-Interface Unicast/Multicast/Broadcast Packer Counters
+        - name: intfLagPktCounter
+          path: /Smash/interface/counter/lag/current/(counter)/(?P<intf>.+)/statistics/(?P<direction>(?:in|out))(?P<type>(?:Ucast|Multicast|Broadcast))(Pkt)
+          help: Per-Lag Unicast/Multicast/Broadcast Packer Counters
+        - name: intfPfcClassCounter
+          path: /Smash/counters/ethIntf/SandCounters/current/(counter)/(?P<intf>.+)/ethStatistics/(?P<direction>(?:in|out))(PfcClassFrames)
+          help: Per-Interface Input/Output PFC Frames Counters
+        - name: tempSensor
+          path: /Sysdb/environment/archer/temperature/status/(?P<sensor>.+)/((?:maxT|t)emperature)/value
+          help: Temperature and Maximum Temperature
+        - name: tempSensorAlert
+          path: /Sysdb/environment/archer/temperature/status/(?P<sensor>.+)/(alertRaisedCount)
+          help: Temperature Alerts Counter
+        - name: currentSensor
+          path: /Sysdb/environment/archer/power/status/currentSensor/(?P<sensor>.+)/(current)
+          help: Current Levels
+        - name: powerSensor
+          path: /Sysdb/environment/archer/power/status/powerSupply/(?P<sensor>.+)/(?P<pwrdir>(?:input|output))Power
+          help: Input/Output Power Levels
+        - name: voltageSensor
+          path: /Sysdb/environment/archer/power/status/voltageSensor/(?:cell/.+|system)/(?P<sensor>VoltageSensor.+)/(voltage)/value
+          help: Voltage Levels
+        - name: railCurrentSensor
+          path: /Sysdb/environment/archer/power/status/voltageSensor/(?:cell/.+|system)/(?P<sensor>VoltageSensor.+)/(current)/value
+          help: Rail Current Levels
+        - name: fanSpeed
+          path: /Sysdb/environment/archer/(cooling)/status/(?P<fan>.+)/speed/value
+          help: Fan Speed
+        - name: qsfpModularRxPower
+          path: /Sysdb/hardware/archer/(xcvr)/status/slice/(?P<linecard>.+)/(?P<intf>.+)/domRegisterData/lane(?P<lane>\d)(OpticalRxPower)
+          help: qsfpModularRxPower
+        - name: qsfpFixedRxPower
+          path: /Sysdb/hardware/archer/(xcvr)/status/all/(?P<intf>.+)/domRegisterData/lane(?P<lane>\d)(OpticalRxPower)
+          help: qsfpFixedRxPower
+        - name: qsfpModularTxPower
+          path: /Sysdb/hardware/archer/(xcvr)/status/slice/(?P<linecard>.+)/(?P<intf>.+)/domRegisterData/lane(?P<lane>\d)(OpticalTxPower)
+          help: qsfpModularTxPower
+        - name: qsfpFixedTxPower
+          path: /Sysdb/hardware/archer/(xcvr)/status/all/(?P<intf>.+)/domRegisterData/lane(?P<lane>\d)(OpticalTxPower)
+          help: qsfpFixedTxPower
+        - name: sfpModularTemperature
+          path: /Sysdb/hardware/archer/(xcvr)/status/slice/(?P<linecard>.+)/(?P<intf>.+)/lastDomUpdateTime/(temperature)
+          help: sfpModularTemperature
+        - name: sfpFixedTemperature
+          path: /Sysdb/hardware/archer/(xcvr)/status/all/(?P<intf>.+)/lastDomUpdateTime/(temperature)
+          help: sfpFixedTemperature
+        - name: sfpFixedRxTxPower
+          path: /Sysdb/hardware/archer/xcvr/status/all/(?P<intf>.+)/lastDomUpdateTime/(?P<powerdirection>(?:rx|tx))(Power)
+          help: sfpFixedRxTxPower
+        - name: kernelprocrss
+          path: /Kernel/proc/stat/(?P<pid>.+)/rss
+          help: KernelProcRSS
+        - name: kernelproc
+          path: /Kernel/proc/stat/(?P<pid>.+)/comm
+          help: KernelProcStat
+          valuelabel: command
+          defaultvalue: 1
+        - name: memoryinfo
+          path: /Kernel/proc/meminfo/(?P<usageType>(?:memAvailable|memFree|memTotal|buffers|cached))
+          help: memory info
+        - name: cputotal
+          path: /Kernel/proc/cpu/utilization/total/(?P<usageType>(?:idle|nice|system|user|util))
+          help: CPUtotal
+        - name: connectivityMonitor
+          path: /Sysdb/connectivityMonitor/status/hostStatus/(?P<originhost>.+)/defaultStats/(?P<traceType>(?:jitter|latency|packetLoss|httpResponseTime))
+        - name: inputBinCtr
+          path: /Smash/counters/ethIntf/SandCounters/current/(counter)/(?P<intf>.+)/ethStatistics/(?P<type>(?:in1024To1522OctetFrames|in128To255OctetFrames|in256To511OctetFrames|in512To1023OctetFrames|in64OctetFrames|in65To127OctetFrames))
+        - name: interfaceLinkStatus
+          path: /Sysdb/interface/status/eth/phy/slice/1/intfStatus/(?P<intf>.+)/linkStatus$
+          help: LinkStatus
+          valuelabel: linkStatus
+          defaultvalue: 4
+        - name: interfaceOperStatus
+          path: /Sysdb/interface/status/eth/phy/slice/1/intfStatus/(?P<intr>.+)/operStatus$
+          help: OperStatus
+          valuelabel: operStatus
+          defaultvalue: 2
+        - name: interfaceSpeed
+          path: /interfaces/interface\[name=(?P<intf>[^\]]+)\]/ethernet/state/port-speed
+          decodeenum:
+            #                                         G  M  K
+            "openconfig-if-ethernet:SPEED_10MB":        10000000
+            "openconfig-if-ethernet:SPEED_100MB":      100000000
+            "openconfig-if-ethernet:SPEED_1GB":       1000000000
+            "openconfig-if-ethernet:SPEED_2500MB":    2500000000
+            "openconfig-if-ethernet:SPEED_5GB":       5000000000
+            "openconfig-if-ethernet:SPEED_10GB":     10000000000
+            "openconfig-if-ethernet:SPEED_25GB":     25000000000
+            "openconfig-if-ethernet:SPEED_40GB":     40000000000
+            "openconfig-if-ethernet:SPEED_50GB":     50000000000
+            "openconfig-if-ethernet:SPEED_100GB":   100000000000
+            "openconfig-if-ethernet:SPEED_200GB":   200000000000
+            "openconfig-if-ethernet:SPEED_400GB":   400000000000
+            "openconfig-if-ethernet:SPEED_600GB":   600000000000
+            "openconfig-if-ethernet:SPEED_800GB":   800000000000
+            "openconfig-if-ethernet:SPEED_UNKNOWN":            0