cmd/scollector: Covert gauges to counters for os.cpu

bosun-monitor · Dec 3, 2015 · f4d9667 · f4d9667
1 parent 14af5aa
commit f4d9667
Show file tree

Hide file tree

Showing 4 changed files with 74 additions and 22 deletions.
diff --git a/cmd/scollector/collectors/collectors.go b/cmd/scollector/collectors/collectors.go
@@ -325,3 +325,17 @@ func skipMetric(index string) bool {
 	}
 	return true
 }
+
+type tsIntegrator func(int64, float64) float64
+
+func getTsIntegrator() tsIntegrator {
+	var total float64
+	var lastTimestamp int64
+	return func(timestamp int64, v float64) float64 {
+		if lastTimestamp > 0 {
+			total += v * float64(timestamp-lastTimestamp)
+		}
+		lastTimestamp = timestamp
+		return total
+	}
+}
diff --git a/cmd/scollector/collectors/snmp_cisco.go b/cmd/scollector/collectors/snmp_cisco.go
@@ -2,6 +2,7 @@ package collectors
 
 import (
 	"fmt"
+	"strconv"
 	"time"
 
 	"bosun.org/cmd/scollector/conf"
@@ -13,14 +14,6 @@ import (
 func SNMPCisco(cfg conf.SNMP) {
 	mib := conf.MIB{
 		BaseOid: "1.3.6.1.4.1.9.9",
-		Metrics: []conf.MIBMetric{
-			{
-				Metric:      "cisco.cpu",
-				Oid:         ".109.1.1.1.1.6",
-				FallbackOid: ".109.1.1.1.1.6.1",
-				Unit:        metadata.Pct,
-			},
-		},
 		Trees: []conf.MIBTree{
 			{
 				BaseOid: ".48.1.1.1",
@@ -40,12 +33,50 @@ func SNMPCisco(cfg conf.SNMP) {
 			},
 		},
 	}
-
-	collectors = append(collectors, &IntervalCollector{
-		F: func() (opentsdb.MultiDataPoint, error) {
-			return GenericSnmp(cfg, mib)
+	cpuIntegrator := getTsIntegrator()
+	collectors = append(collectors,
+		&IntervalCollector{
+			F: func() (opentsdb.MultiDataPoint, error) {
+				return GenericSnmp(cfg, mib)
+			},
+			Interval: time.Second * 30,
+			name:     fmt.Sprintf("snmp-cisco-%s", cfg.Host),
+		},
+		&IntervalCollector{
+			F: func() (opentsdb.MultiDataPoint, error) {
+				return c_cisco_cpu(cfg.Host, cfg.Community, cpuIntegrator)
+			},
+			Interval: time.Second * 30,
+			name:     fmt.Sprintf("snmp-cisco-cpu-%s", cfg.Host),
 		},
-		Interval: time.Second * 30,
-		name:     fmt.Sprintf("snmp-cisco-%s", cfg.Host),
-	})
+	)
+}
+
+const (
+	cpmCPUTotal5secRev = ".1.3.6.1.4.1.9.9.109.1.1.1.1.6"
+)
+
+func c_cisco_cpu(host, community string, cpuIntegrator tsIntegrator) (opentsdb.MultiDataPoint, error) {
+	var md opentsdb.MultiDataPoint
+	cpuRaw, err := snmp_subtree(host, community, cpmCPUTotal5secRev)
+	if err != nil {
+		return md, err
+	}
+	tags := opentsdb.TagSet{"host": host}
+	cpu := make(map[string]int)
+	for k, v := range cpuRaw {
+		pct, err := strconv.Atoi(fmt.Sprintf("%v", v))
+		if err != nil {
+			return md, err
+		}
+		cpu[k] = pct
+	}
+	if len(cpu) > 1 {
+		return md, fmt.Errorf("expected only one cpu when monitoring cisco cpu via cpmCPUTotal5secRev")
+	}
+	for _, pct := range cpu {
+		Add(&md, "cisco.cpu", pct, tags, metadata.Gauge, metadata.Pct, "")
+		Add(&md, osCPU, cpuIntegrator(time.Now().Unix(), float64(pct)), tags, metadata.Counter, metadata.Pct, "")
+	}
+	return md, nil
 }
diff --git a/cmd/scollector/collectors/snmp_sys.go b/cmd/scollector/collectors/snmp_sys.go
@@ -16,14 +16,14 @@ const sysUpTime = ".1.3.6.1.2.1.1.3.0" // "The time (in hundredths of a second)
 func SNMPSys(cfg conf.SNMP) {
 	collectors = append(collectors, &IntervalCollector{
 		F: func() (opentsdb.MultiDataPoint, error) {
-			return c_snmp_sys(cfg.Community, cfg.Host)
+			return c_snmp_sys(cfg.Host, cfg.Community)
 		},
 		Interval: time.Minute * 1,
 		name:     fmt.Sprintf("snmp-sys-%s", cfg.Host),
 	})
 }
 
-func c_snmp_sys(community, host string) (opentsdb.MultiDataPoint, error) {
+func c_snmp_sys(host, community string) (opentsdb.MultiDataPoint, error) {
 	var md opentsdb.MultiDataPoint
 	uptime, err := snmp_oid(host, community, sysUpTime)
 	if err != nil {

diff --git a/cmd/scollector/collectors/vsphere.go b/cmd/scollector/collectors/vsphere.go
@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"strconv"
+	"time"
 
 	"bosun.org/metadata"
 	"bosun.org/opentsdb"
@@ -19,22 +20,23 @@ func Vsphere(user, pwd, host string) error {
 	if host == "" || user == "" || pwd == "" {
 		return fmt.Errorf("empty Host, User, or Password in Vsphere")
 	}
+	cpuIntegrators := make(map[string]tsIntegrator)
 	collectors = append(collectors, &IntervalCollector{
 		F: func() (opentsdb.MultiDataPoint, error) {
-			return c_vsphere(user, pwd, host)
+			return c_vsphere(user, pwd, host, cpuIntegrators)
 		},
 		name: fmt.Sprintf("vsphere-%s", host),
 	})
 	return nil
 }
 
-func c_vsphere(user, pwd, host string) (opentsdb.MultiDataPoint, error) {
+func c_vsphere(user, pwd, host string, cpuIntegrators map[string]tsIntegrator) (opentsdb.MultiDataPoint, error) {
 	v, err := vsphere.Connect(host, user, pwd)
 	if err != nil {
 		return nil, err
 	}
 	var md opentsdb.MultiDataPoint
-	if err := vsphereHost(v, &md); err != nil {
+	if err := vsphereHost(v, &md, cpuIntegrators); err != nil {
 		return nil, err
 	}
 	if err := vsphereDatastore(v, &md); err != nil {
@@ -111,7 +113,7 @@ type HostSystemIdentificationInfo struct {
 	} `xml:"identifierType"`
 }
 
-func vsphereHost(v *vsphere.Vsphere, md *opentsdb.MultiDataPoint) error {
+func vsphereHost(v *vsphere.Vsphere, md *opentsdb.MultiDataPoint, cpuIntegrators map[string]tsIntegrator) error {
 	res, err := v.Info("HostSystem", []string{
 		"name",
 		"summary.hardware.cpuMhz",
@@ -205,7 +207,12 @@ func vsphereHost(v *vsphere.Vsphere, md *opentsdb.MultiDataPoint) error {
 		if cpuMhz > 0 && cpuUse > 0 && cpuCores > 0 {
 			cpuTotal := cpuMhz * cpuCores
 			Add(md, "vsphere.cpu", cpuTotal-cpuUse, opentsdb.TagSet{"host": name, "type": "idle"}, metadata.Gauge, metadata.MHz, "")
-			Add(md, "vsphere.cpu.pct", float64(cpuUse)/float64(cpuTotal)*100, tags, metadata.Gauge, metadata.Pct, "")
+			pct := float64(cpuUse) / float64(cpuTotal) * 100
+			Add(md, "vsphere.cpu.pct", pct, tags, metadata.Gauge, metadata.Pct, "")
+			if _, ok := cpuIntegrators[name]; !ok {
+				cpuIntegrators[name] = getTsIntegrator()
+			}
+			Add(md, osCPU, cpuIntegrators[name](time.Now().Unix(), pct), tags, metadata.Counter, metadata.Pct, "")
 		}
 	}
 	return Error