Add "nomad.client.host.cpu.total_ticks" to metrics

hashicorp · Jun 17, 2023 · a7a62d7 · a7a62d7
1 parent 80e1ad6
commit a7a62d7
Show file tree

Hide file tree

Showing 5 changed files with 24 additions and 15 deletions.
diff --git a/client/client.go b/client/client.go
@@ -3039,7 +3039,10 @@ func (c *Client) setGaugeForCPUStats(nodeID string, hStats *stats.HostStats, bas
 			Value: cpu.CPU,
 		})
 
-		metrics.SetGaugeWithLabels([]string{"client", "host", "cpu", "total"}, float32(cpu.Total), labels)
+		// Keep "total" around to remain compatible with older consumers of the metrics
+		metrics.SetGaugeWithLabels([]string{"client", "host", "cpu", "total"}, float32(cpu.TotalPercent), labels)
+		metrics.SetGaugeWithLabels([]string{"client", "host", "cpu", "total_percent"}, float32(cpu.TotalPercent), labels)
+		metrics.SetGaugeWithLabels([]string{"client", "host", "cpu", "total_ticks"}, float32(cpu.TotalTicks), labels)
 		metrics.SetGaugeWithLabels([]string{"client", "host", "cpu", "user"}, float32(cpu.User), labels)
 		metrics.SetGaugeWithLabels([]string{"client", "host", "cpu", "idle"}, float32(cpu.Idle), labels)
 		metrics.SetGaugeWithLabels([]string{"client", "host", "cpu", "system"}, float32(cpu.System), labels)

diff --git a/client/stats/cpu.go b/client/stats/cpu.go
@@ -79,14 +79,16 @@ func (h *HostStatsCollector) collectCPUStats() (cpus []*CPUStats, totalTicks flo
 			h.statsCalculator[cpuStat.CPU] = percentCalculator
 		}
 		idle, user, system, total := percentCalculator.Calculate(cpuStat)
+		ticks := (total / 100.0) * (float64(shelpers.TotalTicksAvailable()) / float64(len(cpuStats)))
 		cs[idx] = &CPUStats{
-			CPU:    cpuStat.CPU,
-			User:   user,
-			System: system,
-			Idle:   idle,
-			Total:  total,
+			CPU:          cpuStat.CPU,
+			User:         user,
+			System:       system,
+			Idle:         idle,
+			TotalPercent: total,
+			TotalTicks:   ticks,
 		}
-		ticksConsumed += (total / 100.0) * (float64(shelpers.TotalTicksAvailable()) / float64(len(cpuStats)))
+		ticksConsumed += ticks
 	}
 
 	return cs, ticksConsumed, nil

diff --git a/client/stats/cpu_test.go b/client/stats/cpu_test.go
@@ -55,12 +55,14 @@ func TestHostStats_CPU(t *testing.T) {
 
 	for _, cpu := range stats.CPU {
 		assert.False(math.IsNaN(cpu.Idle))
-		assert.False(math.IsNaN(cpu.Total))
+		assert.False(math.IsNaN(cpu.TotalPercent))
+		assert.False(math.IsNaN(cpu.TotalTicks))
 		assert.False(math.IsNaN(cpu.System))
 		assert.False(math.IsNaN(cpu.User))
 
 		assert.False(math.IsInf(cpu.Idle, 0))
-		assert.False(math.IsInf(cpu.Total, 0))
+		assert.False(math.IsInf(cpu.TotalPercent, 0))
+		assert.False(math.IsInf(cpu.TotalTicks, 0))
 		assert.False(math.IsInf(cpu.System, 0))
 		assert.False(math.IsInf(cpu.User, 0))
 	}

diff --git a/client/stats/host.go b/client/stats/host.go
@@ -39,11 +39,12 @@ type MemoryStats struct {
 
 // CPUStats represents stats related to cpu usage
 type CPUStats struct {
-	CPU    string
-	User   float64
-	System float64
-	Idle   float64
-	Total  float64
+	CPU          string
+	User         float64
+	System       float64
+	Idle         float64
+	TotalPercent float64
+	TotalTicks   float64
 }
 
 // DiskStats represents stats related to disk usage

diff --git a/website/content/docs/operations/metrics-reference.mdx b/website/content/docs/operations/metrics-reference.mdx
@@ -164,7 +164,8 @@ Nomad will emit [tagged metrics][tagged-metrics], in the below format:
 | `nomad.client.allocs.oom_killed`        | Number of allocations OOM killed                                                     | Integer    | Gauge | datacenter, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status       |
 | `nomad.client.host.cpu.idle`            | CPU utilization in idle state                                                        | Percentage | Gauge | cpu, datacenter, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status  |
 | `nomad.client.host.cpu.system`          | CPU utilization in system space                                                      | Percentage | Gauge | cpu, datacenter, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status  |
-| `nomad.client.host.cpu.total`           | Total CPU utilization                                                                | Percentage | Gauge | cpu, datacenter, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status  |
+| `nomad.client.host.cpu.total_percent`   | Total CPU utilization in percentage                                                  | Percentage | Gauge | cpu, datacenter, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status  |
+| `nomad.client.host.cpu.total_ticks`     | Total CPU utilization in ticks                                                       | Integer    | Gauge | cpu, datacenter, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status  |
 | `nomad.client.host.cpu.user`            | CPU utilization in user space                                                        | Percentage | Gauge | cpu, datacenter, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status  |
 | `nomad.client.host.disk.available`      | Amount of space which is available                                                   | Bytes      | Gauge | datacenter, disk, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status |
 | `nomad.client.host.disk.inodes_percent` | Disk space consumed by the inodes                                                    | Percentage | Gauge | datacenter, disk, host, node_class, node_id, node_pool, node_scheduling_eligibility, node_status |