wip: Fix it?

Signed-off-by: Dave Tucker <dave@dtucker.co.uk>
sustainable-computing-io · May 16, 2024 · 7e4408b · 7e4408b
1 parent 53589c5
commit 7e4408b
Show file tree

Hide file tree

Showing 5 changed files with 36 additions and 30 deletions.
diff --git a/pkg/bpf/exporter.go b/pkg/bpf/exporter.go
@@ -181,21 +181,22 @@ func (e *exporter) attach() error {
 	}
 
 	if config.ExposeIRQCounterMetrics {
-		// attach softirq_entry tracepoint to kepler_irq_trace function
-		irq_prog, err := e.module.GetProgram("kepler_irq_trace")
-		if err != nil {
-			klog.Warningf("could not get kepler_irq_trace: %v", err)
-			// disable IRQ metric
-			config.ExposeIRQCounterMetrics = false
-		} else {
-			if _, err := irq_prog.AttachGeneric(); err != nil {
-				klog.Warningf("could not attach irq/softirq_entry: %v", err)
-				// disable IRQ metric
-				config.ExposeIRQCounterMetrics = false
+		err := func() error {
+			// attach softirq_entry tracepoint to kepler_irq_trace function
+			irq_prog, err := e.module.GetProgram("kepler_irq_trace")
+			if err != nil {
+				return fmt.Errorf("could not get kepler_irq_trace: %v", err)
 			}
-			for _, event := range SoftIRQEvents {
-				e.enabledSoftwareCounters[event] = struct{}{}
+			if _, err := irq_prog.AttachGeneric(); err != nil {
+				return fmt.Errorf("could not attach irq/softirq_entry: %v", err)
 			}
+			e.enabledSoftwareCounters[config.IRQNetTXLabel] = struct{}{}
+			e.enabledSoftwareCounters[config.IRQNetRXLabel] = struct{}{}
+			e.enabledSoftwareCounters[config.IRQBlockLabel] = struct{}{}
+			return nil
+		}()
+		if err != nil {
+			klog.Warningf("IRQ tracing disabled: %v", err)
 		}
 	}
 

diff --git a/pkg/bpf/test_utils.go b/pkg/bpf/test_utils.go
@@ -21,9 +21,9 @@ func defaultHardwareCounters() map[string]struct{} {
 func defaultSoftwareCounters() map[string]struct{} {
 	swCounters := map[string]struct{}{config.CPUTime: {}, config.PageCacheHit: {}}
 	if config.ExposeIRQCounterMetrics {
-		for _, event := range SoftIRQEvents {
-			swCounters[event] = struct{}{}
-		}
+		swCounters[config.IRQNetTXLabel] = struct{}{}
+		swCounters[config.IRQNetRXLabel] = struct{}{}
+		swCounters[config.IRQBlockLabel] = struct{}{}
 	}
 	return swCounters
 }

diff --git a/pkg/bpf/types.go b/pkg/bpf/types.go
@@ -20,8 +20,6 @@ import (
 	"github.com/sustainable-computing-io/kepler/pkg/config"
 )
 
-var SoftIRQEvents = []string{config.IRQNetTXLabel, config.IRQNetRXLabel, config.IRQBlockLabel}
-
 type Exporter interface {
 	SupportedMetrics() SupportedMetrics
 	Detach()

diff --git a/pkg/collector/resourceutilization/bpf/process_bpf_collector.go b/pkg/collector/resourceutilization/bpf/process_bpf_collector.go
@@ -33,15 +33,26 @@ import (
 type ProcessBPFMetrics = bpf.ProcessBPFMetrics
 
 // update software counter metrics
-func updateSWCounters(key uint64, ct *ProcessBPFMetrics, processStats map[uint64]*stats.ProcessStats) {
+func updateSWCounters(key uint64, ct *ProcessBPFMetrics, processStats map[uint64]*stats.ProcessStats, bpfSupportedMetrics bpf.SupportedMetrics) {
 	// update ebpf metrics
 	// first update CPU time and Page Cache Hit
-	processStats[key].ResourceUsage[config.CPUTime].AddDeltaStat(utils.GenericSocketID, ct.ProcessRunTime)
-	processStats[key].ResourceUsage[config.TaskClock].AddDeltaStat(utils.GenericSocketID, ct.TaskClockTime)
-	processStats[key].ResourceUsage[config.PageCacheHit].AddDeltaStat(utils.GenericSocketID, ct.PageCacheHit/(1000*1000))
-	// update IRQ vector. Soft IRQ events has the events ordered
-	for i, event := range bpf.SoftIRQEvents {
-		processStats[key].ResourceUsage[event].AddDeltaStat(utils.GenericSocketID, uint64(ct.VecNR[i]))
+	for counterKey := range bpfSupportedMetrics.SoftwareCounters {
+		switch counterKey {
+		case config.CPUTime:
+			processStats[key].ResourceUsage[config.CPUTime].AddDeltaStat(utils.GenericSocketID, ct.ProcessRunTime)
+		case config.TaskClock:
+			processStats[key].ResourceUsage[config.TaskClock].AddDeltaStat(utils.GenericSocketID, ct.TaskClockTime)
+		case config.PageCacheHit:
+			processStats[key].ResourceUsage[config.PageCacheHit].AddDeltaStat(utils.GenericSocketID, ct.PageCacheHit/(1000*1000))
+		case config.IRQNetTXLabel:
+			processStats[key].ResourceUsage[config.IRQNetTXLabel].AddDeltaStat(utils.GenericSocketID, uint64(ct.VecNR[bpf.IRQNetTX]))
+		case config.IRQNetRXLabel:
+			processStats[key].ResourceUsage[config.IRQNetRXLabel].AddDeltaStat(utils.GenericSocketID, uint64(ct.VecNR[bpf.IRQNetRX]))
+		case config.IRQBlockLabel:
+			processStats[key].ResourceUsage[config.IRQBlockLabel].AddDeltaStat(utils.GenericSocketID, uint64(ct.VecNR[bpf.IRQBlock]))
+		default:
+			klog.Errorf("counter %s is not supported\n", counterKey)
+		}
 	}
 }
 
@@ -126,7 +137,7 @@ func UpdateProcessBPFMetrics(bpfExporter bpf.Exporter, processStats map[uint64]*
 		// when the process metrics are updated, reset the idle counter
 		pStat.IdleCounter = 0
 
-		updateSWCounters(mapKey, &ct, processStats)
+		updateSWCounters(mapKey, &ct, processStats, bpfSupportedMetrics)
 		updateHWCounters(mapKey, &ct, processStats, bpfSupportedMetrics)
 	}
 }
diff --git a/pkg/config/config.go b/pkg/config/config.go
@@ -457,10 +457,6 @@ func IsCgroupMetricsEnabled() bool {
 	return ExposeCgroupMetrics
 }
 
-func IsIRQCounterMetricsEnabled() bool {
-	return ExposeIRQCounterMetrics
-}
-
 func SetGpuUsageMetric(metric string) {
 	GpuUsageMetric = metric
 }