Add metric for instrumented processes

docs/sources/metrics.md

@@ -109,3 +109,4 @@ Beyla can be [configured to report internal metrics]({{< relref "./configure/opt
 | `otel_trace_exports` | Counter | Length of the trace batches submitted to the remote OTEL collector |
 | `otel_trace_export_errors` | CounterVec | Error count on each failed OTEL trace export, by error type |
 | `prometheus_http_requests` | CounterVec | Number of requests towards the Prometheus Scrape endpoint, faceted by HTTP port and path |
+| `beyla_instrumented_processes` | GaugeVec | Instrumented processes by Beyla, with process name |

pkg/internal/discover/attacher.go

@@ -89,13 +89,15 @@ func (ta *TraceAttacher) getTracer(ie *Instrumentable) (*ebpf.ProcessTracer, boo
  ie.FileInfo.Service.SDKLanguage = ie.Type
  // allowing the tracer to forward traces from the new PID and its children processes
  monitorPIDs(tracer, ie)
+ ta.Metrics.InstrumentProcess(ie.FileInfo.ExecutableName())
  if tracer.Type == ebpf.Generic {
  monitorPIDs(ta.reusableTracer, ie)
  }
  ta.log.Debug(".done")
  return nil, false
  }
  ta.log.Info("instrumenting process", "cmd", ie.FileInfo.CmdExePath, "pid", ie.FileInfo.Pid)
+ ta.Metrics.InstrumentProcess(ie.FileInfo.ExecutableName())
 
  // builds a tracer for that executable
  var programs []ebpf.Tracer
@@ -205,6 +207,7 @@ func (ta *TraceAttacher) notifyProcessDeletion(ie *Instrumentable) {
  // notifying the tracer to block any trace from that PID
  // to avoid that a new process reusing this PID could send traces
  // unless explicitly allowed
+ ta.Metrics.UninstrumentProcess(ie.FileInfo.ExecutableName())
  tracer.BlockPID(uint32(ie.FileInfo.Pid), ie.FileInfo.Ns)
 
  // if there are no more trace instances for a Go program, we need to notify that

pkg/internal/imetrics/imetrics.go

@@ -28,6 +28,10 @@ type Reporter interface {
  OTELTraceExportError(err error)
  // PrometheusRequest is invoked every time the Prometheus exporter is invoked, for a given port and path
  PrometheusRequest(port, path string)
+ // InstrumentProcess is invoked every time a new process is instrumented
+ InstrumentProcess(processName string)
+ // UninstrumentProcess is invoked every time a process is removed from the instrumented processed
+ UninstrumentProcess(processName string)
 }
 
 // NoopReporter is a metrics Reporter that just does nothing
@@ -40,3 +44,5 @@ func (n NoopReporter) OTELMetricExportError(_ error) {}
 func (n NoopReporter) OTELTraceExport(_ int) {}
 func (n NoopReporter) OTELTraceExportError(_ error) {}
 func (n NoopReporter) PrometheusRequest(_, _ string) {}
+func (n NoopReporter) InstrumentProcess(_ string) {}
+func (n NoopReporter) UninstrumentProcess(_ string) {}

pkg/internal/imetrics/iprom.go

@@ -21,13 +21,14 @@ type PrometheusConfig struct {
 
 // PrometheusReporter is an internal metrics Reporter that exports to Prometheus
 type PrometheusReporter struct {
- connector *connector.PrometheusManager
- tracerFlushes prometheus.Histogram
- otelMetricExports prometheus.Counter
- otelMetricExportErrs *prometheus.CounterVec
- otelTraceExports prometheus.Counter
- otelTraceExportErrs *prometheus.CounterVec
- prometheusRequests *prometheus.CounterVec
+ connector *connector.PrometheusManager
+ tracerFlushes prometheus.Histogram
+ otelMetricExports prometheus.Counter
+ otelMetricExportErrs *prometheus.CounterVec
+ otelTraceExports prometheus.Counter
+ otelTraceExportErrs *prometheus.CounterVec
+ prometheusRequests *prometheus.CounterVec
+ instrumentedProcesses *prometheus.GaugeVec
 }
 
 func NewPrometheusReporter(cfg *PrometheusConfig, manager *connector.PrometheusManager) *PrometheusReporter {
@@ -61,14 +62,19 @@ func NewPrometheusReporter(cfg *PrometheusConfig, manager *connector.PrometheusM
  Name: "prometheus_http_requests",
  Help: "requests towards the Prometheus Scrape endpoint",
  }, []string{"port", "path"}),
+ instrumentedProcesses: prometheus.NewGaugeVec(prometheus.GaugeOpts{
+ Name: "beyla_instrumented_processes",
+ Help: "instrumented processes by Beyla",
+ }, []string{"process_name"}),
  }
  manager.Register(cfg.Port, cfg.Path,
  pr.tracerFlushes,
  pr.otelMetricExports,
  pr.otelMetricExportErrs,
  pr.otelTraceExports,
  pr.otelTraceExportErrs,
- pr.prometheusRequests)
+ pr.prometheusRequests,
+ pr.instrumentedProcesses)
 
  return pr
 }
@@ -100,3 +106,11 @@ func (p *PrometheusReporter) OTELTraceExportError(err error) {
 func (p *PrometheusReporter) PrometheusRequest(port, path string) {
  p.prometheusRequests.WithLabelValues(port, path).Inc()
 }
+
+func (p *PrometheusReporter) InstrumentProcess(processName string) {
+ p.instrumentedProcesses.WithLabelValues(processName).Inc()
+}
+
+func (p *PrometheusReporter) UninstrumentProcess(processName string) {
+ p.instrumentedProcesses.WithLabelValues(processName).Dec()
+}

test/integration/configs/prometheus-config.yml

@@ -11,3 +11,8 @@ scrape_configs:
  static_configs:
  - targets:
  - 'otelcol:8888'
+ - job_name: autoinstrumenter-collector
+ honor_labels: true
+ static_configs:
+ - targets:
+ - '172.17.0.1:8999'

test/integration/docker-compose-multiexec.yml

@@ -134,6 +134,10 @@ services:
  BEYLA_METRICS_REPORT_TARGET: "true"
  BEYLA_METRICS_REPORT_PEER: "true"
  BEYLA_HOSTNAME: "beyla"
+ BEYLA_INTERNAL_METRICS_PROMETHEUS_PORT: 8999
+ BEYLA_INTERNAL_METRICS_PROMETHEUS_PATH: /metrics
+ ports:
+ - "8999:8999" 
 
  # OpenTelemetry Collector
  otelcol:
@@ -166,6 +170,7 @@ services:
  - --storage.tsdb.path=/prometheus
  - --web.enable-lifecycle
  - --web.route-prefix=/
+ - --log.level=debug
  volumes:
  - ./configs/:/etc/prometheus
  ports:

test/integration/multiprocess_test.go

@@ -3,8 +3,10 @@
 package integration
 
 import (
+ "fmt"
  "net/http"
  "path"
+ "strconv"
  "testing"
  "time"
 
@@ -96,6 +98,10 @@ func TestMultiProcess(t *testing.T) {
  })
  }
 
+ t.Run("Instrumented processes metric", func(t *testing.T) {
+ checkInstrumentedProcessesMetric(t)
+ })
+
  t.Run("BPF pinning folders mounted", func(t *testing.T) {
  // 1 beyla pinned map folder for all processes
  testBPFPinningMounted(t)
@@ -130,3 +136,30 @@ func checkReportedOnlyOnce(t *testing.T, baseURL, serviceName string) {
  }, test.Interval(1000*time.Millisecond))
 
 }
+
+func checkInstrumentedProcessesMetric(t *testing.T) {
+ pq := prom.Client{HostPort: prometheusHostPort}
+ test.Eventually(t, testTimeout, func(t require.TestingT) {
+ // we expected to have this in Prometheus at this point
+ processes := map[string]int{
+ "python3.11": 10,
+ "greetings": 2,
+ "java": 1,
+ "node": 2,
+ "ruby": 2,
+ "duped_service": 1,
+ "testserver": 2,
+ "rename1": 1,
+ }
+
+ for processName, expectedCount := range processes {
+ results, err := pq.Query(fmt.Sprintf(`beyla_instrumented_processes{process_name="%s"}`, processName))
+ require.NoError(t, err)
+ value, err := strconv.Atoi(results[0].Value[1].(string))
+ require.NoError(t, err)
+ assert.Equal(t, expectedCount, value)
+ }
+
+ }, test.Interval(1000*time.Millisecond))
+
+}