Cross platform support for the 'processes' plugin

closes #798

etc/telegraf.conf

@@ -123,6 +123,10 @@
 [[inputs.mem]]
   # no configuration
 
+# Get the number of processes and group them by status
+[[inputs.processes]]
+  # no configuration
+
 # Read metrics about swap memory usage
 [[inputs.swap]]
   # no configuration

plugins/inputs/system/processes.go

@@ -1,59 +1,165 @@
+// +build !windows
+
 package system
 
 import (
+	"bytes"
 	"fmt"
+	"io/ioutil"
 	"log"
+	"os"
+	"os/exec"
+	"path"
+	"runtime"
+	"strconv"
 
 	"github.com/influxdata/telegraf"
 	"github.com/influxdata/telegraf/plugins/inputs"
-	"github.com/shirou/gopsutil/process"
 )
 
 type Processes struct {
 }
 
 func (_ *Processes) Description() string {
-	return "Get the number of processes and group them by status (Linux only)"
+	return "Get the number of processes and group them by status"
 }
 
 func (_ *Processes) SampleConfig() string { return "" }
 
 func (s *Processes) Gather(acc telegraf.Accumulator) error {
-	pids, err := process.Pids()
+	fields := map[string]interface{}{
+		"blocked":  int64(0),
+		"zombie":   int64(0),
+		"stopped":  int64(0),
+		"running":  int64(0),
+		"sleeping": int64(0),
+		"total":    int64(0),
+	}
+
+	usePS := true
+	switch runtime.GOOS {
+	case "freebsd":
+		fields["idle"] = int64(0)
+		fields["wait"] = int64(0)
+	case "darwin":
+		fields["idle"] = int64(0)
+	case "openbsd":
+		fields["idle"] = int64(0)
+	case "linux":
+		usePS = false
+		fields["paging"] = int64(0)
+		fields["total_threads"] = int64(0)
+	}
+
+	if usePS {
+		if err := fromPS(fields); err != nil {
+			return err
+		}
+	} else {
+		if err := fromProc(fields); err != nil {
+			return err
+		}
+	}
+
+	acc.AddFields("processes", fields, nil)
+	return nil
+}
+
+// exec `ps` to get all process states
+func fromPS(fields map[string]interface{}) error {
+	bin, err := exec.LookPath("ps")
 	if err != nil {
-		return fmt.Errorf("error getting pids list: %s", err)
+		return err
 	}
-	// TODO handle other OS (Windows/BSD/Solaris/OSX)
-	fields := map[string]interface{}{
-		"paging":   uint64(0),
-		"blocked":  uint64(0),
-		"zombie":   uint64(0),
-		"stopped":  uint64(0),
-		"running":  uint64(0),
-		"sleeping": uint64(0),
-	}
-	for _, pid := range pids {
-		process, err := process.NewProcess(pid)
-		if err != nil {
-			log.Printf("Can not get process %d status: %s", pid, err)
+
+	out, err := exec.Command(bin, "axo", "state").Output()
+	if err != nil {
+		return err
+	}
+
+	for _, status := range bytes.Fields(out) {
+		switch status[0] {
+		case 'W':
+			fields["wait"] = fields["wait"].(int64) + int64(1)
+		case 'U', 'D':
+			// Also known as uninterruptible sleep or disk sleep
+			fields["blocked"] = fields["blocked"].(int64) + int64(1)
+		case 'Z':
+			fields["zombie"] = fields["zombie"].(int64) + int64(1)
+		case 'T':
+			fields["stopped"] = fields["stopped"].(int64) + int64(1)
+		case 'R':
+			fields["running"] = fields["running"].(int64) + int64(1)
+		case 'S':
+			fields["sleeping"] = fields["sleeping"].(int64) + int64(1)
+		case 'I':
+			fields["idle"] = fields["idle"].(int64) + int64(1)
+		default:
+			log.Printf("processes: Unknown state [ %s ] from ps",
+				string(status[0]))
+		}
+		fields["total"] = fields["total"].(int64) + int64(1)
+	}
+	return nil
+}
+
+// get process states from /proc/(pid)/stat files
+func fromProc(fields map[string]interface{}) error {
+	files, err := ioutil.ReadDir("/proc")
+	if err != nil {
+		return err
+	}
+
+	for _, file := range files {
+		if !file.IsDir() {
 			continue
 		}
-		status, err := process.Status()
-		if err != nil {
-			log.Printf("Can not get process %d status: %s\n", pid, err)
+
+		statFile := path.Join("/proc", file.Name(), "stat")
+		if _, err := os.Stat(statFile); os.IsNotExist(err) {
 			continue
+		} else if err != nil {
+			return err
+		}
+
+		data, err := ioutil.ReadFile(statFile)
+		if err != nil {
+			return err
+		}
+
+		stats := bytes.Fields(data)
+		if len(stats) < 3 {
+			return fmt.Errorf("Something is terribly wrong with %s", statFile)
+		}
+		switch stats[2][0] {
+		case 'R':
+			fields["running"] = fields["running"].(int64) + int64(1)
+		case 'S':
+			fields["sleeping"] = fields["sleeping"].(int64) + int64(1)
+		case 'D':
+			fields["blocked"] = fields["blocked"].(int64) + int64(1)
+		case 'Z':
+			fields["zombies"] = fields["zombies"].(int64) + int64(1)
+		case 'T', 't':
+			fields["stopped"] = fields["stopped"].(int64) + int64(1)
+		case 'W':
+			fields["paging"] = fields["paging"].(int64) + int64(1)
+		default:
+			log.Printf("processes: Unknown state [ %s ] in file %s",
+				string(stats[2][0]), statFile)
 		}
-		_, exists := fields[status]
-		if !exists {
-			log.Printf("Status '%s' for process with pid: %d\n", status, pid)
+		fields["total"] = fields["total"].(int64) + int64(1)
+
+		threads, err := strconv.Atoi(string(stats[19]))
+		if err != nil {
+			log.Printf("processes: Error parsing thread count: %s", err)
 			continue
 		}
-		fields[status] = fields[status].(uint64) + uint64(1)
+		fields["total_threads"] = fields["total_threads"].(int64) + int64(threads)
 	}
-
-	acc.AddFields("processes", fields, nil)
 	return nil
 }
+
 func init() {
 	inputs.Add("processes", func() telegraf.Input {
 		return &Processes{}

plugins/inputs/system/processes_test.go

@@ -15,7 +15,7 @@ func TestProcesses(t *testing.T) {
 	err := processes.Gather(&acc)
 	require.NoError(t, err)
 
-	assert.True(t, acc.HasUIntField("processes", "running"))
-	assert.True(t, acc.HasUIntField("processes", "sleeping"))
-	assert.True(t, acc.HasUIntField("processes", "stopped"))
+	assert.True(t, acc.HasIntField("processes", "running"))
+	assert.True(t, acc.HasIntField("processes", "sleeping"))
+	assert.True(t, acc.HasIntField("processes", "stopped"))
 }

scripts/circle-test.sh

@@ -68,7 +68,7 @@ telegraf -sample-config > $tmpdir/config.toml
 exit_if_fail telegraf -config $tmpdir/config.toml \
     -test -input-filter cpu:mem
 
-mv $GOPATH/bin/telegraf $CIRCLE_ARTIFACTS
+cat $GOPATH/bin/telegraf | gzip > $CIRCLE_ARTIFACTS/telegraf.gz
 
 eval "git describe --exact-match HEAD"
 if [ $? -eq 0 ]; then