rewrite check_fd plugin by golang

Author: Ruijian Zhang

Dockerfile.in

@@ -28,6 +28,7 @@ RUN test -h /etc/localtime && rm -f /etc/localtime && cp /usr/share/zoneinfo/UTC
 
 ADD ./bin/node-problem-detector /node-problem-detector
 ADD ./bin/log-counter /home/kubernetes/bin/log-counter
+ADD ./bin/check-fd /home/kubernetes/bin/check-fd
 ADD config /config
 RUN chmod +x /config/plugin/*.sh
 ENTRYPOINT ["/node-problem-detector", "--system-log-monitors=/config/kernel-monitor.json"]

Makefile

@@ -102,6 +102,14 @@ version:
 		$(BUILD_TAGS) \
 		./cmd/nodeproblemdetector
 
+./bin/check-fd: $(PKG_SOURCES)
+	CGO_ENABLED=$(CGO_ENABLED) GOOS=linux GO111MODULE=on go build \
+		-mod vendor \
+		-o bin/check-fd \
+		-ldflags '-X $(PKG)/pkg/version.version=$(VERSION)' \
+		$(BUILD_TAGS) \
+		./cmd/checkfd
+
 Dockerfile: Dockerfile.in
 	sed -e 's|@BASEIMAGE@|$(BASEIMAGE)|g' $< >$@
 
@@ -118,12 +126,12 @@ e2e-test: vet fmt build-tar
 	-boskos-project-type=$(BOSKOS_PROJECT_TYPE) -job-name=$(JOB_NAME) \
 	-artifacts-dir=$(ARTIFACTS)
 
-build-binaries: ./bin/node-problem-detector ./bin/log-counter
+build-binaries: ./bin/node-problem-detector ./bin/log-counter ./bin/check-fd
 
 build-container: build-binaries Dockerfile
 	docker build -t $(IMAGE) .
 
-build-tar: ./bin/node-problem-detector ./bin/log-counter
+build-tar: ./bin/node-problem-detector ./bin/log-counter ./bin/check-fd
 	tar -zcvf $(TARBALL) bin/ config/ test/e2e-install.sh
 	sha1sum $(TARBALL)
 	md5sum $(TARBALL)
@@ -150,4 +158,5 @@ push: push-container push-tar
 clean:
 	rm -f bin/log-counter
 	rm -f bin/node-problem-detector
+	rm -f bin/check-fd
 	rm -f node-problem-detector-*.tar.gz

cmd/checkfd/check_fd.go

@@ -0,0 +1,111 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"io/ioutil"
+	"os"
+	"regexp"
+	"strconv"
+	"strings"
+	"sync"
+)
+
+const (
+	OK      = 0
+	NONOK   = 1
+	UNKNOWN = 2
+)
+
+var (
+	procPath           string
+	printCount         bool
+	percentThreshold   int
+)
+
+func init() {
+	flag.StringVar(&procPath, "p", "/proc", "actual path of /proc")
+	flag.BoolVar(&printCount,"c", false, "print numbers of used df and max")
+	flag.IntVar(&percentThreshold, "t", 80, "Warning threshold of percentage of fd usage to max limitation")
+}
+
+func main() {
+	flag.Parse()
+
+	if percentThreshold >= 100 || percentThreshold <=0 {
+		fmt.Fprintf(os.Stderr, "value of -t must between 0 and 100")
+		os.Exit(UNKNOWN)
+	}
+
+	maxPath := procPath + "/sys/fs/file-max"
+	f, err := os.Open(maxPath)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "%v", err)
+		os.Exit(UNKNOWN)
+	}
+	defer f.Close()
+	maxBytes, err := ioutil.ReadAll(f)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "%v", err)
+		os.Exit(UNKNOWN)
+	}
+	fdMax, err := strconv.Atoi(strings.TrimSpace(string(maxBytes)))
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "%v", err)
+		os.Exit(UNKNOWN)
+	}
+
+	files, err := ioutil.ReadDir(procPath)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "%v", err)
+		os.Exit(UNKNOWN)
+	}
+
+	ch := make(chan int)
+	wg := sync.WaitGroup{}
+	re := regexp.MustCompile(`[0-9][0-9]*`)
+	for _, f := range files {
+		if f.IsDir() {
+			if re.MatchString(f.Name()) {
+				wg.Add(1)
+				go countFD(procPath+"/"+f.Name()+"/fd", &wg, ch)
+			}
+		}
+	}
+
+	go func() {
+		wg.Wait()
+		close(ch)
+	}()
+
+	fdTotal := 0
+	for {
+		c, ok := <-ch
+		if !ok {
+			break
+		}
+		fdTotal += c
+	}
+
+	if printCount {
+		fmt.Fprintf(os.Stdout, "current fd usage is %d and limitaion is %d\n", fdTotal, fdMax)
+	}
+	if fdTotal > fdMax / 100 * percentThreshold {
+		fmt.Fprintf(os.Stdout, "current fd usage is %d and is over %d of limition %d, \n",
+			fdTotal, percentThreshold, fdMax)
+		os.Exit(NONOK)
+	} else {
+		fmt.Fprintf(os.Stdout, "node has no fd pressure\n")
+		os.Exit(OK)
+	}
+}
+
+func countFD(path string, wg *sync.WaitGroup, fNum chan<- int) {
+	defer wg.Done()
+	files, err := ioutil.ReadDir(path)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "%v", err)
+		return
+	}
+	fNum <- len(files)
+}