X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/db0472215ce261406503d7336006bcd64c9feacd..f8067dd18b72705f3317e85745e87cffc9e25313:/services/crunchstat/crunchstat.go diff --git a/services/crunchstat/crunchstat.go b/services/crunchstat/crunchstat.go index 1531ec39a9..e35e98aa59 100644 --- a/services/crunchstat/crunchstat.go +++ b/services/crunchstat/crunchstat.go @@ -12,6 +12,7 @@ import ( "os" "os/exec" "os/signal" + "strconv" "strings" "syscall" "time" @@ -33,11 +34,19 @@ type Cgroup struct { cid string } -func CopyPipeToChan(in io.Reader, out chan string, done chan<- bool) { +func CopyPipeToChan(in io.ReadCloser, out chan string, done chan<- bool) { + defer in.Close() + + // TODO(twp): handle long input records gracefully, if possible + // without killing the child task (#4889) + // s := bufio.NewScanner(in) for s.Scan() { out <- s.Text() } + if s.Err() != nil { + out <- fmt.Sprintf("crunchstat: line buffering error: %s", s.Err()) + } done <- true } @@ -47,22 +56,19 @@ func CopyChanToPipe(in <-chan string, out io.Writer) { } } -func OpenAndReadAll(filename string, log_chan chan<- string) ([]byte, error) { - in, err := os.Open(filename) - if err != nil { - if log_chan != nil { - log_chan <- fmt.Sprintf("crunchstat: open %s: %s", filename, err) - } - return nil, err +var logChan chan string + +func LogPrintf(format string, args ...interface{}) { + if logChan == nil { + return } - defer in.Close() - return ReadAllOrWarn(in, log_chan) + logChan <- fmt.Sprintf("crunchstat: "+format, args...) } -func ReadAllOrWarn(in *os.File, log_chan chan<- string) ([]byte, error) { +func ReadAllOrWarn(in *os.File) ([]byte, error) { content, err := ioutil.ReadAll(in) - if err != nil && log_chan != nil { - log_chan <- fmt.Sprintf("crunchstat: read %s: %s", in.Name(), err) + if err != nil { + LogPrintf("read %s: %s", in.Name(), err) } return content, err } @@ -77,12 +83,13 @@ var reportedStatFile = map[string]string{} // container, and read /proc/PID/cgroup to determine the appropriate // cgroup root for the given statgroup. (This will avoid falling back // to host-level stats during container setup and teardown.) -func OpenStatFile(stderr chan<- string, cgroup Cgroup, statgroup string, stat string) (*os.File, error) { - var paths = []string{} - paths = append(paths, fmt.Sprintf("%s/%s/%s/%s/%s", cgroup.root, statgroup, cgroup.parent, cgroup.cid, stat)) - paths = append(paths, fmt.Sprintf("%s/%s/%s/%s", cgroup.root, cgroup.parent, cgroup.cid, stat)) - paths = append(paths, fmt.Sprintf("%s/%s/%s", cgroup.root, statgroup, stat)) - paths = append(paths, fmt.Sprintf("%s/%s", cgroup.root, stat)) +func OpenStatFile(cgroup Cgroup, statgroup string, stat string) (*os.File, error) { + var paths = []string{ + fmt.Sprintf("%s/%s/%s/%s/%s", cgroup.root, statgroup, cgroup.parent, cgroup.cid, stat), + fmt.Sprintf("%s/%s/%s/%s", cgroup.root, cgroup.parent, cgroup.cid, stat), + fmt.Sprintf("%s/%s/%s", cgroup.root, statgroup, stat), + fmt.Sprintf("%s/%s", cgroup.root, stat), + } var path string var file *os.File var err error @@ -103,16 +110,16 @@ func OpenStatFile(stderr chan<- string, cgroup Cgroup, statgroup string, stat st // [b] after all contained processes have exited. reportedStatFile[stat] = path if path == "" { - stderr <- fmt.Sprintf("crunchstat: did not find stats file: stat %s, statgroup %s, cid %s, parent %s, root %s", stat, statgroup, cgroup.cid, cgroup.parent, cgroup.root) + LogPrintf("did not find stats file: stat %s, statgroup %s, cid %s, parent %s, root %s", stat, statgroup, cgroup.cid, cgroup.parent, cgroup.root) } else { - stderr <- fmt.Sprintf("crunchstat: reading stats from %s", path) + LogPrintf("reading stats from %s", path) } } return file, err } -func GetContainerNetStats(stderr chan<- string, cgroup Cgroup) (io.Reader, error) { - procsFile, err := OpenStatFile(stderr, cgroup, "cpuacct", "cgroup.procs") +func GetContainerNetStats(cgroup Cgroup) (io.Reader, error) { + procsFile, err := OpenStatFile(cgroup, "cpuacct", "cgroup.procs") if err != nil { return nil, err } @@ -121,8 +128,9 @@ func GetContainerNetStats(stderr chan<- string, cgroup Cgroup) (io.Reader, error for reader.Scan() { taskPid := reader.Text() statsFilename := fmt.Sprintf("/proc/%s/net/dev", taskPid) - stats, err := OpenAndReadAll(statsFilename, stderr) + stats, err := ioutil.ReadFile(statsFilename) if err != nil { + LogPrintf("read %s: %s", statsFilename, err) continue } return strings.NewReader(string(stats)), nil @@ -136,8 +144,8 @@ type IoSample struct { rxBytes int64 } -func DoBlkIoStats(stderr chan<- string, cgroup Cgroup, lastSample map[string]IoSample) { - c, err := OpenStatFile(stderr, cgroup, "blkio", "blkio.io_service_bytes") +func DoBlkIoStats(cgroup Cgroup, lastSample map[string]IoSample) { + c, err := OpenStatFile(cgroup, "blkio", "blkio.io_service_bytes") if err != nil { return } @@ -175,7 +183,7 @@ func DoBlkIoStats(stderr chan<- string, cgroup Cgroup, lastSample map[string]IoS sample.txBytes-prev.txBytes, sample.rxBytes-prev.rxBytes) } - stderr <- fmt.Sprintf("crunchstat: blkio:%s %d write %d read%s", dev, sample.txBytes, sample.rxBytes, delta) + LogPrintf("blkio:%s %d write %d read%s", dev, sample.txBytes, sample.rxBytes, delta) lastSample[dev] = sample } } @@ -185,15 +193,15 @@ type MemSample struct { memStat map[string]int64 } -func DoMemoryStats(stderr chan<- string, cgroup Cgroup) { - c, err := OpenStatFile(stderr, cgroup, "memory", "memory.stat") +func DoMemoryStats(cgroup Cgroup) { + c, err := OpenStatFile(cgroup, "memory", "memory.stat") if err != nil { return } defer c.Close() b := bufio.NewScanner(c) thisSample := MemSample{time.Now(), make(map[string]int64)} - wantStats := [...]string{"cache", "pgmajfault", "rss"} + wantStats := [...]string{"cache", "swap", "pgmajfault", "rss"} for b.Scan() { var stat string var val int64 @@ -208,58 +216,49 @@ func DoMemoryStats(stderr chan<- string, cgroup Cgroup) { outstat.WriteString(fmt.Sprintf(" %d %s", val, key)) } } - stderr <- fmt.Sprintf("crunchstat: mem%s", outstat.String()) + LogPrintf("mem%s", outstat.String()) } -func DoNetworkStats(stderr chan<- string, cgroup Cgroup, lastSample map[string]IoSample) { +func DoNetworkStats(cgroup Cgroup, lastSample map[string]IoSample) { sampleTime := time.Now() - stats, err := GetContainerNetStats(stderr, cgroup) + stats, err := GetContainerNetStats(cgroup) if err != nil { return } scanner := bufio.NewScanner(stats) -Iface: for scanner.Scan() { var ifName string var rx, tx int64 - words := bufio.NewScanner(strings.NewReader(scanner.Text())) - words.Split(bufio.ScanWords) - wordIndex := 0 - for words.Scan() { - word := words.Text() - switch wordIndex { - case 0: - ifName = strings.TrimRight(word, ":") - case 1: - if _, err := fmt.Sscanf(word, "%d", &rx); err != nil { - continue Iface - } - case 9: - if _, err := fmt.Sscanf(word, "%d", &tx); err != nil { - continue Iface - } - } - wordIndex++ + words := strings.Fields(scanner.Text()) + if len(words) != 17 { + // Skip lines with wrong format + continue } - if ifName == "lo" || ifName == "" || wordIndex != 17 { + ifName = strings.TrimRight(words[0], ":") + if ifName == "lo" || ifName == "" { // Skip loopback interface and lines with wrong format continue } + if tx, err = strconv.ParseInt(words[9], 10, 64); err != nil { + continue + } + if rx, err = strconv.ParseInt(words[1], 10, 64); err != nil { + continue + } nextSample := IoSample{} nextSample.sampleTime = sampleTime nextSample.txBytes = tx nextSample.rxBytes = rx var delta string - if lastSample, ok := lastSample[ifName]; ok { - interval := nextSample.sampleTime.Sub(lastSample.sampleTime).Seconds() + if prev, ok := lastSample[ifName]; ok { + interval := nextSample.sampleTime.Sub(prev.sampleTime).Seconds() delta = fmt.Sprintf(" -- interval %.4f seconds %d tx %d rx", interval, - tx-lastSample.txBytes, - rx-lastSample.rxBytes) + tx-prev.txBytes, + rx-prev.rxBytes) } - stderr <- fmt.Sprintf("crunchstat: net:%s %d tx %d rx%s", - ifName, tx, rx, delta) + LogPrintf("net:%s %d tx %d rx%s", ifName, tx, rx, delta) lastSample[ifName] = nextSample } } @@ -274,13 +273,13 @@ type CpuSample struct { // Return the number of CPUs available in the container. Return 0 if // we can't figure out the real number of CPUs. -func GetCpuCount(stderr chan<- string, cgroup Cgroup) int64 { - cpusetFile, err := OpenStatFile(stderr, cgroup, "cpuset", "cpuset.cpus") +func GetCpuCount(cgroup Cgroup) int64 { + cpusetFile, err := OpenStatFile(cgroup, "cpuset", "cpuset.cpus") if err != nil { return 0 } defer cpusetFile.Close() - b, err := ReadAllOrWarn(cpusetFile, stderr) + b, err := ReadAllOrWarn(cpusetFile) sp := strings.Split(string(b), ",") cpus := int64(0) for _, v := range sp { @@ -295,18 +294,18 @@ func GetCpuCount(stderr chan<- string, cgroup Cgroup) int64 { return cpus } -func DoCpuStats(stderr chan<- string, cgroup Cgroup, lastSample *CpuSample) { - statFile, err := OpenStatFile(stderr, cgroup, "cpuacct", "cpuacct.stat") +func DoCpuStats(cgroup Cgroup, lastSample *CpuSample) { + statFile, err := OpenStatFile(cgroup, "cpuacct", "cpuacct.stat") if err != nil { return } defer statFile.Close() - b, err := ReadAllOrWarn(statFile, stderr) + b, err := ReadAllOrWarn(statFile) if err != nil { return } - nextSample := CpuSample{true, time.Now(), 0, 0, GetCpuCount(stderr, cgroup)} + nextSample := CpuSample{true, time.Now(), 0, 0, GetCpuCount(cgroup)} var userTicks, sysTicks int64 fmt.Sscanf(string(b), "user %d\nsystem %d", &userTicks, &sysTicks) user_hz := float64(C.sysconf(C._SC_CLK_TCK)) @@ -320,12 +319,12 @@ func DoCpuStats(stderr chan<- string, cgroup Cgroup, lastSample *CpuSample) { nextSample.user-lastSample.user, nextSample.sys-lastSample.sys) } - stderr <- fmt.Sprintf("crunchstat: cpu %.4f user %.4f sys %d cpus%s", + LogPrintf("cpu %.4f user %.4f sys %d cpus%s", nextSample.user, nextSample.sys, nextSample.cpus, delta) *lastSample = nextSample } -func PollCgroupStats(cgroup Cgroup, stderr chan string, poll int64, stop_poll_chan <-chan bool) { +func PollCgroupStats(cgroup Cgroup, poll int64, stop_poll_chan <-chan bool) { var lastNetSample = map[string]IoSample{} var lastDiskSample = map[string]IoSample{} var lastCpuSample = CpuSample{} @@ -346,10 +345,10 @@ func PollCgroupStats(cgroup Cgroup, stderr chan string, poll int64, stop_poll_ch case <-poll_chan: // Emit stats, then select again. } - DoMemoryStats(stderr, cgroup) - DoCpuStats(stderr, cgroup, &lastCpuSample) - DoBlkIoStats(stderr, cgroup, lastDiskSample) - DoNetworkStats(stderr, cgroup, lastNetSample) + DoMemoryStats(cgroup) + DoCpuStats(cgroup, &lastCpuSample) + DoBlkIoStats(cgroup, lastDiskSample) + DoNetworkStats(cgroup, lastNetSample) } } @@ -375,12 +374,12 @@ func run(logger *log.Logger) error { logger.Fatal("Must provide -cgroup-root") } - stderr_chan := make(chan string, 1) - defer close(stderr_chan) + logChan = make(chan string, 1) + defer close(logChan) finish_chan := make(chan bool) defer close(finish_chan) - go CopyChanToPipe(stderr_chan, os.Stderr) + go CopyChanToPipe(logChan, os.Stderr) var cmd *exec.Cmd @@ -412,7 +411,7 @@ func run(logger *log.Logger) error { if err != nil { logger.Fatal(err) } - go CopyPipeToChan(stderr_pipe, stderr_chan, finish_chan) + go CopyPipeToChan(stderr_pipe, logChan, finish_chan) // Run subprocess if err := cmd.Start(); err != nil { @@ -431,7 +430,7 @@ func run(logger *log.Logger) error { ok := false var i time.Duration for i = 0; i < time.Duration(wait)*time.Second; i += (100 * time.Millisecond) { - cid, err := OpenAndReadAll(cgroup_cidfile, nil) + cid, err := ioutil.ReadFile(cgroup_cidfile) if err == nil && len(cid) > 0 { ok = true container_id = string(cid) @@ -446,7 +445,7 @@ func run(logger *log.Logger) error { stop_poll_chan := make(chan bool, 1) cgroup := Cgroup{cgroup_root, cgroup_parent, container_id} - go PollCgroupStats(cgroup, stderr_chan, poll, stop_poll_chan) + go PollCgroupStats(cgroup, poll, stop_poll_chan) // When the child exits, tell the polling goroutine to stop. defer func() { stop_poll_chan <- true }()