services/crunch/crunchstat/src/arvados.org/crunchstat/crunchstat.go

   1 package main
   2
   3 import (
   4         "bufio"
   5         "flag"
   6         "fmt"
   7         "io"
   8         "io/ioutil"
   9         "log"
  10         "os"
  11         "os/exec"
  12         "os/signal"
  13         "strings"
  14         "syscall"
  15         "time"
  16 )
  17
  18 func ReadLineByLine(inp io.ReadCloser, out chan string, finish chan bool) {
  19         s := bufio.NewScanner(inp)
  20         for s.Scan() {
  21                 out <- s.Text()
  22         }
  23         finish <- true
  24 }
  25
  26 func OutputChannel(stdout chan string, stderr chan string) {
  27         for {
  28                 select {
  29                 case s, ok := <-stdout:
  30                         if ok {
  31                                 fmt.Fprintln(os.Stdout, s)
  32                         } else {
  33                                 return
  34                         }
  35                 case s, ok := <-stderr:
  36                         if ok {
  37                                 fmt.Fprintln(os.Stderr, s)
  38                         } else {
  39                                 return
  40                         }
  41                 }
  42         }
  43 }
  44
  45 func FindStat(cgroup_path string, statgroup string, stat string) string {
  46         path := fmt.Sprintf("%s/%s.%s", cgroup_path, statgroup, stat)
  47         if _, err := os.Stat(path); err != nil {
  48                 return path
  49         }
  50         path = fmt.Sprintf("%s/%s/%s.%s", cgroup_path, statgroup, statgroup, stat)
  51         if _, err := os.Stat(path); err != nil {
  52                 return path
  53         }
  54         return ""
  55 }
  56
  57 func PollCgroupStats(cgroup_path string, stderr chan string, poll int64) {
  58         //var last_usage int64 = 0
  59         var last_user int64 = 0
  60         var last_sys int64 = 0
  61         var last_cpucount int64 = 0
  62
  63         type Disk struct {
  64                 last_read  int64
  65                 next_read  int64
  66                 last_write int64
  67                 next_write int64
  68         }
  69
  70         disk := make(map[string]*Disk)
  71
  72         //cpuacct_usage := FindStat(cgroup_path, "cpuacct", "usage")
  73         cpuacct_stat := FindStat(cgroup_path, "cpuacct", "stat")
  74         blkio_io_service_bytes := FindStat(cgroup_path, "blkio", "io_service_bytes")
  75         cpuset_cpus := FindStat(cgroup_path, "cpuset", "cpus")
  76         memory_stat := FindStat(cgroup_path, "memory", "stat")
  77
  78         var elapsed int64 = poll
  79
  80         for {
  81                 /*{
  82                         c, _ := os.Open(cpuacct_usage)
  83                         b, _ := ioutil.ReadAll(c)
  84                         var next int64
  85                         fmt.Sscanf(string(b), "%d", &next)
  86                         if last_usage != 0 {
  87                                 stderr <- fmt.Sprintf("crunchstat: cpuacct.usage %v", (next-last_usage)/10000000)
  88                         }
  89                         //fmt.Printf("usage %d %d %d %d%%\n", last_usage, next, next-last_usage, (next-last_usage)/10000000)
  90                         last_usage = next
  91                         c.Close()
  92                 }*/
  93                 var cpus int64 = 0
  94                 if cpuset_cpus != "" {
  95                         c, _ := os.Open(cpuset_cpus)
  96                         b, _ := ioutil.ReadAll(c)
  97                         sp := strings.Split(string(b), ",")
  98                         for _, v := range sp {
  99                                 var min, max int64
 100                                 n, _ := fmt.Sscanf(v, "%d-%d", &min, &max)
 101                                 if n == 2 {
 102                                         cpus += (max - min) + 1
 103                                 } else {
 104                                         cpus += 1
 105                                 }
 106                         }
 107
 108                         if cpus != last_cpucount {
 109                                 stderr <- fmt.Sprintf("crunchstat: cpuset.cpus %v", cpus)
 110                         }
 111                         last_cpucount = cpus
 112
 113                         c.Close()
 114                 }
 115                 if cpus == 0 {
 116                         cpus = 1
 117                 }
 118                 if cpuacct_stat != "" {
 119                         c, _ := os.Open(cpuacct_stat)
 120                         b, _ := ioutil.ReadAll(c)
 121                         var next_user int64
 122                         var next_sys int64
 123                         fmt.Sscanf(string(b), "user %d\nsystem %d", &next_user, &next_sys)
 124                         c.Close()
 125
 126                         if last_user != 0 {
 127                                 user_diff := next_user - last_user
 128                                 sys_diff := next_sys - last_sys
 129                                 // Assume we're reading stats based on 100
 130                                 // jiffies per second.  Because the ellaspsed
 131                                 // time is in milliseconds, we need to boost
 132                                 // that to 1000 jiffies per second, then boost
 133                                 // it by another 100x to get a percentage, then
 134                                 // finally divide by the actual elapsed time
 135                                 // and the number of cpus to get average load
 136                                 // over the polling period.
 137                                 user_pct := (user_diff * 10 * 100) / (elapsed * cpus)
 138                                 sys_pct := (sys_diff * 10 * 100) / (elapsed * cpus)
 139
 140                                 stderr <- fmt.Sprintf("crunchstat: cpuacct.stat user %v", user_pct)
 141                                 stderr <- fmt.Sprintf("crunchstat: cpuacct.stat sys %v", sys_pct)
 142                         }
 143
 144                         /*fmt.Printf("user %d %d %d%%\n", last_user, next_user, next_user-last_user)
 145                         fmt.Printf("sys %d %d %d%%\n", last_sys, next_sys, next_sys-last_sys)
 146                         fmt.Printf("sum %d%%\n", (next_user-last_user)+(next_sys-last_sys))*/
 147                         last_user = next_user
 148                         last_sys = next_sys
 149                 }
 150                 if blkio_io_service_bytes != "" {
 151                         c, _ := os.Open(blkio_io_service_bytes)
 152                         b := bufio.NewScanner(c)
 153                         var device, op string
 154                         var next int64
 155                         for b.Scan() {
 156                                 if _, err := fmt.Sscanf(string(b.Text()), "%s %s %d", &device, &op, &next); err == nil {
 157                                         if disk[device] == nil {
 158                                                 disk[device] = new(Disk)
 159                                         }
 160                                         if op == "Read" {
 161                                                 disk[device].last_read = disk[device].next_read
 162                                                 disk[device].next_read = next
 163                                                 if disk[device].last_read > 0 {
 164                                                         stderr <- fmt.Sprintf("crunchstat: blkio.io_service_bytes %s read %v", device, disk[device].next_read-disk[device].last_read)
 165                                                 }
 166                                         }
 167                                         if op == "Write" {
 168                                                 disk[device].last_write = disk[device].next_write
 169                                                 disk[device].next_write = next
 170                                                 if disk[device].last_write > 0 {
 171                                                         stderr <- fmt.Sprintf("crunchstat: blkio.io_service_bytes %s write %v", device, disk[device].next_write-disk[device].last_write)
 172                                                 }
 173                                         }
 174                                 }
 175                         }
 176                         c.Close()
 177                 }
 178
 179                 if memory_stat != "" {
 180                         c, _ := os.Open(memory_stat)
 181                         b := bufio.NewScanner(c)
 182                         var stat string
 183                         var val int64
 184                         for b.Scan() {
 185                                 if _, err := fmt.Sscanf(string(b.Text()), "%s %d", &stat, &val); err == nil {
 186                                         if stat == "rss" {
 187                                                 stderr <- fmt.Sprintf("crunchstat: memory.stat rss %v", val)
 188                                         }
 189                                 }
 190                         }
 191                         c.Close()
 192                 }
 193
 194                 bedtime := time.Now()
 195                 time.Sleep(time.Duration(poll) * time.Millisecond)
 196                 morning := time.Now()
 197                 elapsed = morning.Sub(bedtime).Nanoseconds() / int64(time.Millisecond)
 198         }
 199 }
 200
 201 func main() {
 202
 203         var (
 204                 cgroup_path    string
 205                 cgroup_parent  string
 206                 cgroup_cidfile string
 207                 wait           int64
 208                 poll           int64
 209         )
 210
 211         flag.StringVar(&cgroup_path, "cgroup-path", "", "Direct path to cgroup")
 212         flag.StringVar(&cgroup_parent, "cgroup-parent", "", "Path to parent cgroup")
 213         flag.StringVar(&cgroup_cidfile, "cgroup-cid", "", "Path to container id file")
 214         flag.Int64Var(&wait, "wait", 5, "Maximum time (in seconds) to wait for cid file to show up")
 215         flag.Int64Var(&poll, "poll", 1000, "Polling frequency, in milliseconds")
 216
 217         flag.Parse()
 218
 219         logger := log.New(os.Stderr, "crunchstat: ", 0)
 220
 221         if cgroup_path == "" && cgroup_cidfile == "" {
 222                 logger.Fatal("Must provide either -cgroup-path or -cgroup-cid")
 223         }
 224
 225         // Make output channel
 226         stdout_chan := make(chan string)
 227         stderr_chan := make(chan string)
 228         finish_chan := make(chan bool)
 229         defer close(stdout_chan)
 230         defer close(stderr_chan)
 231         defer close(finish_chan)
 232
 233         go OutputChannel(stdout_chan, stderr_chan)
 234
 235         var cmd *exec.Cmd
 236
 237         if len(flag.Args()) > 0 {
 238                 // Set up subprocess
 239                 cmd = exec.Command(flag.Args()[0], flag.Args()[1:]...)
 240
 241                 logger.Print("Running ", flag.Args())
 242
 243                 // Forward SIGINT and SIGTERM to inner process
 244                 term := make(chan os.Signal, 1)
 245                 go func(sig <-chan os.Signal) {
 246                         catch := <-sig
 247                         if cmd.Process != nil {
 248                                 cmd.Process.Signal(catch)
 249                         }
 250                         logger.Print("caught signal:", catch)
 251                 }(term)
 252                 signal.Notify(term, syscall.SIGTERM)
 253                 signal.Notify(term, syscall.SIGINT)
 254
 255                 // Funnel stdout and stderr from subprocess to output channels
 256                 stdout_pipe, err := cmd.StdoutPipe()
 257                 if err != nil {
 258                         logger.Fatal(err)
 259                 }
 260                 go ReadLineByLine(stdout_pipe, stdout_chan, finish_chan)
 261
 262                 stderr_pipe, err := cmd.StderrPipe()
 263                 if err != nil {
 264                         logger.Fatal(err)
 265                 }
 266                 go ReadLineByLine(stderr_pipe, stderr_chan, finish_chan)
 267
 268                 // Run subprocess
 269                 if err := cmd.Start(); err != nil {
 270                         logger.Fatal(err)
 271                 }
 272         }
 273
 274         // Read the cid file
 275         if cgroup_cidfile != "" {
 276                 // wait up to 'wait' seconds for the cid file to appear
 277                 var i time.Duration
 278                 for i = 0; i < time.Duration(wait)*time.Second; i += (100 * time.Millisecond) {
 279                         f, err := os.Open(cgroup_cidfile)
 280                         if err == nil {
 281                                 cid, err2 := ioutil.ReadAll(f)
 282                                 if err2 == nil && len(cid) > 0 {
 283                                         cgroup_path = string(cid)
 284                                         f.Close()
 285                                         break
 286                                 }
 287                         }
 288                         time.Sleep(100 * time.Millisecond)
 289                 }
 290                 if cgroup_path == "" {
 291                         logger.Printf("Could not read cid file %s", cgroup_cidfile)
 292                 }
 293         }
 294
 295         // add the parent prefix
 296         if cgroup_parent != "" {
 297                 cgroup_path = fmt.Sprintf("%s/%s", cgroup_parent, cgroup_path)
 298         }
 299
 300         logger.Print("Using cgroup ", cgroup_path)
 301
 302         go PollCgroupStats(cgroup_path, stderr_chan, poll)
 303
 304         // Wait for each of stdout and stderr to drain
 305         <-finish_chan
 306         <-finish_chan
 307
 308         if err := cmd.Wait(); err != nil {
 309                 if exiterr, ok := err.(*exec.ExitError); ok {
 310                         // The program has exited with an exit code != 0
 311
 312                         // This works on both Unix and Windows. Although package
 313                         // syscall is generally platform dependent, WaitStatus is
 314                         // defined for both Unix and Windows and in both cases has
 315                         // an ExitStatus() method with the same signature.
 316                         if status, ok := exiterr.Sys().(syscall.WaitStatus); ok {
 317                                 os.Exit(status.ExitStatus())
 318                         }
 319                 } else {
 320                         logger.Fatalf("cmd.Wait: %v", err)
 321                 }
 322         }
 323 }