services/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 CopyPipeToChan(in io.Reader, out chan string, done chan<- bool) {
  19         s := bufio.NewScanner(in)
  20         for s.Scan() {
  21                 out <- s.Text()
  22         }
  23         done <- true
  24 }
  25
  26 func CopyChanToPipe(in <-chan string, out io.Writer) {
  27         for s := range in {
  28                 fmt.Fprintln(out, s)
  29         }
  30 }
  31
  32 func FindStat(cgroup_root string, cgroup_parent string, container_id string, statgroup string, stat string) string {
  33         var path string
  34         path = fmt.Sprintf("%s/%s/%s/%s/%s.%s", cgroup_root, statgroup, cgroup_parent, container_id, statgroup, stat)
  35         if _, err := os.Stat(path); err == nil {
  36                 return path
  37         }
  38         path = fmt.Sprintf("%s/%s/%s/%s.%s", cgroup_root, cgroup_parent, container_id, statgroup, stat)
  39         if _, err := os.Stat(path); err == nil {
  40                 return path
  41         }
  42         path = fmt.Sprintf("%s/%s/%s.%s", cgroup_root, statgroup, statgroup, stat)
  43         if _, err := os.Stat(path); err == nil {
  44                 return path
  45         }
  46         path = fmt.Sprintf("%s/%s.%s", cgroup_root, statgroup, stat)
  47         if _, err := os.Stat(path); err == nil {
  48                 return path
  49         }
  50         return ""
  51 }
  52
  53 func PollCgroupStats(cgroup_root string, cgroup_parent string, container_id string, stderr chan string, poll int64, stop_poll_chan <-chan bool) {
  54         //var last_usage int64 = 0
  55         var last_user int64 = -1
  56         var last_sys int64 = -1
  57         var last_cpucount int64 = 0
  58
  59         type Disk struct {
  60                 last_read  int64
  61                 next_read  int64
  62                 last_write int64
  63                 next_write int64
  64         }
  65
  66         disk := make(map[string]*Disk)
  67
  68         //cpuacct_usage := FindStat(cgroup_path, "cpuacct", "usage")
  69         cpuacct_stat := FindStat(cgroup_root, cgroup_parent, container_id, "cpuacct", "stat")
  70         blkio_io_service_bytes := FindStat(cgroup_root, cgroup_parent, container_id, "blkio", "io_service_bytes")
  71         cpuset_cpus := FindStat(cgroup_root, cgroup_parent, container_id, "cpuset", "cpus")
  72         memory_stat := FindStat(cgroup_root, cgroup_parent, container_id, "memory", "stat")
  73
  74         if cpuacct_stat != "" {
  75                 stderr <- fmt.Sprintf("crunchstat: reading stats from %s", cpuacct_stat)
  76         }
  77         if blkio_io_service_bytes != "" {
  78                 stderr <- fmt.Sprintf("crunchstat: reading stats from %s", blkio_io_service_bytes)
  79         }
  80         if cpuset_cpus != "" {
  81                 stderr <- fmt.Sprintf("crunchstat: reading stats from %s", cpuset_cpus)
  82         }
  83         if memory_stat != "" {
  84                 stderr <- fmt.Sprintf("crunchstat: reading stats from %s", memory_stat)
  85         }
  86
  87         poll_chan := make(chan bool, 1)
  88         go func() {
  89                 // Send periodic poll events.
  90                 poll_chan <- true
  91                 for {
  92                         time.Sleep(time.Duration(poll) * time.Millisecond)
  93                         poll_chan <- true
  94                 }
  95         }()
  96         for {
  97                 bedtime := time.Now()
  98                 select {
  99                 case <-stop_poll_chan:
 100                         return
 101                 case <-poll_chan:
 102                         // Emit stats, then select again.
 103                 }
 104                 morning := time.Now()
 105                 elapsed := morning.Sub(bedtime).Nanoseconds() / int64(time.Millisecond)
 106                 /*{
 107                         c, _ := os.Open(cpuacct_usage)
 108                         b, _ := ioutil.ReadAll(c)
 109                         var next int64
 110                         fmt.Sscanf(string(b), "%d", &next)
 111                         if last_usage != 0 {
 112                                 stderr <- fmt.Sprintf("crunchstat: cpuacct.usage %v", (next-last_usage)/10000000)
 113                         }
 114                         //fmt.Printf("usage %d %d %d %d%%\n", last_usage, next, next-last_usage, (next-last_usage)/10000000)
 115                         last_usage = next
 116                         c.Close()
 117                 }*/
 118                 var cpus int64 = 0
 119                 if cpuset_cpus != "" {
 120                         c, err := os.Open(cpuset_cpus)
 121                         if err != nil {
 122                                 stderr <- fmt.Sprintf("open %s: %s", cpuset_cpus, err)
 123                                 continue
 124                         }
 125                         b, _ := ioutil.ReadAll(c)
 126                         sp := strings.Split(string(b), ",")
 127                         for _, v := range sp {
 128                                 var min, max int64
 129                                 n, _ := fmt.Sscanf(v, "%d-%d", &min, &max)
 130                                 if n == 2 {
 131                                         cpus += (max - min) + 1
 132                                 } else {
 133                                         cpus += 1
 134                                 }
 135                         }
 136
 137                         if cpus != last_cpucount {
 138                                 stderr <- fmt.Sprintf("crunchstat: cpuset.cpus %v", cpus)
 139                         }
 140                         last_cpucount = cpus
 141
 142                         c.Close()
 143                 }
 144                 if cpus == 0 {
 145                         cpus = 1
 146                 }
 147                 if cpuacct_stat != "" {
 148                         c, err := os.Open(cpuacct_stat)
 149                         if err != nil {
 150                                 stderr <- fmt.Sprintf("open %s: %s", cpuacct_stat, err)
 151                                 // Next time around, last_user would
 152                                 // be >1 interval old, so stats will
 153                                 // be incorrect. Start over instead.
 154                                 last_user = -1
 155                                 continue
 156                         }
 157                         b, _ := ioutil.ReadAll(c)
 158                         var next_user int64
 159                         var next_sys int64
 160                         fmt.Sscanf(string(b), "user %d\nsystem %d", &next_user, &next_sys)
 161                         c.Close()
 162
 163                         if elapsed > 0 && last_user != -1 {
 164                                 user_diff := next_user - last_user
 165                                 sys_diff := next_sys - last_sys
 166                                 // Assume we're reading stats based on 100
 167                                 // jiffies per second.  Because the elapsed
 168                                 // time is in milliseconds, we need to boost
 169                                 // that to 1000 jiffies per second, then boost
 170                                 // it by another 100x to get a percentage, then
 171                                 // finally divide by the actual elapsed time
 172                                 // and the number of cpus to get average load
 173                                 // over the polling period.
 174                                 user_pct := (user_diff * 10 * 100) / (elapsed * cpus)
 175                                 sys_pct := (sys_diff * 10 * 100) / (elapsed * cpus)
 176
 177                                 stderr <- fmt.Sprintf("crunchstat: cpuacct.stat user %v", user_pct)
 178                                 stderr <- fmt.Sprintf("crunchstat: cpuacct.stat sys %v", sys_pct)
 179                         }
 180
 181                         /*fmt.Printf("user %d %d %d%%\n", last_user, next_user, next_user-last_user)
 182                         fmt.Printf("sys %d %d %d%%\n", last_sys, next_sys, next_sys-last_sys)
 183                         fmt.Printf("sum %d%%\n", (next_user-last_user)+(next_sys-last_sys))*/
 184                         last_user = next_user
 185                         last_sys = next_sys
 186                 }
 187                 if blkio_io_service_bytes != "" {
 188                         c, err := os.Open(blkio_io_service_bytes)
 189                         if err != nil {
 190                                 stderr <- fmt.Sprintf("open %s: %s", blkio_io_service_bytes, err)
 191                                 continue
 192                         }
 193                         b := bufio.NewScanner(c)
 194                         var device, op string
 195                         var next int64
 196                         for b.Scan() {
 197                                 if _, err := fmt.Sscanf(string(b.Text()), "%s %s %d", &device, &op, &next); err == nil {
 198                                         if disk[device] == nil {
 199                                                 disk[device] = new(Disk)
 200                                         }
 201                                         if op == "Read" {
 202                                                 disk[device].last_read = disk[device].next_read
 203                                                 disk[device].next_read = next
 204                                                 if disk[device].last_read > 0 && (disk[device].next_read != disk[device].last_read) {
 205                                                         stderr <- fmt.Sprintf("crunchstat: blkio.io_service_bytes %s read %v", device, disk[device].next_read-disk[device].last_read)
 206                                                 }
 207                                         }
 208                                         if op == "Write" {
 209                                                 disk[device].last_write = disk[device].next_write
 210                                                 disk[device].next_write = next
 211                                                 if disk[device].last_write > 0 && (disk[device].next_write != disk[device].last_write) {
 212                                                         stderr <- fmt.Sprintf("crunchstat: blkio.io_service_bytes %s write %v", device, disk[device].next_write-disk[device].last_write)
 213                                                 }
 214                                         }
 215                                 }
 216                         }
 217                         c.Close()
 218                 }
 219
 220                 if memory_stat != "" {
 221                         c, err := os.Open(memory_stat)
 222                         if err != nil {
 223                                 stderr <- fmt.Sprintf("open %s: %s", memory_stat, err)
 224                                 continue
 225                         }
 226                         b := bufio.NewScanner(c)
 227                         var stat string
 228                         var val int64
 229                         for b.Scan() {
 230                                 if _, err := fmt.Sscanf(string(b.Text()), "%s %d", &stat, &val); err == nil {
 231                                         if stat == "rss" {
 232                                                 stderr <- fmt.Sprintf("crunchstat: memory.stat rss %v", val)
 233                                         }
 234                                 }
 235                         }
 236                         c.Close()
 237                 }
 238         }
 239 }
 240
 241 func run(logger *log.Logger) error {
 242
 243         var (
 244                 cgroup_root    string
 245                 cgroup_parent  string
 246                 cgroup_cidfile string
 247                 wait           int64
 248                 poll           int64
 249         )
 250
 251         flag.StringVar(&cgroup_root, "cgroup-root", "", "Root of cgroup tree")
 252         flag.StringVar(&cgroup_parent, "cgroup-parent", "", "Name of container parent under cgroup")
 253         flag.StringVar(&cgroup_cidfile, "cgroup-cid", "", "Path to container id file")
 254         flag.Int64Var(&wait, "wait", 5, "Maximum time (in seconds) to wait for cid file to show up")
 255         flag.Int64Var(&poll, "poll", 1000, "Polling frequency, in milliseconds")
 256
 257         flag.Parse()
 258
 259         if cgroup_root == "" {
 260                 logger.Fatal("Must provide -cgroup-root")
 261         }
 262
 263         stderr_chan := make(chan string, 1)
 264         defer close(stderr_chan)
 265         finish_chan := make(chan bool)
 266         defer close(finish_chan)
 267
 268         go CopyChanToPipe(stderr_chan, os.Stderr)
 269
 270         var cmd *exec.Cmd
 271
 272         if len(flag.Args()) > 0 {
 273                 // Set up subprocess
 274                 cmd = exec.Command(flag.Args()[0], flag.Args()[1:]...)
 275
 276                 logger.Print("Running ", flag.Args())
 277
 278                 // Child process will use our stdin and stdout pipes
 279                 // (we close our copies below)
 280                 cmd.Stdin = os.Stdin
 281                 cmd.Stdout = os.Stdout
 282
 283                 // Forward SIGINT and SIGTERM to inner process
 284                 term := make(chan os.Signal, 1)
 285                 go func(sig <-chan os.Signal) {
 286                         catch := <-sig
 287                         if cmd.Process != nil {
 288                                 cmd.Process.Signal(catch)
 289                         }
 290                         logger.Print("caught signal: ", catch)
 291                 }(term)
 292                 signal.Notify(term, syscall.SIGTERM)
 293                 signal.Notify(term, syscall.SIGINT)
 294
 295                 // Funnel stderr through our channel
 296                 stderr_pipe, err := cmd.StderrPipe()
 297                 if err != nil {
 298                         logger.Fatal(err)
 299                 }
 300                 go CopyPipeToChan(stderr_pipe, stderr_chan, finish_chan)
 301
 302                 // Run subprocess
 303                 if err := cmd.Start(); err != nil {
 304                         logger.Fatal(err)
 305                 }
 306
 307                 // Close stdin/stdout in this (parent) process
 308                 os.Stdin.Close()
 309                 os.Stdout.Close()
 310         }
 311
 312         // Read the cid file
 313         var container_id string
 314         if cgroup_cidfile != "" {
 315                 // wait up to 'wait' seconds for the cid file to appear
 316                 ok := false
 317                 var i time.Duration
 318                 for i = 0; i < time.Duration(wait)*time.Second; i += (100 * time.Millisecond) {
 319                         f, err := os.Open(cgroup_cidfile)
 320                         if err == nil {
 321                                 defer f.Close()
 322                                 cid, err2 := ioutil.ReadAll(f)
 323                                 if err2 == nil && len(cid) > 0 {
 324                                         ok = true
 325                                         container_id = string(cid)
 326                                         break
 327                                 }
 328                         }
 329                         time.Sleep(100 * time.Millisecond)
 330                 }
 331                 if !ok {
 332                         logger.Printf("Could not read cid file %s", cgroup_cidfile)
 333                 }
 334         }
 335
 336         stop_poll_chan := make(chan bool, 1)
 337         go PollCgroupStats(cgroup_root, cgroup_parent, container_id, stderr_chan, poll, stop_poll_chan)
 338
 339         // When the child exits, tell the polling goroutine to stop.
 340         defer func() { stop_poll_chan <- true }()
 341
 342         // Wait for CopyPipeToChan to consume child's stderr pipe
 343         <-finish_chan
 344
 345         return cmd.Wait()
 346 }
 347
 348 func main() {
 349         logger := log.New(os.Stderr, "crunchstat: ", 0)
 350         if err := run(logger); err != nil {
 351                 if exiterr, ok := err.(*exec.ExitError); ok {
 352                         // The program has exited with an exit code != 0
 353
 354                         // This works on both Unix and
 355                         // Windows. Although package syscall is
 356                         // generally platform dependent, WaitStatus is
 357                         // defined for both Unix and Windows and in
 358                         // both cases has an ExitStatus() method with
 359                         // the same signature.
 360                         if status, ok := exiterr.Sys().(syscall.WaitStatus); ok {
 361                                 os.Exit(status.ExitStatus())
 362                         }
 363                 } else {
 364                         logger.Fatalf("cmd.Wait: %v", err)
 365                 }
 366         }
 367 }