Don't print blkio stats if no reads or writes have occurred since the last poll.

[arvados.git] / services / crunch / crunchstat / src / arvados.org / crunchstat / crunchstat.go

diff --git a/services/crunch/crunchstat/src/arvados.org/crunchstat/crunchstat.go b/services/crunch/crunchstat/src/arvados.org/crunchstat/crunchstat.go
index 09a99e5df2a5b72eb38671df4e250aa9d109d983..d61871da6475dcdde342789069492e9e8dcfedf2 100644 (file)
--- a/services/crunch/crunchstat/src/arvados.org/crunchstat/crunchstat.go
+++ b/services/crunch/crunchstat/src/arvados.org/crunchstat/crunchstat.go
@@ -42,19 +42,28 @@ func OutputChannel(stdout chan string, stderr chan string) {
        }
 }
 
-func FindStat(cgroup_path string, statgroup string, stat string) string {
-       path := fmt.Sprintf("%s/%s.%s", cgroup_path, statgroup, stat)
-       if _, err := os.Stat(path); err != nil {
+func FindStat(cgroup_root string, cgroup_parent string, container_id string, statgroup string, stat string) string {
+       var path string
+       path = fmt.Sprintf("%s/%s/%s/%s/%s.%s", cgroup_root, statgroup, cgroup_parent, container_id, statgroup, stat)
+       if _, err := os.Stat(path); err == nil {
                return path
        }
-       path = fmt.Sprintf("%s/%s/%s.%s", cgroup_path, statgroup, statgroup, stat)
-       if _, err := os.Stat(path); err != nil {
+       path = fmt.Sprintf("%s/%s/%s/%s.%s", cgroup_root, cgroup_parent, container_id, statgroup, stat)
+       if _, err := os.Stat(path); err == nil {
+               return path
+       }
+       path = fmt.Sprintf("%s/%s/%s.%s", cgroup_root, statgroup, statgroup, stat)
+       if _, err := os.Stat(path); err == nil {
+               return path
+       }
+       path = fmt.Sprintf("%s/%s.%s", cgroup_root, statgroup, stat)
+       if _, err := os.Stat(path); err == nil {
                return path
        }
        return ""
 }
 
-func PollCgroupStats(cgroup_path string, stderr chan string, poll int64) {
+func PollCgroupStats(cgroup_root string, cgroup_parent string, container_id string, stderr chan string, poll int64) {
        //var last_usage int64 = 0
        var last_user int64 = 0
        var last_sys int64 = 0
@@ -70,10 +79,23 @@ func PollCgroupStats(cgroup_path string, stderr chan string, poll int64) {
        disk := make(map[string]*Disk)
 
        //cpuacct_usage := FindStat(cgroup_path, "cpuacct", "usage")
-       cpuacct_stat := FindStat(cgroup_path, "cpuacct", "stat")
-       blkio_io_service_bytes := FindStat(cgroup_path, "blkio", "io_service_bytes")
-       cpuset_cpus := FindStat(cgroup_path, "cpuset", "cpus")
-       memory_stat := FindStat(cgroup_path, "memory", "stat")
+       cpuacct_stat := FindStat(cgroup_root, cgroup_parent, container_id, "cpuacct", "stat")
+       blkio_io_service_bytes := FindStat(cgroup_root, cgroup_parent, container_id, "blkio", "io_service_bytes")
+       cpuset_cpus := FindStat(cgroup_root, cgroup_parent, container_id, "cpuset", "cpus")
+       memory_stat := FindStat(cgroup_root, cgroup_parent, container_id, "memory", "stat")
+
+       if cpuacct_stat != "" {
+               stderr <- fmt.Sprintf("crunchstat: reading stats from %s", cpuacct_stat)
+       }
+       if blkio_io_service_bytes != "" {
+               stderr <- fmt.Sprintf("crunchstat: reading stats from %s", blkio_io_service_bytes)
+       }
+       if cpuset_cpus != "" {
+               stderr <- fmt.Sprintf("crunchstat: reading stats from %s", cpuset_cpus)
+       }
+       if memory_stat != "" {
+               stderr <- fmt.Sprintf("crunchstat: reading stats from %s", memory_stat)
+       }
 
        var elapsed int64 = poll
 
@@ -160,14 +182,14 @@ func PollCgroupStats(cgroup_path string, stderr chan string, poll int64) {
                                        if op == "Read" {
                                                disk[device].last_read = disk[device].next_read
                                                disk[device].next_read = next
-                                               if disk[device].last_read > 0 {
+                                               if disk[device].last_read > 0 && (disk[device].next_read != disk[device].last_read) {
                                                        stderr <- fmt.Sprintf("crunchstat: blkio.io_service_bytes %s read %v", device, disk[device].next_read-disk[device].last_read)
                                                }
                                        }
                                        if op == "Write" {
                                                disk[device].last_write = disk[device].next_write
                                                disk[device].next_write = next
-                                               if disk[device].last_write > 0 {
+                                               if disk[device].last_write > 0 && (disk[device].next_write != disk[device].last_write) {
                                                        stderr <- fmt.Sprintf("crunchstat: blkio.io_service_bytes %s write %v", device, disk[device].next_write-disk[device].last_write)
                                                }
                                        }
@@ -201,15 +223,15 @@ func PollCgroupStats(cgroup_path string, stderr chan string, poll int64) {
 func main() {
 
        var (
-               cgroup_path    string
+               cgroup_root    string
                cgroup_parent  string
                cgroup_cidfile string
                wait           int64
                poll           int64
        )
 
-       flag.StringVar(&cgroup_path, "cgroup-path", "", "Direct path to cgroup")
-       flag.StringVar(&cgroup_parent, "cgroup-parent", "", "Path to parent cgroup")
+       flag.StringVar(&cgroup_root, "cgroup-root", "", "Root of cgroup tree")
+       flag.StringVar(&cgroup_parent, "cgroup-parent", "", "Name of container parent under cgroup")
        flag.StringVar(&cgroup_cidfile, "cgroup-cid", "", "Path to container id file")
        flag.Int64Var(&wait, "wait", 5, "Maximum time (in seconds) to wait for cid file to show up")
        flag.Int64Var(&poll, "poll", 1000, "Polling frequency, in milliseconds")
@@ -218,8 +240,8 @@ func main() {
 
        logger := log.New(os.Stderr, "crunchstat: ", 0)
 
-       if cgroup_path == "" && cgroup_cidfile == "" {
-               logger.Fatal("Must provide either -cgroup-path or -cgroup-cid")
+       if cgroup_root == "" {
+               logger.Fatal("Must provide either -cgroup-root")
        }
 
        // Make output channel
@@ -272,6 +294,7 @@ func main() {
        }
 
        // Read the cid file
+       var container_id string
        if cgroup_cidfile != "" {
                // wait up to 'wait' seconds for the cid file to appear
                var i time.Duration
@@ -280,26 +303,19 @@ func main() {
                        if err == nil {
                                cid, err2 := ioutil.ReadAll(f)
                                if err2 == nil && len(cid) > 0 {
-                                       cgroup_path = string(cid)
+                                       container_id = string(cid)
                                        f.Close()
                                        break
                                }
                        }
                        time.Sleep(100 * time.Millisecond)
                }
-               if cgroup_path == "" {
+               if cgroup_root == "" {
                        logger.Printf("Could not read cid file %s", cgroup_cidfile)
                }
        }
 
-       // add the parent prefix
-       if cgroup_parent != "" {
-               cgroup_path = fmt.Sprintf("%s/%s", cgroup_parent, cgroup_path)
-       }
-
-       logger.Print("Using cgroup ", cgroup_path)
-
-       go PollCgroupStats(cgroup_path, stderr_chan, poll)
+       go PollCgroupStats(cgroup_root, cgroup_parent, container_id, stderr_chan, poll)
 
        // Wait for each of stdout and stderr to drain
        <-finish_chan