11309: Prefix errors with "notice:" or "warning:" to aid parsing.

[arvados.git] / lib / crunchstat / crunchstat.go

diff --git a/lib/crunchstat/crunchstat.go b/lib/crunchstat/crunchstat.go
index e0ef058835ad6652009e5b2dbb40dbbf7a63da58..056ef0d185e61c7bbc52b692abd21ea61d9afdd4 100644 (file)
--- a/lib/crunchstat/crunchstat.go
+++ b/lib/crunchstat/crunchstat.go
@@ -1,3 +1,7 @@
+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
 // Package crunchstat reports resource usage (CPU, memory, disk,
 // network) for a cgroup.
 package crunchstat
@@ -40,7 +44,7 @@ type Reporter struct {
 
        // Where cgroup accounting files live on this system, e.g.,
        // "/sys/fs/cgroup".
-       CgroupRoot   string
+       CgroupRoot string
 
        // Parent cgroup, e.g., "docker".
        CgroupParent string
@@ -56,7 +60,8 @@ type Reporter struct {
        lastDiskSample   map[string]ioSample
        lastCPUSample    cpuSample
 
-       done chan struct{}
+       done    chan struct{} // closed when we should stop reporting
+       flushed chan struct{} // closed when we have made our last report
 }
 
 // Start starts monitoring in a new goroutine, and returns
@@ -72,6 +77,7 @@ type Reporter struct {
 // Callers should not modify public data fields after calling Start.
 func (r *Reporter) Start() {
        r.done = make(chan struct{})
+       r.flushed = make(chan struct{})
        go r.run()
 }
 
@@ -81,12 +87,13 @@ func (r *Reporter) Start() {
 // Nothing will be logged after Stop returns.
 func (r *Reporter) Stop() {
        close(r.done)
+       <-r.flushed
 }
 
 func (r *Reporter) readAllOrWarn(in io.Reader) ([]byte, error) {
        content, err := ioutil.ReadAll(in)
        if err != nil {
-               r.Logger.Print(err)
+               r.Logger.Printf("warning: %v", err)
        }
        return content, err
 }
@@ -162,7 +169,7 @@ func (r *Reporter) getContainerNetStats() (io.Reader, error) {
                statsFilename := fmt.Sprintf("/proc/%s/net/dev", taskPid)
                stats, err := ioutil.ReadFile(statsFilename)
                if err != nil {
-                       r.Logger.Print(err)
+                       r.Logger.Printf("notice: %v", err)
                        continue
                }
                return strings.NewReader(string(stats)), nil
@@ -312,6 +319,9 @@ func (r *Reporter) getCPUCount() int64 {
        }
        defer cpusetFile.Close()
        b, err := r.readAllOrWarn(cpusetFile)
+       if err != nil {
+               return 0
+       }
        sp := strings.Split(string(b), ",")
        cpus := int64(0)
        for _, v := range sp {
@@ -337,12 +347,16 @@ func (r *Reporter) doCPUStats() {
                return
        }
 
-       nextSample := cpuSample{true, time.Now(), 0, 0, r.getCPUCount()}
        var userTicks, sysTicks int64
        fmt.Sscanf(string(b), "user %d\nsystem %d", &userTicks, &sysTicks)
        userHz := float64(C.sysconf(C._SC_CLK_TCK))
-       nextSample.user = float64(userTicks) / userHz
-       nextSample.sys = float64(sysTicks) / userHz
+       nextSample := cpuSample{
+               hasData:    true,
+               sampleTime: time.Now(),
+               user:       float64(userTicks) / userHz,
+               sys:        float64(sysTicks) / userHz,
+               cpus:       r.getCPUCount(),
+       }
 
        delta := ""
        if r.lastCPUSample.hasData {
@@ -356,9 +370,11 @@ func (r *Reporter) doCPUStats() {
        r.lastCPUSample = nextSample
 }
 
-// Report stats periodically until r.done indicates someone called
-// Stop.
+// Report stats periodically until we learn (via r.done) that someone
+// called Stop.
 func (r *Reporter) run() {
+       defer close(r.flushed)
+
        r.reportedStatFile = make(map[string]string)
 
        if !r.waitForCIDFile() || !r.waitForCgroup() {
@@ -383,7 +399,7 @@ func (r *Reporter) run() {
 }
 
 // If CID is empty, wait for it to appear in CIDFile. Return true if
-// we get it before r.done indicates someone called Stop.
+// we get it before we learn (via r.done) that someone called Stop.
 func (r *Reporter) waitForCIDFile() bool {
        if r.CID != "" || r.CIDFile == "" {
                return true
@@ -400,7 +416,7 @@ func (r *Reporter) waitForCIDFile() bool {
                select {
                case <-ticker.C:
                case <-r.done:
-                       r.Logger.Printf("CID never appeared in %+q: %v", r.CIDFile, err)
+                       r.Logger.Printf("warning: CID never appeared in %+q: %v", r.CIDFile, err)
                        return false
                }
        }
@@ -423,9 +439,9 @@ func (r *Reporter) waitForCgroup() bool {
                select {
                case <-ticker.C:
                case <-warningTimer:
-                       r.Logger.Printf("cgroup stats files have not appeared after %v (config error?) -- still waiting...", r.PollPeriod)
+                       r.Logger.Printf("warning: cgroup stats files have not appeared after %v (config error?) -- still waiting...", r.PollPeriod)
                case <-r.done:
-                       r.Logger.Printf("cgroup stats files never appeared for %v", r.CID)
+                       r.Logger.Printf("warning: cgroup stats files never appeared for %v", r.CID)
                        return false
                }
        }