MemstatCollector: remove mem_shared from mem_used calculation. Fixes #226

collectors/likwidMetric.go

@@ -46,7 +46,6 @@ const (
 	LIKWID_LIB_DL_FLAGS   = dl.RTLD_LAZY | dl.RTLD_GLOBAL
 	LIKWID_DEF_ACCESSMODE = "direct"
 	LIKWID_DEF_LOCKFILE   = "/var/run/likwid.lock"
-	LIKWID_DEF_WARMUP_DEL = "0s"
 )
 
 type LikwidCollectorMetricConfig struct {
@@ -84,7 +83,6 @@ type LikwidCollectorConfig struct {
 	DaemonPath     string                          `json:"accessdaemon_path,omitempty"`
 	LibraryPath    string                          `json:"liblikwid_path,omitempty"`
 	LockfilePath   string                          `json:"lockfile_path,omitempty"`
-	WarmupDelay    string                          `json:"warmup_delay"`
 }
 
 type LikwidCollector struct {
@@ -106,7 +104,6 @@ type LikwidCollector struct {
 	likwidGroups  map[C.int]LikwidEventsetConfig
 	lock          sync.Mutex
 	measureThread thread.Thread
-	warmupDelay   time.Duration
 }
 
 type LikwidMetric struct {
@@ -209,7 +206,6 @@ func (m *LikwidCollector) Init(config json.RawMessage) error {
 	m.config.AccessMode = LIKWID_DEF_ACCESSMODE
 	m.config.LibraryPath = LIKWID_LIB_NAME
 	m.config.LockfilePath = LIKWID_DEF_LOCKFILE
-	m.config.WarmupDelay = LIKWID_DEF_WARMUP_DEL
 	if len(config) > 0 {
 		d := json.NewDecoder(bytes.NewReader(config))
 		d.DisallowUnknownFields()
@@ -246,17 +242,6 @@ func (m *LikwidCollector) Init(config json.RawMessage) error {
 		m.cpu2tid[c] = i
 	}
 
-	if len(m.config.WarmupDelay) > 0 {
-		t, err := time.ParseDuration(m.config.WarmupDelay)
-		if err != nil {
-			return fmt.Errorf("%s Init(): Cannot parse WarmupDelay: %w", m.name, err)
-		}
-		if t < 0 {
-			return fmt.Errorf("%s Init(): WarmupDelay must not be negative", m.name)
-		}
-		m.warmupDelay = t
-	}
-
 	m.likwidGroups = make(map[C.int]LikwidEventsetConfig)
 
 	// This is for the global metrics computation test
@@ -510,27 +495,6 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	if ret != 0 {
 		return true, fmt.Errorf("failed to start events '%s', error %d", evset.go_estr, ret)
 	}
-
-	// warmup measuring
-	if m.warmupDelay > 0 {
-		select {
-		case <-sigchan:
-			ret = -1
-		case e := <-watcher.Events:
-			if e.Op != fsnotify.Chmod {
-				ret = C.perfmon_readCounters()
-			}
-		default:
-			ret = C.perfmon_readCounters()
-		}
-		if ret != 0 {
-			return true, fmt.Errorf("failed to read events '%s', error %d", evset.go_estr, ret)
-		}
-
-		time.Sleep(m.warmupDelay)
-	}
-
-	// begin measuring
 	select {
 	case <-sigchan:
 		ret = -1
@@ -548,7 +512,7 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	// Wait
 	time.Sleep(interval)
 
-	// end measuring
+	// Read counters
 	select {
 	case <-sigchan:
 		ret = -1

collectors/likwidMetric.md

@@ -63,7 +63,6 @@ Additional options:
 - `accessdaemon_path`: Folder of the accessDaemon `likwid-accessD` (like `/usr/local/sbin`)
 - `liblikwid_path`: Location of `liblikwid.so` including file name like `/usr/local/lib/liblikwid.so`
 - `lockfile_path`: Location of LIKWID's lock file if multiple tools should access the hardware counters. Default `/var/run/likwid.lock`
-- `warmup_delay`: Run an additional measurement of the specified length before the actual measurement. This can be used as a workaround for CPU starvation (during high load) in the measurement thread. Default is disabled (i.e. `0s`).
 
 ### Available metric types
 

collectors/memstatMetric.go

@@ -84,7 +84,7 @@ func getStats(filename string) map[string]MemstatStats {
 		case 5:
 			v, err := strconv.ParseFloat(linefields[3], 64)
 			if err == nil {
-				cclog.ComponentDebug("getStats", strings.Trim(linefields[2], ":"), v, linefields[4])
+				cclog.ComponentDebug("MemstatCollector", strings.Trim(linefields[2], ":"), v, linefields[4])
 				stats[strings.Trim(linefields[2], ":")] = MemstatStats{
 					value: v,
 					unit:  linefields[4],
@@ -92,6 +92,9 @@ func getStats(filename string) map[string]MemstatStats {
 			}
 		}
 	}
+	if scanner.Err() != nil {
+		cclog.ComponentError("MemstatCollector", "Failed to get memory stats")
+	}
 	return stats
 }
 
@@ -247,12 +250,6 @@ func (m *MemstatCollector) Read(interval time.Duration, output chan lp.CCMessage
 								unit = cacheVal.unit
 							}
 						}
-						if shmemVal, shmem := stats["Shmem"]; shmem {
-							memUsed -= shmemVal.value
-							if len(shmemVal.unit) > 0 && len(unit) == 0 {
-								unit = shmemVal.unit
-							}
-						}
 					}
 				}
 			}