Acquiring locks before iterating allocations and tasks

client/alloc_runner.go

@@ -486,6 +486,8 @@ func (r *AllocRunner) StatsReporter() AllocStatsReporter {
 // AllocStats returns the stats reporter of all the tasks running in the
 // allocation
 func (r *AllocRunner) AllocStats() map[string]TaskStatsReporter {
+	r.taskLock.Lock()
+	defer r.taskLock.RUnlock()
 	res := make(map[string]TaskStatsReporter)
 	for task, tr := range r.tasks {
 		res[task] = tr.StatsReporter()
@@ -498,7 +500,7 @@ func (r *AllocRunner) AllocStats() map[string]TaskStatsReporter {
 func (r *AllocRunner) TaskStats(task string) (TaskStatsReporter, error) {
 	tr, ok := r.tasks[task]
 	if !ok {
-		return nil, fmt.Errorf("task %q not running", task)
+		return nil, fmt.Errorf("task %q not running in allocation %v", task, r.alloc.ID)
 	}
 
 	return tr.StatsReporter(), nil

client/client.go

@@ -72,6 +72,10 @@ const (
 	// consulSyncInterval is the interval at which the client syncs with consul
 	// to remove services and checks which are no longer valid
 	consulSyncInterval = 15 * time.Second
+
+	// hostStatsCollectorIntv is the interval on which we collect host resource
+	// usage stats
+	hostStatsCollectorIntv = 1 * time.Second
 )
 
 // DefaultConfig returns the default configuration
@@ -213,7 +217,7 @@ func NewClient(cfg *config.Config) (*Client, error) {
 	go c.run()
 
 	// Start collecting stats
-	go c.monitorUsage()
+	go c.monitorHostStats()
 
 	// Start the consul sync
 	go c.syncConsul()
@@ -430,7 +434,8 @@ func (c *Client) StatsReporter() ClientStatsReporter {
 // Nomad client
 func (c *Client) AllocStats() map[string]AllocStatsReporter {
 	res := make(map[string]AllocStatsReporter)
-	for alloc, ar := range c.allocs {
+	allocRunners := c.getAllocRunners()
+	for alloc, ar := range allocRunners {
 		res[alloc] = ar
 	}
 	return res
@@ -1277,9 +1282,10 @@ func (c *Client) syncConsul() {
 	}
 }
 
-func (c *Client) monitorUsage() {
+// monitorHostStats collects host resource usage stats periodically
+func (c *Client) monitorHostStats() {
+	next := time.NewTimer(hostStatsCollectorIntv)
 	for {
-		next := time.NewTimer(1 * time.Second)
 		select {
 		case <-next.C:
 			ru, err := c.hostStatsCollector.Collect()