Expose Device Stats in /client/stats API endpoint

client/client.go

@@ -274,22 +274,6 @@ func NewClient(cfg *config.Config, consulCatalog consul.CatalogAPI, consulServic
 		return nil, fmt.Errorf("failed to initialize ACL state: %v", err)
 	}
 
-	// Add the stats collector
-	statsCollector := stats.NewHostStatsCollector(c.logger, c.config.AllocDir)
-	c.hostStatsCollector = statsCollector
-
-	// Add the garbage collector
-	gcConfig := &GCConfig{
-		MaxAllocs:           cfg.GCMaxAllocs,
-		DiskUsageThreshold:  cfg.GCDiskUsageThreshold,
-		InodeUsageThreshold: cfg.GCInodeUsageThreshold,
-		Interval:            cfg.GCInterval,
-		ParallelDestroys:    cfg.GCParallelDestroys,
-		ReservedDiskMB:      cfg.Node.Reserved.DiskMB,
-	}
-	c.garbageCollector = NewAllocGarbageCollector(c.logger, statsCollector, c, gcConfig)
-	go c.garbageCollector.Run()
-
 	// Setup the node
 	if err := c.setupNode(); err != nil {
 		return nil, fmt.Errorf("node setup failed: %v", err)
@@ -324,6 +308,22 @@ func NewClient(cfg *config.Config, consulCatalog consul.CatalogAPI, consulServic
 	c.devicemanager = devicemanager.New(devConfig)
 	go c.devicemanager.Run()
 
+	// Add the stats collector
+	statsCollector := stats.NewHostStatsCollector(c.logger, c.config.AllocDir, c.devicemanager.AllStats)
+	c.hostStatsCollector = statsCollector
+
+	// Add the garbage collector
+	gcConfig := &GCConfig{
+		MaxAllocs:           cfg.GCMaxAllocs,
+		DiskUsageThreshold:  cfg.GCDiskUsageThreshold,
+		InodeUsageThreshold: cfg.GCInodeUsageThreshold,
+		Interval:            cfg.GCInterval,
+		ParallelDestroys:    cfg.GCParallelDestroys,
+		ReservedDiskMB:      cfg.Node.Reserved.DiskMB,
+	}
+	c.garbageCollector = NewAllocGarbageCollector(c.logger, statsCollector, c, gcConfig)
+	go c.garbageCollector.Run()
+
 	// Set the preconfigured list of static servers
 	c.configLock.RLock()
 	if len(c.configCopy.Servers) > 0 {

client/stats/host.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	hclog "github.com/hashicorp/go-hclog"
+	"github.com/hashicorp/nomad/plugins/device"
 	"github.com/shirou/gopsutil/cpu"
 	"github.com/shirou/gopsutil/disk"
 	"github.com/shirou/gopsutil/host"
@@ -20,6 +21,7 @@ type HostStats struct {
 	CPU              []*CPUStats
 	DiskStats        []*DiskStats
 	AllocDirStats    *DiskStats
+	DeviceStats      []*DeviceGroupStats
 	Uptime           uint64
 	Timestamp        int64
 	CPUTicksConsumed float64
@@ -53,6 +55,12 @@ type DiskStats struct {
 	InodesUsedPercent float64
 }
 
+// DeviceGroupStats represents stats related to device group
+type DeviceGroupStats = device.DeviceGroupStats
+
+// DeviceStatsCollector is used to retrieve all the latest statistics for all devices.
+type DeviceStatsCollector func() []*DeviceGroupStats
+
 // NodeStatsCollector is an interface which is used for the purposes of mocking
 // the HostStatsCollector in the tests
 type NodeStatsCollector interface {
@@ -62,11 +70,12 @@ type NodeStatsCollector interface {
 
 // HostStatsCollector collects host resource usage stats
 type HostStatsCollector struct {
-	numCores        int
-	statsCalculator map[string]*HostCpuStatsCalculator
-	hostStats       *HostStats
-	hostStatsLock   sync.RWMutex
-	allocDir        string
+	numCores             int
+	statsCalculator      map[string]*HostCpuStatsCalculator
+	hostStats            *HostStats
+	hostStatsLock        sync.RWMutex
+	allocDir             string
+	deviceStatsCollector DeviceStatsCollector
 
 	// badParts is a set of partitions whose usage cannot be read; used to
 	// squelch logspam.
@@ -78,16 +87,17 @@ type HostStatsCollector struct {
 // NewHostStatsCollector returns a HostStatsCollector. The allocDir is passed in
 // so that we can present the disk related statistics for the mountpoint where
 // the allocation directory lives
-func NewHostStatsCollector(logger hclog.Logger, allocDir string) *HostStatsCollector {
+func NewHostStatsCollector(logger hclog.Logger, allocDir string, deviceStatsCollector DeviceStatsCollector) *HostStatsCollector {
 	logger = logger.Named("host_stats")
 	numCores := runtime.NumCPU()
 	statsCalculator := make(map[string]*HostCpuStatsCalculator)
 	collector := &HostStatsCollector{
-		statsCalculator: statsCalculator,
-		numCores:        numCores,
-		logger:          logger,
-		allocDir:        allocDir,
-		badParts:        make(map[string]struct{}),
+		statsCalculator:      statsCalculator,
+		numCores:             numCores,
+		logger:               logger,
+		allocDir:             allocDir,
+		badParts:             make(map[string]struct{}),
+		deviceStatsCollector: deviceStatsCollector,
 	}
 	return collector
 }
@@ -140,6 +150,10 @@ func (h *HostStatsCollector) collectLocked() error {
 	}
 	hs.AllocDirStats = h.toDiskStats(usage, nil)
 
+	// Collect devices stats
+	deviceStats := h.collectDeviceGroupStats()
+	hs.DeviceStats = deviceStats
+
 	// Update the collected status object.
 	h.hostStats = hs
 
@@ -189,6 +203,10 @@ func (h *HostStatsCollector) collectDiskStats() ([]*DiskStats, error) {
 	return diskStats, nil
 }
 
+func (h *HostStatsCollector) collectDeviceGroupStats() []*DeviceGroupStats {
+	return h.deviceStatsCollector()
+}
+
 // Stats returns the host stats that has been collected
 func (h *HostStatsCollector) Stats() *HostStats {
 	h.hostStatsLock.RLock()

plugins/shared/structs/stats.go

@@ -16,23 +16,23 @@ type StatObject struct {
 type StatValue struct {
 	// FloatNumeratorVal exposes a floating point value. If denominator is set
 	// it is assumed to be a fractional value, otherwise it is a scalar.
-	FloatNumeratorVal   *float64
-	FloatDenominatorVal *float64
+	FloatNumeratorVal   *float64 `json:",omitempty"`
+	FloatDenominatorVal *float64 `json:",omitempty"`
 
 	// IntNumeratorVal exposes a int value. If denominator is set it is assumed
 	// to be a fractional value, otherwise it is a scalar.
-	IntNumeratorVal   *int64
-	IntDenominatorVal *int64
+	IntNumeratorVal   *int64 `json:",omitempty"`
+	IntDenominatorVal *int64 `json:",omitempty"`
 
 	// StringVal exposes a string value. These are likely annotations.
-	StringVal *string
+	StringVal *string `json:",omitempty"`
 
 	// BoolVal exposes a boolean statistic.
-	BoolVal *bool
+	BoolVal *bool `json:",omitempty"`
 
 	// Unit gives the unit type: °F, %, MHz, MB, etc.
-	Unit string
+	Unit string `json:",omitempty"`
 
 	// Desc provides a human readable description of the statistic.
-	Desc string
+	Desc string `json:",omitempty"`
 }