Moving Nvidia interfaces to stats package so that they can be used outside of the accelerators package

Signed-off-by: Maciej "Iwan" Iwanowski <maciej.iwanowski@intel.com>

accelerators/nvidia.go

@@ -16,6 +16,7 @@ package accelerators
 import (
 	"bufio"
 	"fmt"
+	"github.com/google/cadvisor/stats"
 	"io/ioutil"
 	"os"
 	"path/filepath"
@@ -30,7 +31,7 @@ import (
 	"k8s.io/klog"
 )
 
-type NvidiaManager struct {
+type nvidiaManager struct {
 	sync.Mutex
 
 	// true if there are NVIDIA devices present on the node
@@ -47,8 +48,12 @@ var sysFsPCIDevicesPath = "/sys/bus/pci/devices/"
 
 const nvidiaVendorId = "0x10de"
 
+func NewNvidiaManager() stats.Manager {
+	return &nvidiaManager{}
+}
+
 // Setup initializes NVML if nvidia devices are present on the node.
-func (nm *NvidiaManager) Setup() {
+func (nm *nvidiaManager) Setup() {
 	if !detectDevices(nvidiaVendorId) {
 		klog.V(4).Info("No NVIDIA devices found.")
 		return
@@ -84,7 +89,7 @@ func detectDevices(vendorId string) bool {
 
 // initializeNVML initializes the NVML library and sets up the nvmlDevices map.
 // This is defined as a variable to help in testing.
-var initializeNVML = func(nm *NvidiaManager) {
+var initializeNVML = func(nm *nvidiaManager) {
 	if err := gonvml.Initialize(); err != nil {
 		// This is under a logging level because otherwise we may cause
 		// log spam if the drivers/nvml is not installed on the system.
@@ -115,7 +120,7 @@ var initializeNVML = func(nm *NvidiaManager) {
 }
 
 // Destroy shuts down NVML.
-func (nm *NvidiaManager) Destroy() {
+func (nm *nvidiaManager) Destroy() {
 	if nm.nvmlInitialized {
 		gonvml.Shutdown()
 	}
@@ -123,7 +128,7 @@ func (nm *NvidiaManager) Destroy() {
 
 // GetCollector returns a collector that can fetch nvidia gpu metrics for nvidia devices
 // present in the devices.list file in the given devicesCgroupPath.
-func (nm *NvidiaManager) GetCollector(devicesCgroupPath string) (AcceleratorCollector, error) {
+func (nm *nvidiaManager) GetCollector(devicesCgroupPath string) (stats.Collector, error) {
 	nc := &NvidiaCollector{}
 
 	if !nm.devicesPresent {

accelerators/nvidia_test.go

@@ -72,13 +72,13 @@ func TestGetCollector(t *testing.T) {
 	}
 	parseDevicesCgroup = mockParser
 	originalInitializeNVML := initializeNVML
-	initializeNVML = func(_ *NvidiaManager) {}
+	initializeNVML = func(_ *nvidiaManager) {}
 	defer func() {
 		parseDevicesCgroup = originalParser
 		initializeNVML = originalInitializeNVML
 	}()
 
-	nm := &NvidiaManager{}
+	nm := &nvidiaManager{}
 
 	// When devicesPresent is false, empty collector should be returned.
 	ac, err := nm.GetCollector("does-not-matter")

manager/container.go

@@ -17,6 +17,7 @@ package manager
 import (
 	"flag"
 	"fmt"
+	"github.com/google/cadvisor/stats"
 	"io/ioutil"
 	"math"
 	"math/rand"
@@ -29,7 +30,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/google/cadvisor/accelerators"
 	"github.com/google/cadvisor/cache/memory"
 	"github.com/google/cadvisor/collector"
 	"github.com/google/cadvisor/container"
@@ -90,7 +90,7 @@ type containerData struct {
 	collectorManager collector.CollectorManager
 
 	// nvidiaCollector updates stats for Nvidia GPUs attached to the container.
-	nvidiaCollector accelerators.AcceleratorCollector
+	nvidiaCollector stats.Collector
 }
 
 // jitter returns a time.Duration between duration and duration + maxFactor * duration,

manager/manager.go

@@ -18,6 +18,7 @@ package manager
 import (
 	"flag"
 	"fmt"
+	"github.com/google/cadvisor/stats"
 	"net/http"
 	"os"
 	"path"
@@ -181,7 +182,7 @@ func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, maxHousekeepingIn
 		containerWatchers:                     []watcher.ContainerWatcher{},
 		eventsChannel:                         eventsChannel,
 		collectorHttpClient:                   collectorHttpClient,
-		nvidiaManager:                         &accelerators.NvidiaManager{},
+		nvidiaManager:                         accelerators.NewNvidiaManager(),
 		rawContainerCgroupPathPrefixWhiteList: rawContainerCgroupPathPrefixWhiteList,
 	}
 
@@ -230,7 +231,7 @@ type manager struct {
 	containerWatchers        []watcher.ContainerWatcher
 	eventsChannel            chan watcher.ContainerEvent
 	collectorHttpClient      *http.Client
-	nvidiaManager            accelerators.AcceleratorManager
+	nvidiaManager            stats.Manager
 	// List of raw container cgroup path prefix whitelist.
 	rawContainerCgroupPathPrefixWhiteList []string
 }

stats/types.go

@@ -1,4 +1,4 @@
-// Copyright 2017 Google Inc. All Rights Reserved.
+// Copyright 2020 Google Inc. All Rights Reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -11,22 +11,25 @@
 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 // See the License for the specific language governing permissions and
 // limitations under the License.
-package accelerators
+
+// Handling statistics that are fully controlled in cAdvisor
+package stats
 
 import info "github.com/google/cadvisor/info/v1"
 
-// This is supposed to store global state about an accelerator metrics collector.
-// cadvisor manager will call Setup() when it starts and Destroy() when it stops.
-// For each container detected by the cadvisor manager, it will call
+// This is supposed to store global state about an cAdvisor metrics collector.
+// cAdvisor manager will call Setup() when it starts and Destroy() when it stops.
+// For each container detected by the cAdvisor manager, it will call
 // GetCollector() with the devices cgroup path for that container.
 // GetCollector() is supposed to return an object that can update
 // accelerator stats for that container.
-type AcceleratorManager interface {
+type Manager interface {
 	Setup()
 	Destroy()
-	GetCollector(deviceCgroup string) (AcceleratorCollector, error)
+	GetCollector(deviceCgroup string) (Collector, error)
 }
 
-type AcceleratorCollector interface {
+// Collector can update ContainerStats by adding more metrics.
+type Collector interface {
 	UpdateStats(*info.ContainerStats) error
 }