Refactor the code to remove duplications (#1089)

* collector: refator metric package to stats Signed-off-by: Marcelo Amaral <[email protected]> * collector: move prometheus collector to metrics package Signed-off-by: Marcelo Amaral <[email protected]> * model: calculate processes energy and aggregate per container or vm Signed-off-by: Marcelo Amaral <[email protected]> * libvirt: find vm id from cgroup path Signed-off-by: Marcelo Amaral <[email protected]> * sensors: rename power package to sensors Signed-off-by: Marcelo Amaral <[email protected]> * libbpf: rebuild Signed-off-by: Marcelo Amaral <[email protected]> * bcc: fix cgroup id Signed-off-by: Marcelo Amaral <[email protected]> * e2e: wait kepler start Signed-off-by: Marcelo Amaral <[email protected]> * code: rebase Signed-off-by: Marcelo Amaral <[email protected]> --------- Signed-off-by: Marcelo Amaral <[email protected]>
sustainable-computing-io · Dec 19, 2023 · 7c99198 · 7c99198
1 parent 158305a
commit 7c99198
Show file tree

Hide file tree

Showing 133 changed files with 3,606 additions and 5,736 deletions.
diff --git a/.github/workflows/unit_test.yml b/.github/workflows/unit_test.yml
@@ -27,17 +27,26 @@ jobs:
       env:
           GOPATH: /home/runner/go
           GOBIN: /home/runner/go/bin
-    - name: install libbpf
-      uses: sustainable-computing-io/[email protected]
-      with:
-        ebpfprovider: libbpf
     - name: Prepare environment 
       run: |
-          sudo apt-get install -y cpuid
+          sudo apt-get install -y cpuid clang
           cd doc/ && sudo ./dev/prepare_dev_env.sh && cd - 
           git config --global --add safe.directory /kepler
+    - name: install libbpf
+      uses: sustainable-computing-io/[email protected]
+      with:
+        ebpfprovider: libbpf
     - name: Run 
       run: |
+          sudo apt remove libbpf-dev
+          mkdir temp-libbpf
+          cd temp-libbpf
+          git clone https://github.com/libbpf/libbpf
+          cd libbpf/src
+          sudo make install_headers
+          sudo make install_uapi_headers
+          sudo prefix=/usr  BUILD_STATIC_ONLY=y make install
+          cd ../../../
           ATTACHER_TAG=libbpf make test-verbose
           go tool cover -func=coverage.out -o=coverage.out
     - name: Go Coverage Badge  # Pass the `coverage.out` output to this action

diff --git a/.golangci.yml b/.golangci.yml
@@ -94,4 +94,4 @@ issues:
       text: "could not import C"
     - linters:
         - dupl
-      text: ".*pkg/collector/metric/types/types_test.go.*" # false positive, https://github.com/mibk/dupl/issues/20
+      text: ".*pkg/collector/stats/types/types_test.go.*" # false positive, https://github.com/mibk/dupl/issues/20
diff --git a/Makefile b/Makefile
@@ -9,6 +9,7 @@ export TIMESTAMP ?=$(shell echo $(BIN_TIMESTAMP) | tr -d ':' | tr 'T' '-' | tr -
 # restrict included verify-* targets to only process project files
 SOURCE_GIT_TAG     := $(shell git describe --tags --always --abbrev=7 --match 'v*')
 SRC_ROOT           := $(shell pwd)
+ARCH               := $(shell arch)
 OUTPUT_DIR         := _output
 CROSS_BUILD_BINDIR := $(OUTPUT_DIR)/bin
 GIT_VERSION        := $(shell git describe --dirty --tags --always --match='v*')

diff --git a/README.md b/README.md
@@ -1,6 +1,7 @@
 <img align="right" width="250px" src="https://user-images.githubusercontent.com/17484350/138557170-d8079b94-a517-4366-ade8-8d473e3f3f1d.jpg">
 
 ![GitHub Workflow Status (event)](https://img.shields.io/github/actions/workflow/status/sustainable-computing-io/kepler/unit_test.yml?branch=main&label=CI)
+
 ![Coverage](https://img.shields.io/badge/Coverage-42.1%25-yellow)
 [![OpenSSF Best Practices](https://bestpractices.coreinfrastructure.org/projects/7391/badge)](https://bestpractices.coreinfrastructure.org/projects/7391)
 <!--

diff --git a/bpfassets/bcc/bcc.c b/bpfassets/bcc/bcc.c
@@ -215,7 +215,7 @@ int kprobe__finish_task_switch(struct pt_regs *ctx, struct task_struct *prev)
 #ifdef SET_GROUP_ID
     u64 cgroup_id = bpf_get_current_cgroup_id();
 #else
-    u64 cgroup_id = 0;
+    u64 cgroup_id = task->cgroups->subsys[0]->cgroup->id;
 #endif
 
     u64 cur_ts = bpf_ktime_get_ns();

diff --git a/bpfassets/libbpf/bpf.o/amd64_kepler.bpf.o b/bpfassets/libbpf/bpf.o/amd64_kepler.bpf.o
diff --git a/bpfassets/libbpf/src/kepler.bpf.c b/bpfassets/libbpf/src/kepler.bpf.c
@@ -225,6 +225,7 @@ int kepler_trace(struct sched_switch_args *ctx)
 
     u32 cur_pid = bpf_get_current_pid_tgid();
     u64 cgroup_id = bpf_get_current_cgroup_id(); // the cgroup id is the cgroup id of the running process (this is not next_pid or prev_pid)
+
     u64 cur_ts = bpf_ktime_get_ns();
     u32 cpu_id = bpf_get_smp_processor_id();
     u32 prev_pid = ctx->prev_pid;

diff --git a/cmd/exporter/exporter.go b/cmd/exporter/exporter.go
@@ -29,18 +29,16 @@ import (
 	"time"
 
 	"github.com/sustainable-computing-io/kepler/pkg/bpfassets/attacher"
-	collector_metric "github.com/sustainable-computing-io/kepler/pkg/collector/metric"
+	"github.com/sustainable-computing-io/kepler/pkg/collector/stats"
 	"github.com/sustainable-computing-io/kepler/pkg/config"
 	"github.com/sustainable-computing-io/kepler/pkg/manager"
-	"github.com/sustainable-computing-io/kepler/pkg/power/accelerator/gpu"
-	"github.com/sustainable-computing-io/kepler/pkg/power/accelerator/qat"
-	"github.com/sustainable-computing-io/kepler/pkg/power/components"
-	"github.com/sustainable-computing-io/kepler/pkg/power/platform"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/accelerator/gpu"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/accelerator/qat"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/components"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/platform"
 	kversion "github.com/sustainable-computing-io/kepler/pkg/version"
 
-	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
-	"github.com/prometheus/common/version"
 
 	"k8s.io/klog/v2"
 )
@@ -197,7 +195,7 @@ func main() {
 	components.InitPowerImpl()
 	platform.InitPowerImpl()
 
-	collector_metric.InitAvailableParamAndMetrics()
+	stats.InitAvailableParamAndMetrics()
 
 	if config.EnabledGPU {
 		klog.Infof("Initializing the GPU collector")
@@ -218,7 +216,7 @@ func main() {
 		}
 	}
 
-	if config.EnabledQAT {
+	if config.IsExposeQATMetricsEnabled() {
 		klog.Infof("Initializing the QAT collector")
 		err := qat.Init()
 		if err == nil {
@@ -229,9 +227,8 @@ func main() {
 	}
 
 	m := manager.New()
-	prometheus.MustRegister(version.NewCollector("kepler_exporter"))
-	prometheus.MustRegister(m.PrometheusCollector)
-	defer m.MetricCollector.Destroy()
+	reg := m.PrometheusCollector.RegisterMetrics()
+	defer m.StatsCollector.Destroy()
 	defer components.StopPower()
 
 	// starting a new gorotine to collect data and report metrics
@@ -242,7 +239,12 @@ func main() {
 	metricPathConfig := config.GetMetricPath(*metricsPath)
 	bindAddressConfig := config.GetBindAddress(*address)
 
-	http.Handle(metricPathConfig, promhttp.Handler())
+	http.Handle(metricPathConfig, promhttp.HandlerFor(
+		reg,
+		promhttp.HandlerOpts{
+			Registry: reg,
+		},
+	))
 	http.HandleFunc("/healthz", healthProbe)
 	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
 		_, err := w.Write([]byte(`<html>

diff --git a/cmd/validator/validator.go b/cmd/validator/validator.go
@@ -29,9 +29,9 @@ import (
 	"time"
 
 	"github.com/jaypipes/ghw"
-	"github.com/sustainable-computing-io/kepler/pkg/power/components"
-	"github.com/sustainable-computing-io/kepler/pkg/power/components/source"
-	"github.com/sustainable-computing-io/kepler/pkg/power/platform"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/components"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/components/source"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/platform"
 )
 
 const (

diff --git a/e2e/integration-test/e2e_metric_test.go b/e2e/integration-test/e2e_metric_test.go
@@ -93,6 +93,7 @@ func getEnvOrDefault(envName, defaultValue string) string {
 }
 
 var _ = Describe("Metrics check should pass", Ordered, func() {
+
 	var keplerMetric *TestKeplerMetric
 
 	_ = BeforeAll(func() {
@@ -177,7 +178,6 @@ var _ = Describe("Metrics check should pass", Ordered, func() {
 		Entry(nil, "kepler_node_core_joules_total"),     // node level check by instance
 		Entry(nil, "kepler_node_dram_joules_total"),     // node level check by instance
 		Entry(nil, "kepler_node_info"),                  // node level missing labels
-		Entry(nil, "kepler_node_other_joules_total"),    // node level check by instance
 		Entry(nil, "kepler_node_package_joules_total"),  // node levelcheck by instance
 		Entry(nil, "kepler_node_platform_joules_total"), // node levelcheck by instance
 		Entry(nil, "kepler_node_uncore_joules_total"),   // node levelcheck by instance
@@ -214,7 +214,6 @@ var _ = Describe("Metrics check should pass", Ordered, func() {
 		Entry(nil, "kepler_container_core_joules_total"),    // pod level
 		Entry(nil, "kepler_container_dram_joules_total"),    // pod level
 		Entry(nil, "kepler_container_joules_total"),         // pod level
-		Entry(nil, "kepler_container_other_joules_total"),   // pod level
 		Entry(nil, "kepler_container_package_joules_total"), // pod level
 		Entry(nil, "kepler_container_uncore_joules_total"),  // pod level
 	)

diff --git a/hack/cluster-deploy.sh b/hack/cluster-deploy.sh
@@ -40,10 +40,7 @@ build_manifest() {
 		return 0
 	}
 	header "Build Kepler Manifest"
-	run make build-manifest \
-		OPTS="$OPTS" \
-		IMAGE_REPO="$IMAGE_REPO" \
-		IMAGE_TAG="$IMAGE_TAG"
+	run make build-manifest OPTS="$OPTS"
 }
 
 build_kepler() {

diff --git a/pkg/bpfassets/attacher/bpf_perf.go b/pkg/bpfassets/attacher/bpf_perf.go
@@ -82,8 +82,8 @@ func getCounters() map[string]perfCounter {
 func GetEnabledBPFHWCounters() []string {
 	Counters = getCounters()
 	var metrics []string
-	klog.V(5).Infof("hardeware counter metrics config %t", config.ExposeHardwareCounterMetrics)
-	if !config.ExposeHardwareCounterMetrics {
+	klog.V(5).Infof("hardeware counter metrics config %t", config.IsHCMetricsEnabled())
+	if !config.IsHCMetricsEnabled() {
 		klog.V(5).Info("hardeware counter metrics not enabled")
 		return metrics
 	}

diff --git a/pkg/bpfassets/attacher/libbpf_attacher_stub.go b/pkg/bpfassets/attacher/libbpf_attacher_stub.go
@@ -30,6 +30,9 @@ const (
 var libbpfCounters = map[string]perfCounter{}
 
 func attachLibbpfModule() (interface{}, error) {
+	if LibbpfBuilt {
+		return nil, nil
+	}
 	return nil, fmt.Errorf("no libbpf build tag")
 }
 

diff --git a/pkg/cgroup/cgroup_interface.go b/pkg/cgroup/cgroup_interface.go
@@ -16,7 +16,7 @@ limitations under the License.
 
 package cgroup
 
-import "github.com/sustainable-computing-io/kepler/pkg/collector/metric/types"
+import "github.com/sustainable-computing-io/kepler/pkg/collector/stats/types"
 
 type CCgroupStatHandler interface {
 	SetCGroupStat(containerID string, CgroupStatMap map[string]*types.UInt64StatCollection) error

diff --git a/pkg/cgroup/cgroup_stats_linux.go b/pkg/cgroup/cgroup_stats_linux.go
@@ -28,7 +28,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/cgroups/fs2"
 	"github.com/opencontainers/runc/libcontainer/configs"
 
-	"github.com/sustainable-computing-io/kepler/pkg/collector/metric/types"
+	"github.com/sustainable-computing-io/kepler/pkg/collector/stats/types"
 	"github.com/sustainable-computing-io/kepler/pkg/config"
 )
 

diff --git a/pkg/cgroup/resolve_container.go b/pkg/cgroup/resolve_container.go
@@ -76,23 +76,22 @@ func GetContainerID(cGroupID, pid uint64, withCGroupID bool) (string, error) {
 	return info.ContainerID, err
 }
 
-func GetContainerMetrics() (containerCPU, containerMem map[string]float64, retErr error) {
-	return podLister.ListMetrics()
-}
-
-func GetAvailableKubeletMetrics() []string {
-	return podLister.GetAvailableMetrics()
-}
-
 func GetContainerInfo(cGroupID, pid uint64, withCGroupID bool) (*ContainerInfo, error) {
 	var err error
 	var containerID string
 
+	name := utils.SystemProcessName
+	namespace := utils.SystemProcessNamespace
+	if cGroupID == 1 {
+		// some kernel processes have cgroup id equal 1 or 0
+		name = utils.KernelProcessName
+		namespace = utils.KernelProcessNamespace
+	}
 	info := &ContainerInfo{
-		ContainerID:   utils.SystemProcessName,
-		ContainerName: utils.SystemProcessName,
-		PodName:       utils.SystemProcessName,
-		Namespace:     utils.SystemProcessNamespace,
+		ContainerID:   name,
+		ContainerName: name,
+		PodName:       name,
+		Namespace:     namespace,
 	}
 
 	if containerID, err = getContainerIDFromPath(cGroupID, pid, withCGroupID); err != nil {
@@ -113,6 +112,9 @@ func ParseContainerIDFromPodStatus(containerID string) string {
 }
 
 func getContainerIDFromPath(cGroupID, pid uint64, withCGroupID bool) (string, error) {
+	if cGroupID == 1 {
+		return utils.KernelProcessName, nil
+	}
 	var err error
 	var containerID string
 	if withCGroupID {

diff --git a/pkg/collector/container_accelerator_collector.go b/pkg/collector/container_accelerator_collector.go
diff --git a/pkg/collector/container_energy_collector.go b/pkg/collector/container_energy_collector.go