chore: rebase node agent with upstream

.github/workflows/ci.yml

@@ -13,7 +13,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions/setup-go@v5
         with:
-          go-version: '1.21'
+          go-version: '1.23'
       - run: sudo apt install -y libsystemd-dev
       - name: gofmt -l .
         run: files=$(gofmt -l .); if [[ -n "$files" ]]; then echo "$files"; exit 1; fi

CONTRIBUTING.md

@@ -6,7 +6,7 @@ Below are some basic guidelines.
 
 ## Requirements
 * Linux ≥v4.16 (amd64, arm64)
-* Go v1.21
+* Go v1.23
 
 
 ## Running

Dockerfile

@@ -1,4 +1,4 @@
-FROM golang:1.21-bullseye AS builder
+FROM golang:1.23-bullseye AS builder
 RUN apt update && apt install -y libsystemd-dev
 WORKDIR /tmp/src
 COPY go.mod .
@@ -8,7 +8,9 @@ COPY . .
 ARG VERSION=unknown
 RUN CGO_ENABLED=1 go build -mod=readonly -ldflags "-X main.version=$VERSION" -o coroot-node-agent .
 
-FROM debian:bullseye-slim
-RUN apt update && apt install -y ca-certificates && apt clean
+FROM registry.access.redhat.com/ubi9/ubi
+
+ARG VERSION=unknown
+
 COPY --from=builder /tmp/src/coroot-node-agent /usr/bin/coroot-node-agent
 ENTRYPOINT ["coroot-node-agent"]

Dockerfile.alpine

@@ -1,4 +1,4 @@
-FROM --platform=linux/amd64 golang:1.22-alpine as build-stage
+FROM --platform=linux/amd64 golang:1.23-alpine as build-stage
 # Set destination for COPY
 WORKDIR /app
 RUN apk add --no-cache --update go gcc g++ bash git

cgroup/cgroup.go

@@ -23,6 +23,7 @@ var (
 	containerdIdRegexp  = regexp.MustCompile(`cri-containerd[-:]([a-z0-9]{64})`)
 	lxcIdRegexp         = regexp.MustCompile(`/lxc/([^/]+)`)
 	systemSliceIdRegexp = regexp.MustCompile(`(/(system|runtime)\.slice/([^/]+))`)
+	talosIdRegexp       = regexp.MustCompile(`/(system|podruntime)/([^/]+)`)
 )
 
 type Version uint8
@@ -43,6 +44,7 @@ const (
 	ContainerTypeLxc
 	ContainerTypeSystemdService
 	ContainerTypeSandbox
+	ContainerTypeTalosRuntime
 )
 
 func (t ContainerType) String() string {
@@ -105,24 +107,25 @@ func NewFromProcessCgroupFile(filePath string) (*Cgroup, error) {
 			cg.subsystems[cgType] = path.Join(baseCgroupPath, parts[2])
 		}
 	}
-	if p := cg.subsystems["name=systemd"]; p != "" {
-		cg.Id = p
-		cg.Version = V1
-	} else if p = cg.subsystems["cpu"]; p != "" {
-		cg.Id = p
-		cg.Version = V1
-	} else {
-		cg.Id = cg.subsystems[""]
+	if cg.Id = cg.subsystems[""]; cg.Id != "" {
 		cg.Version = V2
+	} else if cg.Id = cg.subsystems["cpu"]; cg.Id != "" {
+		cg.Version = V1
+	}
+	if (cg.Id == "" || cg.Id == "/") && cg.subsystems["name=systemd"] != "/" {
+		cg.Id = cg.subsystems["name=systemd"]
 	}
 	if cg.ContainerType, cg.ContainerId, err = containerByCgroup(cg.Id); err != nil {
 		return nil, err
 	}
 	return cg, nil
 }
 
-func containerByCgroup(path string) (ContainerType, string, error) {
-	parts := strings.Split(strings.TrimLeft(path, "/"), "/")
+func containerByCgroup(cgroupPath string) (ContainerType, string, error) {
+	parts := strings.Split(strings.TrimLeft(cgroupPath, "/"), "/")
+	if cgroupPath == "/init" {
+		return ContainerTypeTalosRuntime, "/talos/init", nil
+	}
 	if len(parts) < 2 {
 		return ContainerTypeStandaloneProcess, "", nil
 	}
@@ -134,43 +137,50 @@ func containerByCgroup(path string) (ContainerType, string, error) {
 		return ContainerTypeStandaloneProcess, "", nil
 	}
 	if prefix == "docker" || (prefix == "system.slice" && strings.HasPrefix(parts[1], "docker-")) {
-		matches := dockerIdRegexp.FindStringSubmatch(path)
+		matches := dockerIdRegexp.FindStringSubmatch(cgroupPath)
 		if matches == nil {
-			return ContainerTypeUnknown, "", fmt.Errorf("invalid docker cgroup %s", path)
+			return ContainerTypeUnknown, "", fmt.Errorf("invalid docker cgroup %s", cgroupPath)
 		}
 		return ContainerTypeDocker, matches[1], nil
 	}
-	if strings.Contains(path, "kubepods") {
-		crioMatches := crioIdRegexp.FindStringSubmatch(path)
+	if strings.Contains(cgroupPath, "kubepods") {
+		crioMatches := crioIdRegexp.FindStringSubmatch(cgroupPath)
 		if crioMatches != nil {
 			return ContainerTypeCrio, crioMatches[1], nil
 		}
-		if strings.Contains(path, "crio-conmon-") {
+		if strings.Contains(cgroupPath, "crio-conmon-") {
 			return ContainerTypeUnknown, "", nil
 		}
-		containerdMatches := containerdIdRegexp.FindStringSubmatch(path)
+		containerdMatches := containerdIdRegexp.FindStringSubmatch(cgroupPath)
 		if containerdMatches != nil {
 			return ContainerTypeContainerd, containerdMatches[1], nil
 		}
-		matches := dockerIdRegexp.FindStringSubmatch(path)
+		matches := dockerIdRegexp.FindStringSubmatch(cgroupPath)
 		if matches == nil {
 			return ContainerTypeSandbox, "", nil
 		}
 		return ContainerTypeDocker, matches[1], nil
 	}
 	if prefix == "lxc" {
-		matches := lxcIdRegexp.FindStringSubmatch(path)
+		matches := lxcIdRegexp.FindStringSubmatch(cgroupPath)
 		if matches == nil {
-			return ContainerTypeUnknown, "", fmt.Errorf("invalid lxc cgroup %s", path)
+			return ContainerTypeUnknown, "", fmt.Errorf("invalid lxc cgroup %s", cgroupPath)
 		}
 		return ContainerTypeLxc, matches[1], nil
 	}
+	if prefix == "system" || prefix == "podruntime" {
+		matches := talosIdRegexp.FindStringSubmatch(cgroupPath)
+		if matches == nil {
+			return ContainerTypeUnknown, "", fmt.Errorf("invalid talos runtime cgroup %s", cgroupPath)
+		}
+		return ContainerTypeTalosRuntime, path.Join("/talos/", matches[2]), nil
+	}
 	if prefix == "system.slice" || prefix == "runtime.slice" {
-		matches := systemSliceIdRegexp.FindStringSubmatch(path)
+		matches := systemSliceIdRegexp.FindStringSubmatch(cgroupPath)
 		if matches == nil {
-			return ContainerTypeUnknown, "", fmt.Errorf("invalid systemd cgroup %s", path)
+			return ContainerTypeUnknown, "", fmt.Errorf("invalid systemd cgroup %s", cgroupPath)
 		}
 		return ContainerTypeSystemdService, strings.Replace(matches[1], "\\x2d", "-", -1), nil
 	}
-	return ContainerTypeUnknown, "", fmt.Errorf("unknown container: %s", path)
+	return ContainerTypeUnknown, "", fmt.Errorf("unknown container: %s", cgroupPath)
 }

cgroup/cgroup_test.go

@@ -62,6 +62,27 @@ func TestNewFromProcessCgroupFile(t *testing.T) {
 	assert.Equal(t, "/system.slice/springboot.service", cg.ContainerId)
 	assert.Equal(t, ContainerTypeSystemdService, cg.ContainerType)
 
+	cg, err = NewFromProcessCgroupFile(path.Join("fixtures/proc/700/cgroup"))
+	assert.Nil(t, err)
+	assert.Equal(t, V2, cg.Version)
+	assert.Equal(t, "/podruntime/runtime", cg.Id)
+	assert.Equal(t, "/talos/runtime", cg.ContainerId)
+	assert.Equal(t, ContainerTypeTalosRuntime, cg.ContainerType)
+
+	cg, err = NewFromProcessCgroupFile(path.Join("fixtures/proc/800/cgroup"))
+	assert.Nil(t, err)
+	assert.Equal(t, V2, cg.Version)
+	assert.Equal(t, "/system.slice/docker-cf87ba651579c9231db817909e7865e5747bd7abcac0c57ce23cf4abbaee046b.scope", cg.Id)
+	assert.Equal(t, "cf87ba651579c9231db817909e7865e5747bd7abcac0c57ce23cf4abbaee046b", cg.ContainerId)
+	assert.Equal(t, ContainerTypeDocker, cg.ContainerType)
+
+	cg, err = NewFromProcessCgroupFile(path.Join("fixtures/proc/900/cgroup"))
+	assert.Nil(t, err)
+	assert.Equal(t, V1, cg.Version)
+	assert.Equal(t, "/system.slice/python-app.service", cg.Id)
+	assert.Equal(t, "/system.slice/python-app.service", cg.ContainerId)
+	assert.Equal(t, ContainerTypeSystemdService, cg.ContainerType)
+
 	baseCgroupPath = "/kubepods.slice/kubepods-besteffort.slice/kubepods-besteffort-podc83d0428_58af_41eb_8dba_b9e6eddffe7b.slice/docker-0e612005fd07e7f47e2cd07df99a2b4e909446814d71d0b5e4efc7159dd51252.scope"
 	defer func() {
 		baseCgroupPath = ""
@@ -97,6 +118,11 @@ func TestContainerByCgroup(t *testing.T) {
 	as.Equal("63425c4a8b4291744a79dd9011fddc7a1f8ffda61f65d72196aa01d00cae2e2d", id)
 	as.Nil(err)
 
+	typ, id, err = containerByCgroup("/docker/63425c4a8b4291744a79dd9011fddc7a1f8ffda61f65d72196aa01d00cae2e2d")
+	as.Equal(typ, ContainerTypeDocker)
+	as.Equal("63425c4a8b4291744a79dd9011fddc7a1f8ffda61f65d72196aa01d00cae2e2d", id)
+	as.Nil(err)
+
 	typ, id, err = containerByCgroup("/lxc/mysql-primary-db")
 	as.Equal(typ, ContainerTypeLxc)
 	as.Equal("mysql-primary-db", id)
@@ -136,4 +162,19 @@ func TestContainerByCgroup(t *testing.T) {
 	as.Equal(typ, ContainerTypeContainerd)
 	as.Equal("d4a9f9195eaf7e4a729f24151101e1de61f1398677e7b82acfb936dff0b4ce55", id)
 	as.Nil(err)
+
+	typ, id, err = containerByCgroup("/podruntime/kubelet")
+	as.Equal(typ, ContainerTypeTalosRuntime)
+	as.Equal("/talos/kubelet", id)
+	as.Nil(err)
+
+	typ, id, err = containerByCgroup("/system/dashboard")
+	as.Equal(typ, ContainerTypeTalosRuntime)
+	as.Equal("/talos/dashboard", id)
+	as.Nil(err)
+
+	typ, id, err = containerByCgroup("/init")
+	as.Equal(typ, ContainerTypeTalosRuntime)
+	as.Equal("/talos/init", id)
+	as.Nil(err)
 }

cgroup/cpu.go

@@ -6,6 +6,8 @@ import (
 	"path"
 	"strconv"
 	"strings"
+
+	"github.com/coroot/coroot-node-agent/common"
 )
 
 type CPUStat struct {
@@ -14,27 +16,32 @@ type CPUStat struct {
 	LimitCores           float64
 }
 
-func (cg Cgroup) CpuStat() (*CPUStat, error) {
+func (cg Cgroup) CpuStat() *CPUStat {
 	if cg.Version == V1 {
-		return cg.cpuStatV1()
+		st, _ := cg.cpuStatV1()
+		return st
 	}
-	return cg.cpuStatV2()
+	st, _ := cg.cpuStatV2()
+	return st
 }
 
 func (cg Cgroup) cpuStatV1() (*CPUStat, error) {
+	if cg.subsystems["cpu"] == "/" || cg.subsystems["cpuacct"] == "/" {
+		return nil, nil
+	}
 	throttling, err := readVariablesFromFile(path.Join(cgRoot, "cpu", cg.subsystems["cpu"], "cpu.stat"))
 	if err != nil {
 		return nil, err
 	}
-	usageNs, err := readIntFromFile(path.Join(cgRoot, "cpuacct", cg.subsystems["cpuacct"], "cpuacct.usage"))
+	usageNs, err := common.ReadIntFromFile(path.Join(cgRoot, "cpuacct", cg.subsystems["cpuacct"], "cpuacct.usage"))
 	if err != nil {
 		return nil, err
 	}
-	periodUs, err := readIntFromFile(path.Join(cgRoot, "cpu", cg.subsystems["cpu"], "cpu.cfs_period_us"))
+	periodUs, err := common.ReadIntFromFile(path.Join(cgRoot, "cpu", cg.subsystems["cpu"], "cpu.cfs_period_us"))
 	if err != nil {
 		return nil, err
 	}
-	quotaUs, err := readIntFromFile(path.Join(cgRoot, "cpu", cg.subsystems["cpu"], "cpu.cfs_quota_us"))
+	quotaUs, err := common.ReadIntFromFile(path.Join(cgRoot, "cpu", cg.subsystems["cpu"], "cpu.cfs_quota_us"))
 	if err != nil {
 		return nil, err
 	}

cgroup/cpu_test.go

@@ -11,28 +11,29 @@ func TestCgroup_CpuStat(t *testing.T) {
 	cgRoot = "fixtures/cgroup"
 
 	cg, _ := NewFromProcessCgroupFile(path.Join("fixtures/proc/100/cgroup"))
-	s, err := cg.CpuStat()
-	assert.Nil(t, err)
+	s := cg.CpuStat()
 	assert.Equal(t, 0., s.LimitCores)
 	assert.Equal(t, 26778.913419246, s.UsageSeconds)
 
 	cg, _ = NewFromProcessCgroupFile(path.Join("fixtures/proc/200/cgroup"))
-	s, err = cg.CpuStat()
-	assert.Nil(t, err)
+	s = cg.CpuStat()
 	assert.Equal(t, 1.5, s.LimitCores)
 	assert.Equal(t, 254005.032764376, s.ThrottledTimeSeconds)
 
 	cg, _ = NewFromProcessCgroupFile(path.Join("fixtures/proc/400/cgroup"))
-	s, err = cg.CpuStat()
-	assert.Nil(t, err)
+	s = cg.CpuStat()
 	assert.Equal(t, 0.1, s.LimitCores)
 	assert.Equal(t, 0.363166, s.ThrottledTimeSeconds)
 	assert.Equal(t, 3795.681254, s.UsageSeconds)
 
 	cg, _ = NewFromProcessCgroupFile(path.Join("fixtures/proc/500/cgroup"))
-	s, err = cg.CpuStat()
-	assert.Nil(t, err)
+	s = cg.CpuStat()
 	assert.Equal(t, 0., s.LimitCores)
 	assert.Equal(t, 0., s.ThrottledTimeSeconds)
 	assert.Equal(t, 5531.521992, s.UsageSeconds)
+
+	cg, _ = NewFromProcessCgroupFile(path.Join("fixtures/proc/1000/cgroup"))
+	s, err := cg.cpuStatV1()
+	assert.NoError(t, err)
+	assert.Nil(t, s)
 }

cgroup/fixtures/proc/1000/cgroup

@@ -0,0 +1,11 @@
+11:pids:/
+10:blkio:/
+9:net_cls,net_prio:/
+8:perf_event:/
+7:freezer:/
+6:memory:/
+5:cpu,cpuacct:/
+4:hugetlb:/
+3:cpuset:/
+2:devices:/system.slice/system-getty.slice
+1:name=systemd:/system.slice/system-getty.slice/getty@tty1.service

cgroup/fixtures/proc/700/cgroup

@@ -0,0 +1 @@
+0::/podruntime/runtime

cgroup/fixtures/proc/800/cgroup

@@ -0,0 +1,2 @@
+1:name=systemd:/
+0::/system.slice/docker-cf87ba651579c9231db817909e7865e5747bd7abcac0c57ce23cf4abbaee046b.scope

cgroup/fixtures/proc/900/cgroup

@@ -0,0 +1,2 @@
+1:name=systemd:/system.slice/python-app.service
+7:cpu,cpuacct:/

cgroup/io.go

@@ -16,14 +16,19 @@ type IOStat struct {
 	WrittenBytes uint64
 }
 
-func (cg *Cgroup) IOStat() (map[string]IOStat, error) {
+func (cg *Cgroup) IOStat() map[string]IOStat {
 	if cg.Version == V1 {
-		return cg.ioStatV1()
+		st, _ := cg.ioStatV1()
+		return st
 	}
-	return cg.ioStatV2()
+	st, _ := cg.ioStatV2()
+	return st
 }
 
 func (cg *Cgroup) ioStatV1() (map[string]IOStat, error) {
+	if cg.subsystems["blkio"] == "/" {
+		return nil, nil
+	}
 	ops, err := readBlkioStatFile(path.Join(cgRoot, "blkio", cg.subsystems["blkio"], "blkio.throttle.io_serviced"))
 	if err != nil {
 		return nil, err

cgroup/io_test.go

@@ -11,8 +11,7 @@ func TestCgroup_IOStat(t *testing.T) {
 	cgRoot = "fixtures/cgroup"
 
 	cg, _ := NewFromProcessCgroupFile(path.Join("fixtures/proc/200/cgroup"))
-	stat, err := cg.IOStat()
-	assert.Nil(t, err)
+	stat := cg.IOStat()
 	assert.Equal(t,
 		map[string]IOStat{
 			"8:0":  {ReadOps: 0, WriteOps: 281, ReadBytes: 0, WrittenBytes: 4603904},
@@ -24,13 +23,16 @@ func TestCgroup_IOStat(t *testing.T) {
 		stat)
 
 	cg, _ = NewFromProcessCgroupFile(path.Join("fixtures/proc/400/cgroup"))
-	stat, err = cg.IOStat()
-	assert.Nil(t, err)
+	stat = cg.IOStat()
 	assert.Equal(t,
 		map[string]IOStat{
 			"252:0": {ReadOps: 22, WriteOps: 57111, ReadBytes: 11, WrittenBytes: 630538240},
 			"253:0": {ReadOps: 44, WriteOps: 57056, ReadBytes: 33, WrittenBytes: 630538241},
 		},
 		stat)
 
+	cg, _ = NewFromProcessCgroupFile(path.Join("fixtures/proc/1000/cgroup"))
+	stat, err := cg.ioStatV1()
+	assert.NoError(t, err)
+	assert.Nil(t, stat)
 }