feat: publish sanitized custom.97 source
This commit is contained in:
1022 files changed
+69903
-19207
No files matched your search
@@ -0,0 +1,174 @@
|
||||
package common
|
||||
|
||||
import (
|
||||
"os"
|
||||
"path/filepath"
|
||||
"runtime"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
// 容器自身资源采样。
|
||||
//
|
||||
// 历史实现直接读取宿主机口径(/proc/stat 与 /proc/meminfo),把同节点其它
|
||||
// 工作负载的占用也算成本进程过载,导致正常的 relay 流量被 503 熔断。这里
|
||||
// 改为优先读取当前容器的 cgroup 配额与用量,只有在拿不到 cgroup 数据时才由
|
||||
// 调用方回退到宿主机口径。
|
||||
|
||||
const (
|
||||
defaultCgroupRoot = "/sys/fs/cgroup"
|
||||
// cgroup v1 用接近 int64 上限的值表示"未限制"。
|
||||
unlimitedMemoryBytes = uint64(1) << 60
|
||||
)
|
||||
|
||||
// containerUsage 记录上一次采样点,用来计算区间内的 CPU 占用。
|
||||
type containerUsage struct {
|
||||
cgroupRoot string
|
||||
// schedulableCPUs 是本进程可调度的逻辑 CPU 数(Linux 上含 cpuset/affinity 限制)。
|
||||
schedulableCPUs float64
|
||||
cpuLastUsec uint64
|
||||
cpuLastSampled time.Time
|
||||
}
|
||||
|
||||
func newContainerUsage() *containerUsage {
|
||||
// runtime.NumCPU 在 Linux 上于进程启动时按 sched_getaffinity 计数,不受 GOMAXPROCS 影响。
|
||||
return &containerUsage{cgroupRoot: defaultCgroupRoot, schedulableCPUs: float64(runtime.NumCPU())}
|
||||
}
|
||||
|
||||
// cpuSaturation 返回相对容器可用 CPU 的使用率(百分比)。
|
||||
// 可用 CPU 取 cgroup 配额与可调度 CPU 的较小者:配额大于 cpuset 或节点核数时,
|
||||
// 按配额计算会让容器打满也无法触发保护。
|
||||
// ok=false 表示当前环境没有可用的 cgroup CPU 数据,调用方应回退到宿主机口径。
|
||||
func (u *containerUsage) cpuSaturation(now time.Time) (float64, bool) {
|
||||
limitCores, ok := u.readCgroupCPUQuotaCores()
|
||||
if !ok {
|
||||
return 0, false
|
||||
}
|
||||
if u.schedulableCPUs > 0 {
|
||||
limitCores = min(limitCores, u.schedulableCPUs)
|
||||
}
|
||||
usageUsec, ok := u.readCgroupCPUUsageUsec()
|
||||
if !ok {
|
||||
return 0, false
|
||||
}
|
||||
if u.cpuLastSampled.IsZero() || usageUsec < u.cpuLastUsec || !now.After(u.cpuLastSampled) {
|
||||
u.cpuLastUsec = usageUsec
|
||||
u.cpuLastSampled = now
|
||||
return 0, false
|
||||
}
|
||||
elapsed := now.Sub(u.cpuLastSampled)
|
||||
if elapsed < time.Second {
|
||||
return 0, false
|
||||
}
|
||||
deltaUsec := usageUsec - u.cpuLastUsec
|
||||
u.cpuLastUsec = usageUsec
|
||||
u.cpuLastSampled = now
|
||||
|
||||
usedCores := float64(deltaUsec) / 1e6 / elapsed.Seconds()
|
||||
return usedCores / limitCores * 100, true
|
||||
}
|
||||
|
||||
// memoryUsage 返回容器内存工作集相对上限的使用率(百分比)。
|
||||
// 工作集 = 用量 - 非活跃文件缓存,与 docker stats、kubelet 一致:用量包含可回收的
|
||||
// 页缓存,写日志或文件后会长期贴近上限,不代表内存压力。
|
||||
// ok=false 时调用方回退宿主机口径(同样不含缓存):内存未限制、文件不可读、
|
||||
// 或 memory.stat 缺少非活跃缓存字段。宁可回退旧口径,也不把含缓存的用量当作压力。
|
||||
func (u *containerUsage) memoryUsage() (float64, bool) {
|
||||
usagePath := filepath.Join(u.cgroupRoot, "memory.current")
|
||||
limitPath := filepath.Join(u.cgroupRoot, "memory.max")
|
||||
statPath := filepath.Join(u.cgroupRoot, "memory.stat")
|
||||
inactiveFileKey := "inactive_file"
|
||||
if _, err := os.Stat(usagePath); err != nil {
|
||||
v1Dir := filepath.Join(u.cgroupRoot, "memory")
|
||||
usagePath = filepath.Join(v1Dir, "memory.usage_in_bytes")
|
||||
limitPath = filepath.Join(v1Dir, "memory.limit_in_bytes")
|
||||
statPath = filepath.Join(v1Dir, "memory.stat")
|
||||
inactiveFileKey = "total_inactive_file"
|
||||
}
|
||||
|
||||
// cgroup v2 未限制时 memory.max 为 "max",解析失败即视为不可用。
|
||||
limit, ok := readCgroupUint(limitPath)
|
||||
if !ok || limit == 0 || limit >= unlimitedMemoryBytes {
|
||||
return 0, false
|
||||
}
|
||||
usage, ok := readCgroupUint(usagePath)
|
||||
if !ok {
|
||||
return 0, false
|
||||
}
|
||||
inactiveFile, ok := readCgroupStatValue(statPath, inactiveFileKey)
|
||||
if !ok {
|
||||
return 0, false
|
||||
}
|
||||
// 用量与 memory.stat 不是同一时刻的快照,缓存可能短暂大于用量,按 0 处理防止下溢。
|
||||
workingSet := usage - min(inactiveFile, usage)
|
||||
return float64(workingSet) / float64(limit) * 100, true
|
||||
}
|
||||
|
||||
// readCgroupCPUQuotaCores 返回 cgroup CPU 配额(核)。ok=false 表示未限制或不可读。
|
||||
func (u *containerUsage) readCgroupCPUQuotaCores() (float64, bool) {
|
||||
if raw, err := os.ReadFile(filepath.Join(u.cgroupRoot, "cpu.max")); err == nil {
|
||||
quotaText, periodText, found := strings.Cut(strings.TrimSpace(string(raw)), " ")
|
||||
if !found || quotaText == "max" {
|
||||
return 0, false
|
||||
}
|
||||
quota, err := strconv.ParseUint(quotaText, 10, 64)
|
||||
if err != nil || quota == 0 {
|
||||
return 0, false
|
||||
}
|
||||
period, err := strconv.ParseUint(strings.TrimSpace(periodText), 10, 64)
|
||||
if err != nil || period == 0 {
|
||||
return 0, false
|
||||
}
|
||||
return float64(quota) / float64(period), true
|
||||
}
|
||||
|
||||
// cgroup v1 未限制时 cfs_quota_us 为 -1,解析失败即视为不可用。
|
||||
v1Dir := filepath.Join(u.cgroupRoot, "cpu")
|
||||
quota, ok := readCgroupUint(filepath.Join(v1Dir, "cpu.cfs_quota_us"))
|
||||
if !ok || quota == 0 {
|
||||
return 0, false
|
||||
}
|
||||
period, ok := readCgroupUint(filepath.Join(v1Dir, "cpu.cfs_period_us"))
|
||||
if !ok || period == 0 {
|
||||
return 0, false
|
||||
}
|
||||
return float64(quota) / float64(period), true
|
||||
}
|
||||
|
||||
func (u *containerUsage) readCgroupCPUUsageUsec() (uint64, bool) {
|
||||
statPath := filepath.Join(u.cgroupRoot, "cpu.stat")
|
||||
if _, err := os.Stat(statPath); err == nil {
|
||||
return readCgroupStatValue(statPath, "usage_usec")
|
||||
}
|
||||
// cgroup v1 的 cpuacct.usage 单位是纳秒。
|
||||
nanos, ok := readCgroupUint(filepath.Join(u.cgroupRoot, "cpuacct", "cpuacct.usage"))
|
||||
return nanos / 1000, ok
|
||||
}
|
||||
|
||||
// readCgroupUint 读取只含一个无符号整数的 cgroup 文件。
|
||||
func readCgroupUint(path string) (uint64, bool) {
|
||||
raw, err := os.ReadFile(path)
|
||||
if err != nil {
|
||||
return 0, false
|
||||
}
|
||||
value, err := strconv.ParseUint(strings.TrimSpace(string(raw)), 10, 64)
|
||||
return value, err == nil
|
||||
}
|
||||
|
||||
// readCgroupStatValue 从 "key value" 逐行格式的 cgroup 统计文件(cpu.stat、memory.stat)中读取一项。
|
||||
func readCgroupStatValue(path string, key string) (uint64, bool) {
|
||||
raw, err := os.ReadFile(path)
|
||||
if err != nil {
|
||||
return 0, false
|
||||
}
|
||||
for line := range strings.SplitSeq(string(raw), "\n") {
|
||||
name, valueText, found := strings.Cut(line, " ")
|
||||
if !found || name != key {
|
||||
continue
|
||||
}
|
||||
value, err := strconv.ParseUint(strings.TrimSpace(valueText), 10, 64)
|
||||
return value, err == nil
|
||||
}
|
||||
return 0, false
|
||||
}
|
||||
Reference in new issue
Block a user