fix: make resource metrics platform aware

This commit is contained in:
ji ruo committed 2026-10-30 01:39:39 +08:00
1 parent 08cf862b0c
commit bf2ef120da
7 files changed
+124 -120

No files matched your search

+2 -73
View File
@@ -1,17 +1,8 @@
package server
// Container-level resource metrics. These deliberately read cgroups directly:
// generic host libraries often report the host's capacity from inside Docker.
import (
"encoding/json"
"os"
"net/http"
"path/filepath"
"runtime"
"strconv"
"strings"
"syscall"
"time"
)
@@ -23,6 +14,7 @@ type metricValue struct {
type systemMetricsResponse struct {
Scope string `json:"scope"`
Supported bool `json:"supported"`
Timestamp time.Time `json:"timestamp"`
CPU metricValue `json:"cpu"`
Memory metricValue `json:"memory"`
@@ -32,71 +24,8 @@ type systemMetricsResponse struct {
}
func (s *Server) systemMetrics(w http.ResponseWriter, r *http.Request) {
cpuUse, cpuLimit, ok := cgroupCPU()
cpuPct := 0.0
if ok {
now := time.Now()
s.metricsMu.Lock()
if !s.lastCPUAt.IsZero() && cpuUse >= s.lastCPUUse {
elapsed := now.Sub(s.lastCPUAt).Seconds()
if elapsed > 0 {
cpuPct = float64(cpuUse-s.lastCPUUse) / 1e6 / elapsed / cpuLimit * 100
}
}
s.lastCPUUse, s.lastCPUAt = cpuUse, now
s.metricsMu.Unlock()
}
memUsed, memTotal, memOK := cgroupMemory()
root := diskMetric("/")
data := diskMetric(s.dataDir)
if data.TotalBytes == 0 { data = root }
containerMetrics := ok || memOK
scope := "process"
if containerMetrics {
scope = "container"
}
out := systemMetricsResponse{Scope: scope, Timestamp: time.Now().UTC(), Container: containerMetrics,
CPU: metricValue{Percent: clampPercent(cpuPct)},
Memory: metricValue{UsedBytes: memUsed, TotalBytes: memTotal, Percent: percent(memUsed, memTotal)}, Disk: root, DataDisk: data}
out := systemMetricsSnapshot(s)
w.Header().Set("Cache-Control", "no-store")
w.Header().Set("Content-Type", "application/json")
_ = json.NewEncoder(w).Encode(out)
}
func cgroupCPU() (uint64, float64, bool) {
for _, dir := range []string{"/sys/fs/cgroup", "/sys/fs/cgroup/cpu"} {
b, err := os.ReadFile(filepath.Join(dir, "cpu.stat")); if err != nil { continue }
var use uint64
for _, line := range strings.Split(string(b), "\n") { f := strings.Fields(line); if len(f) == 2 && f[0] == "usage_usec" { use, _ = strconv.ParseUint(f[1], 10, 64) } }
limit := float64(runtime.NumCPU())
if q, err := os.ReadFile(filepath.Join(dir, "cpu.max")); err == nil { f := strings.Fields(string(q)); if len(f) == 2 && f[0] != "max" { quota, _ := strconv.ParseFloat(f[0], 64); period, _ := strconv.ParseFloat(f[1], 64); if period > 0 { limit = quota / period } } }
if use > 0 { return use, limit, true }
}
return 0, 0, false
}
func cgroupMemory() (uint64, uint64, bool) {
for _, dir := range []string{"/sys/fs/cgroup", "/sys/fs/cgroup/memory"} {
cur, e1 := readUint(filepath.Join(dir, "memory.current")); max, e2 := readUint(filepath.Join(dir, "memory.max"))
if e1 == nil && e2 == nil { return cur, max, max > 0 }
if e1 == nil {
if b, err := os.ReadFile(filepath.Join(dir, "memory.max")); err == nil && strings.TrimSpace(string(b)) == "max" {
return cur, hostMemoryTotal(), true
}
}
cur, e1 = readUint(filepath.Join(dir, "memory.usage_in_bytes")); max, e2 = readUint(filepath.Join(dir, "memory.limit_in_bytes")); if e1 == nil && e2 == nil && max < ^uint64(0)/2 { return cur, max, true }
}
return 0, 0, false
}
func hostMemoryTotal() uint64 {
b, err := os.ReadFile("/proc/meminfo"); if err != nil { return 0 }
for _, line := range strings.Split(string(b), "\n") { f := strings.Fields(line); if len(f) >= 2 && f[0] == "MemTotal:" { n, _ := strconv.ParseUint(f[1], 10, 64); return n * 1024 } }
return 0
}
func readUint(path string) (uint64, error) { b, err := os.ReadFile(path); if err != nil { return 0, err }; return strconv.ParseUint(strings.TrimSpace(string(b)), 10, 64) }
func diskMetric(path string) metricValue { var st syscall.Statfs_t; if path == "" || syscall.Statfs(path, &st) != nil { return metricValue{} }; total := st.Blocks * uint64(st.Bsize); free := st.Bavail * uint64(st.Bsize); used := total - free; return metricValue{UsedBytes: used, TotalBytes: total, Percent: percent(used, total)} }
func percent(used, total uint64) float64 { if total == 0 { return 0 }; return clampPercent(float64(used) / float64(total) * 100) }
func clampPercent(v float64) float64 { if v < 0 { return 0 }; if v > 100 { return 100 }; return v }
+62
View File
@@ -0,0 +1,62 @@
//go:build linux
package server
import (
"os"
"path/filepath"
"runtime"
"strconv"
"strings"
"syscall"
"time"
)
func systemMetricsSnapshot(s *Server) systemMetricsResponse {
use, limit, cpuOK := cgroupCPU()
cpuPct := 0.0
if cpuOK {
now := time.Now()
s.metricsMu.Lock()
if !s.lastCPUAt.IsZero() && use >= s.lastCPUUse && limit > 0 {
elapsed := now.Sub(s.lastCPUAt).Seconds()
if elapsed > 0 { cpuPct = float64(use-s.lastCPUUse) / 1e6 / elapsed / limit * 100 }
}
s.lastCPUUse, s.lastCPUAt = use, now
s.metricsMu.Unlock()
}
memUsed, memTotal, memOK := cgroupMemory()
root, data := diskMetric("/"), diskMetric(s.dataDir)
if data.TotalBytes == 0 { data = root }
return systemMetricsResponse{Scope: "container", Supported: true, Timestamp: time.Now().UTC(), Container: cpuOK || memOK,
CPU: metricValue{Percent: clampPercent(cpuPct)}, Memory: metricValue{UsedBytes: memUsed, TotalBytes: memTotal, Percent: percent(memUsed, memTotal)}, Disk: root, DataDisk: data}
}
func cgroupCPU() (uint64, float64, bool) {
for _, dir := range []string{"/sys/fs/cgroup", "/sys/fs/cgroup/cpu"} {
b, err := os.ReadFile(filepath.Join(dir, "cpu.stat")); if err != nil { continue }
var use uint64
for _, line := range strings.Split(string(b), "\n") { f := strings.Fields(line); if len(f) == 2 && f[0] == "usage_usec" { use, _ = strconv.ParseUint(f[1], 10, 64) } }
limit := float64(runtime.NumCPU())
if q, err := os.ReadFile(filepath.Join(dir, "cpu.max")); err == nil { f := strings.Fields(string(q)); if len(f) == 2 && f[0] != "max" { quota, _ := strconv.ParseFloat(f[0], 64); period, _ := strconv.ParseFloat(f[1], 64); if period > 0 { limit = quota / period } } }
if use > 0 { return use, limit, true }
}
return 0, 0, false
}
func cgroupMemory() (uint64, uint64, bool) {
for _, dir := range []string{"/sys/fs/cgroup", "/sys/fs/cgroup/memory"} {
cur, ce := readUint(filepath.Join(dir, "memory.current")); max, me := readUint(filepath.Join(dir, "memory.max"))
if ce == nil && me == nil { return cur, max, max > 0 }
if ce == nil { if b, err := os.ReadFile(filepath.Join(dir, "memory.max")); err == nil && strings.TrimSpace(string(b)) == "max" { return cur, hostMemoryTotal(), true } }
cur, ce = readUint(filepath.Join(dir, "memory.usage_in_bytes")); max, me = readUint(filepath.Join(dir, "memory.limit_in_bytes"))
if ce == nil && me == nil && max < ^uint64(0)/2 { return cur, max, true }
}
return 0, 0, false
}
func hostMemoryTotal() uint64 { b, err := os.ReadFile("/proc/meminfo"); if err != nil { return 0 }; for _, line := range strings.Split(string(b), "\n") { f := strings.Fields(line); if len(f) >= 2 && f[0] == "MemTotal:" { n, _ := strconv.ParseUint(f[1], 10, 64); return n * 1024 } }; return 0 }
func readUint(path string) (uint64, error) { b, err := os.ReadFile(path); if err != nil { return 0, err }; return strconv.ParseUint(strings.TrimSpace(string(b)), 10, 64) }
func diskMetric(path string) metricValue { var st syscall.Statfs_t; if path == "" || syscall.Statfs(path, &st) != nil { return metricValue{} }; total, free := st.Blocks*uint64(st.Bsize), st.Bavail*uint64(st.Bsize); used := total-free; return metricValue{UsedBytes: used, TotalBytes: total, Percent: percent(used, total)} }
func percent(used, total uint64) float64 { if total == 0 { return 0 }; return clampPercent(float64(used)/float64(total)*100) }
func clampPercent(v float64) float64 { if v < 0 { return 0 }; if v > 100 { return 100 }; return v }
+7
View File
@@ -3,6 +3,7 @@ package server
import (
"encoding/json"
"net/http/httptest"
"runtime"
"testing"
)
@@ -21,6 +22,12 @@ func TestSystemMetricsReturnsContainerAndDiskValues(t *testing.T) {
if got.Scope == "" || got.Timestamp.IsZero() {
t.Fatalf("unexpected response metadata: %+v", got)
}
if runtime.GOOS != "linux" {
if got.Supported || got.Scope != "unsupported" {
t.Fatalf("unsupported platform returned metrics: %+v", got)
}
return
}
if got.Disk.TotalBytes == 0 || got.DataDisk.TotalBytes == 0 {
t.Fatalf("disk metrics missing: %+v", got)
}
+9
View File
@@ -0,0 +1,9 @@
//go:build !linux
package server
import "time"
func systemMetricsSnapshot(_ *Server) systemMetricsResponse {
return systemMetricsResponse{Scope: "unsupported", Supported: false, Timestamp: time.Now().UTC()}
}