Dashboard: Processes module (top by CPU/memory, PID, user, threads, SIGTERM), sampled at most every 2 s

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

internal/metrics/metrics.go

@@ -12,8 +12,10 @@ package metrics
 import (
 	"bufio"
 	"context"
+	"errors"
 	"os"
 	"os/exec"
+	"os/user"
 	"path/filepath"
 	"regexp"
 	"sort"
@@ -95,6 +97,17 @@ type UPS struct {
 	LastEvent  string  `json:"last_event,omitempty"`
 }
 
+type Proc struct {
+	PID     int     `json:"pid"`
+	Name    string  `json:"name"`
+	Cmd     string  `json:"cmd,omitempty"`
+	User    string  `json:"user"`
+	CPU     float64 `json:"cpu"` // percent of one core
+	RSS     uint64  `json:"rss"` // bytes
+	Threads int     `json:"threads"`
+	State   string  `json:"state"`
+}
+
 type Sample struct {
 	T        int64      `json:"t"`           // unix ms
 	Interval int64      `json:"interval_ms"` // sampling interval in effect
@@ -109,6 +122,7 @@ type Sample struct {
 	Disks    []Disk     `json:"disks"`
 	FS       []FS       `json:"fs"`
 	UPS      *UPS       `json:"ups"`
+	Procs    []Proc     `json:"procs"` // top by CPU and by RSS, merged
 	Uptime   float64    `json:"uptime"`
 }
 
@@ -136,6 +150,9 @@ type Collector struct {
 	fsAt     time.Time
 	lastUPS  *UPS
 	upsAt    time.Time
+	prevProc map[int]procTimes
+	procAt   time.Time
+	lastProc []Proc
 	upsProbe func() *UPS // nil when no UPS daemon is installed
 }
 
@@ -271,6 +288,11 @@ func (c *Collector) sample() {
 		c.lastUPS, c.upsAt = c.upsProbe(), now
 	}
 	s.UPS = c.lastUPS
+	// Processes: walking /proc is the expensive part, so at most every 2 s regardless of interval.
+	if c.lastProc == nil || now.Sub(c.procAt) >= 2*time.Second {
+		c.lastProc, c.procAt = c.procs(now), now
+	}
+	s.Procs = c.lastProc
 
 	if first {
 		return // deltas are meaningless on the priming pass
@@ -835,6 +857,132 @@ func upsAPC() *UPS {
 	return u
 }
 
+// ---- processes ----
+
+type procTimes struct {
+	ticks uint64
+	at    time.Time
+}
+
+var clkTck = float64(100) // Linux USER_HZ; sysconf(_SC_CLK_TCK) is 100 on every mainstream build
+var uidNames = map[string]string{}
+
+func uidName(uid string) string {
+	if n, ok := uidNames[uid]; ok {
+		return n
+	}
+	n := uid
+	if u, err := user.LookupId(uid); err == nil {
+		n = u.Username
+	}
+	uidNames[uid] = n
+	return n
+}
+
+// procs returns the union of the top 15 by CPU and top 15 by RSS.
+func (c *Collector) procs(now time.Time) []Proc {
+	ents, err := os.ReadDir("/proc")
+	if err != nil {
+		return []Proc{}
+	}
+	if c.prevProc == nil {
+		c.prevProc = map[int]procTimes{}
+	}
+	cur := map[int]procTimes{}
+	var all []Proc
+	pageSize := uint64(os.Getpagesize())
+	for _, e := range ents {
+		pid, err := strconv.Atoi(e.Name())
+		if err != nil || !e.IsDir() {
+			continue
+		}
+		stat := readStr("/proc/" + e.Name() + "/stat")
+		// "pid (comm with spaces) state ppid ..." — comm can contain spaces and parens
+		close := strings.LastIndex(stat, ")")
+		open := strings.Index(stat, "(")
+		if open < 0 || close < 0 {
+			continue
+		}
+		name := stat[open+1 : close]
+		f := strings.Fields(stat[close+2:])
+		if len(f) < 22 {
+			continue
+		}
+		utime, _ := strconv.ParseUint(f[11], 10, 64)
+		stime, _ := strconv.ParseUint(f[12], 10, 64)
+		threads, _ := strconv.Atoi(f[17])
+		rssPages, _ := strconv.ParseUint(f[21], 10, 64)
+		ticks := utime + stime
+		cur[pid] = procTimes{ticks: ticks, at: now}
+		p := Proc{PID: pid, Name: name, State: f[0], Threads: threads, RSS: rssPages * pageSize}
+		if prev, ok := c.prevProc[pid]; ok && prev.ticks <= ticks {
+			dt := now.Sub(prev.at).Seconds()
+			if dt > 0 {
+				p.CPU = round1(float64(ticks-prev.ticks) / clkTck / dt * 100)
+			}
+		}
+		all = append(all, p)
+	}
+	c.prevProc = cur
+	// pick: top by CPU, then fill with top by RSS
+	sort.Slice(all, func(i, j int) bool { return all[i].CPU > all[j].CPU })
+	pick := map[int]bool{}
+	var out []Proc
+	for i := 0; i < len(all) && i < 15; i++ {
+		if all[i].CPU > 0 || i < 5 {
+			pick[all[i].PID] = true
+			out = append(out, all[i])
+		}
+	}
+	sort.Slice(all, func(i, j int) bool { return all[i].RSS > all[j].RSS })
+	for i := 0; i < len(all) && i < 15; i++ {
+		if !pick[all[i].PID] {
+			pick[all[i].PID] = true
+			out = append(out, all[i])
+		}
+	}
+	// enrich only the chosen ones (user, cmdline) — cheap because it's ≤30 reads
+	for i := range out {
+		pidS := strconv.Itoa(out[i].PID)
+		for _, line := range strings.Split(readStr("/proc/"+pidS+"/status"), "\n") {
+			if strings.HasPrefix(line, "Uid:") {
+				if fs := strings.Fields(line); len(fs) > 1 {
+					out[i].User = uidName(fs[1])
+				}
+				break
+			}
+		}
+		if cmd := strings.ReplaceAll(strings.TrimRight(readStr("/proc/"+pidS+"/cmdline"), "\x00"), "\x00", " "); cmd != "" {
+			if len(cmd) > 160 {
+				cmd = cmd[:160] + "…"
+			}
+			out[i].Cmd = cmd
+		}
+	}
+	if out == nil {
+		out = []Proc{}
+	}
+	return out
+}
+
+// Signal sends a signal to a pid; falls back to sudo when the process isn't ours.
+func Signal(pid int, sig syscall.Signal) error {
+	if pid <= 1 {
+		return errors.New("refusing to signal pid " + strconv.Itoa(pid))
+	}
+	if err := syscall.Kill(pid, sig); err == nil {
+		return nil
+	} else if !errors.Is(err, syscall.EPERM) {
+		return err
+	}
+	if lookPath("sudo") {
+		if err := exec.Command("sudo", "-n", "kill", "-"+strconv.Itoa(int(sig)), strconv.Itoa(pid)).Run(); err == nil {
+			return nil
+		}
+	}
+	return errors.New("permission denied")
+}
+
 // ---- small helpers ----
 
 func readStr(p string) string {