In /proc/stat, guest is already inside user (and guest_nice inside nice). Peardata was adding guest into the total again, which shrinks every other slice — including idle.

readFileCached capped at 64KB. On 56-core hosts /proc/interrupts (and similar) often exceeds that — reads were silently truncated. Buffer now grows (up to 1MB). That didn’t break system.cpu idle (aggregate line is first), but it was a real high-core regression.

server/services/collector.js

@@ -161,6 +161,11 @@ function parseProcStat() {
   return { aggregate, cores, intr, ctxt, processes, procsRunning, procsBlocked }
 }
 
+/**
+ * Convert /proc/stat jiffy counters to percentages.
+ * Linux accounts guest in user and guest_nice in nice — subtract before totaling
+ * (same approach as htop/netdata) so idle isn't understated on hypervisor hosts.
+ */
 function cpuDeltaPct(prev, cur) {
   if (!prev || !cur) {
     return {
@@ -176,25 +181,26 @@ function cpuDeltaPct(prev, cur) {
       idle: 100,
     }
   }
-  const keys = [
-    'user',
-    'nice',
-    'system',
-    'idle',
-    'iowait',
-    'irq',
-    'softirq',
-    'steal',
-    'guest',
-    'guest_nice',
-  ]
-  /** @type {Record<string, number>} */
-  const d = {}
-  let total = 0
-  for (const k of keys) {
-    d[k] = Math.max(0, (cur[k] || 0) - (prev[k] || 0))
-    total += d[k]
+  const delta = (k) => Math.max(0, (cur[k] || 0) - (prev[k] || 0))
+  const guest = delta('guest')
+  const guestNice = delta('guest_nice')
+  // guest ⊆ user, guest_nice ⊆ nice in kernel counters
+  const user = Math.max(0, delta('user') - guest)
+  const nice = Math.max(0, delta('nice') - guestNice)
+  const d = {
+    guest_nice: guestNice,
+    guest,
+    steal: delta('steal'),
+    softirq: delta('softirq'),
+    irq: delta('irq'),
+    user,
+    system: delta('system'),
+    nice,
+    iowait: delta('iowait'),
+    idle: delta('idle'),
   }
+  let total = 0
+  for (const v of Object.values(d)) total += v
   if (total <= 0) total = 1
   return {
     guest_nice: pct(d.guest_nice, total),
@@ -423,6 +429,8 @@ export class MetricsCollector extends EventEmitter {
     /** @type {Map<string, Record<string, number|null>>} */
     this.latest = new Map()
     this._chartsRegistered = false
+    this._lastCpuUsage = process.cpuUsage()
+    this._lastCpuTs = Date.now()
   }
 
   start() {
@@ -509,7 +517,20 @@ export class MetricsCollector extends EventEmitter {
       log.error('Collector tick failed', { error: err.message })
     }
     const dur = performance.now() - t0
-    if (dur > 15) log.debug('tick', { collector: 'main', durMs: Math.round(dur * 10) / 10 })
+    if (dur > 15) {
+      const now = Date.now()
+      const cpuDelta = process.cpuUsage(this._lastCpuUsage)
+      const wallMs = now - this._lastCpuTs
+      this._lastCpuUsage = process.cpuUsage()
+      this._lastCpuTs = now
+      const totalMs = (cpuDelta.user + cpuDelta.system) / 1000
+      const pct = wallMs > 0 ? Math.round((totalMs / wallMs) * 100) : 0
+      if (pct > 30) {
+        log.debug('tick', { collector: 'main', durMs: Math.round(dur * 10) / 10, cpuPct: pct, cpuUserMs: Math.round(cpuDelta.user / 1000), cpuSysMs: Math.round(cpuDelta.system / 1000), wallMs })
+      } else {
+        log.debug('tick', { collector: 'main', durMs: Math.round(dur * 10) / 10 })
+      }
+    }
   }
 
   _safeCollect(name, fn) {
@@ -1290,3 +1311,6 @@ export function getCollector() {
   if (!singleton) singleton = new MetricsCollector()
   return singleton
 }
+
+/** @internal exported for unit tests */
+export { cpuDeltaPct }

server/services/collectors/cgroups.js

@@ -49,20 +49,6 @@ function readFile(p) {
   return readFileBuf(p)
 }
 
-let _profileIdx = 0
-function profile(label, fn) {
-  const t0 = performance.now()
-  const r = fn()
-  const dur = performance.now() - t0
-  if (dur > 2) {
-    _profileIdx++
-    if (_profileIdx <= 40) {
-      process.stderr.write(`cgroups:${label} ${Math.round(dur * 10) / 10}ms\n`)
-    }
-  }
-  return r
-}
-
 function cgroupRoot() {
   if (fs.existsSync('/sys/fs/cgroup/cgroup.controllers')) return '/sys/fs/cgroup'
   return null
@@ -294,18 +280,18 @@ export class CgroupsCollector extends EventEmitter {
       const memDetailDef = makeCgroupMemDetailChart(cg.id, title)
       const throttleDef = makeCgroupThrottleChart(cg.id, title)
 
-      const cpu = profile('cpuStat:' + cg.id.slice(0,20), () => parseCpuStat(cg.path))
+      const cpu = parseCpuStat(cg.path)
       let usageUsec = cpu?.usage_usec ?? 0
       let userUsec = cpu?.user_usec ?? 0
       let systemUsec = cpu?.system_usec ?? 0
       const throttledUsec = cpu?.throttled_usec ?? 0
 
-      const memCur = profile('memCur:' + cg.id.slice(0,20), () => Number(readFile(path.join(cg.path, 'memory.current')) || 0))
+      const memCur = Number(readFile(path.join(cg.path, 'memory.current')) || 0)
       const memMaxRaw = readFile(path.join(cg.path, 'memory.max'))
       const memMax =
         memMaxRaw && memMaxRaw.trim() !== 'max' ? Number(memMaxRaw.trim()) || 0 : 0
 
-      const io = profile('ioStat:' + cg.id.slice(0,20), () => parseIoStat(cg.path))
+      const io = parseIoStat(cg.path)
       const prev = this.prev.get(cg.id)
 
       let userPct = 0
@@ -347,7 +333,7 @@ export class CgroupsCollector extends EventEmitter {
         },
       })
 
-      const memStat = profile('memStat:' + cg.id.slice(0,20), () => parseMemoryStat(cg.path))
+      const memStat = parseMemoryStat(cg.path)
       if (memStat) {
         batch.push({
           chart: memDetailDef.id,

server/services/collectors/host-more.js

@@ -219,7 +219,7 @@ export function collectCpuidle(batch, ts, dtSec, state) {
     } catch {
       return
     }
-    /** @type {Record<string, Array<{ nameFile: string, timeFile: string }>>} */
+    /** @type {Record<string, Array<{ id: string, nameFile: string, timeFile: string }>>} */
     const entries = {}
     for (const d of dirs) {
       const coreId = d.slice(3)
@@ -233,6 +233,7 @@ export function collectCpuidle(batch, ts, dtSec, state) {
       const files = []
       for (const s of states) {
         files.push({
+          id: s,
           nameFile: path.join(idleDir, s, 'name'),
           timeFile: path.join(idleDir, s, 'time'),
         })
@@ -251,10 +252,10 @@ export function collectCpuidle(batch, ts, dtSec, state) {
     const files = entries[coreId]
     /** @type {Record<string, number>} */
     const times = {}
-    for (const { nameFile, timeFile } of files) {
-      const name = (readFile(nameFile) || '').trim().toLowerCase().replace(/\s+/g, '_')
+    for (const { id, nameFile, timeFile } of files) {
+      const name = (readFile(nameFile) || id).trim().toLowerCase().replace(/\s+/g, '_')
       const t = Number(readFile(timeFile) || 0) // µs
-      times[name || 'state'] = t
+      times[name || id] = t
     }
     nextAll[coreId] = times
     const prev = prevAll[coreId]