@glassmkr/crucible 0.7.1 → 0.8.1

package/src/collect/fd.ts

@@ -1,31 +0,0 @@
-import { readProcFile } from "../lib/parse.js";
-
-export interface FileDescriptorData {
-  allocated: number;
-  free: number;
-  max: number;
-  percent: number;
-}
-
-export function collectFileDescriptors(): FileDescriptorData {
-  const raw = readProcFile("/proc/sys/fs/file-nr");
-  if (!raw) {
-    return { allocated: 0, free: 0, max: 0, percent: 0 };
-  }
-
-  const parts = raw.trim().split(/\s+/);
-  if (parts.length < 3) {
-    return { allocated: 0, free: 0, max: 0, percent: 0 };
-  }
-
-  const allocated = parseInt(parts[0], 10);
-  const free = parseInt(parts[1], 10);
-  const max = parseInt(parts[2], 10);
-
-  if (isNaN(allocated) || isNaN(max) || max === 0) {
-    return { allocated: 0, free: 0, max: 0, percent: 0 };
-  }
-
-  const percent = Math.round(((allocated / max) * 100) * 10) / 10;
-  return { allocated, free: isNaN(free) ? 0 : free, max, percent };
-}

package/src/collect/io-errors.ts

@@ -1,23 +0,0 @@
-import { run } from "../lib/exec.js";
-
-export async function collectIoErrors(): Promise<{ count: number; devices: string[] } | null> {
-  // Parse dmesg for recent I/O errors (last 10 minutes covers the 5-min collection interval)
-  const output = await run("bash", ["-c", 'dmesg -T --since "10 minutes ago" 2>/dev/null | grep -i "I/O error\\|Buffer I/O error\\|blk_update_request.*error"'], 5000);
-  if (!output || !output.trim()) return null;
-
-  const lines = output.trim().split("\n").filter((l) => l.trim());
-  if (lines.length === 0) return null;
-
-  // Extract device names from error messages
-  const deviceSet = new Set<string>();
-  for (const line of lines) {
-    // "blk_update_request: I/O error, dev sda, sector 12345"
-    const devMatch = line.match(/dev\s+(\w+)/);
-    if (devMatch) deviceSet.add(devMatch[1]);
-    // "Buffer I/O error on device sda1"
-    const bufMatch = line.match(/on device\s+(\w+)/);
-    if (bufMatch) deviceSet.add(bufMatch[1]);
-  }
-
-  return { count: lines.length, devices: Array.from(deviceSet) };
-}

package/src/collect/io-latency.ts

@@ -1,103 +0,0 @@
-import { readProcFile } from "../lib/parse.js";
-
-export interface IoLatencyInfo {
-  device: string;
-  avg_read_latency_ms: number | null;
-  avg_write_latency_ms: number | null;
-  read_iops: number;
-  write_iops: number;
-}
-
-interface DiskstatsCounters {
-  reads_completed: number;
-  read_time_ms: number;
-  writes_completed: number;
-  write_time_ms: number;
-}
-
-// Previous cumulative counters for delta computation
-const previousCounters = new Map<string, DiskstatsCounters>();
-
-// Match physical block devices, not partitions or virtual devices
-function isPhysicalDevice(name: string): boolean {
-  // sd*, vd*, xvd* without trailing partition number
-  if (/^(sd|vd|xvd)[a-z]+$/.test(name)) return true;
-  // nvme*n* without partition suffix (nvme0n1 yes, nvme0n1p1 no)
-  if (/^nvme\d+n\d+$/.test(name)) return true;
-  // md* (RAID arrays)
-  if (/^md\d+$/.test(name)) return true;
-  return false;
-}
-
-function parseDiskstats(): Record<string, DiskstatsCounters> {
-  const raw = readProcFile("/proc/diskstats") || "";
-  const result: Record<string, DiskstatsCounters> = {};
-
-  for (const line of raw.split("\n")) {
-    const parts = line.trim().split(/\s+/);
-    if (parts.length < 11) continue;
-
-    const name = parts[2];
-    if (!isPhysicalDevice(name)) continue;
-
-    result[name] = {
-      reads_completed: Number(parts[3]) || 0,
-      read_time_ms: Number(parts[6]) || 0,
-      writes_completed: Number(parts[7]) || 0,
-      write_time_ms: Number(parts[10]) || 0,
-    };
-  }
-
-  return result;
-}
-
-function delta(current: number, previous: number): number {
-  if (current >= previous) return current - previous;
-  return current; // counter wrapped or reset
-}
-
-export function collectIoLatency(): IoLatencyInfo[] {
-  const current = parseDiskstats();
-  const results: IoLatencyInfo[] = [];
-  const currentDevices = new Set<string>();
-
-  for (const [name, counters] of Object.entries(current)) {
-    currentDevices.add(name);
-    const prev = previousCounters.get(name);
-
-    // Store current for next cycle
-    previousCounters.set(name, { ...counters });
-
-    if (!prev) {
-      // First cycle: no delta, report null latency
-      results.push({
-        device: name,
-        avg_read_latency_ms: null,
-        avg_write_latency_ms: null,
-        read_iops: 0,
-        write_iops: 0,
-      });
-      continue;
-    }
-
-    const deltaReads = delta(counters.reads_completed, prev.reads_completed);
-    const deltaReadTime = delta(counters.read_time_ms, prev.read_time_ms);
-    const deltaWrites = delta(counters.writes_completed, prev.writes_completed);
-    const deltaWriteTime = delta(counters.write_time_ms, prev.write_time_ms);
-
-    results.push({
-      device: name,
-      avg_read_latency_ms: deltaReads > 0 ? Math.round((deltaReadTime / deltaReads) * 100) / 100 : null,
-      avg_write_latency_ms: deltaWrites > 0 ? Math.round((deltaWriteTime / deltaWrites) * 100) / 100 : null,
-      read_iops: deltaReads,
-      write_iops: deltaWrites,
-    });
-  }
-
-  // Remove stale devices
-  for (const name of previousCounters.keys()) {
-    if (!currentDevices.has(name)) previousCounters.delete(name);
-  }
-
-  return results;
-}

package/src/collect/ipmi.ts

@@ -1,207 +0,0 @@
-import { run } from "../lib/exec.js";
-import type { IpmiInfo, SelEvent, FanStatus } from "../lib/types.js";
-
-export async function collectIpmi(): Promise<IpmiInfo> {
-  const sensorRaw = await run("ipmitool", ["sensor"]);
-  if (!sensorRaw) {
-    return { available: false, sensors: [], ecc_errors: { correctable: 0, uncorrectable: 0 }, sel_entries_count: 0, sel_events_recent: [], fans: [] };
-  }
-
-  // Parse sensor readings
-  const sensors: IpmiInfo["sensors"] = [];
-  for (const line of sensorRaw.split("\n")) {
-    const parts = line.split("|").map((s) => s.trim());
-    if (parts.length < 4) continue;
-    const name = parts[0];
-    const rawValue = parts[1];
-    const unit = parts[2];
-    const status = parts[3];
-
-    const numValue = parseFloat(rawValue);
-    const value: number | string = isNaN(numValue) ? rawValue : numValue;
-
-    let upperCritical: number | undefined;
-    if (parts[8]) {
-      const uc = parseFloat(parts[8]);
-      if (!isNaN(uc)) upperCritical = uc;
-    }
-
-    sensors.push({ name, value, unit, status, upper_critical: upperCritical });
-  }
-
-  // ECC errors from memory-type sensors
-  let correctable = 0;
-  let uncorrectable = 0;
-  for (const sensor of sensors) {
-    const name = sensor.name.toLowerCase();
-    if (name.includes("correctable") && typeof sensor.value === "number") {
-      correctable += sensor.value;
-    }
-    if (name.includes("uncorrectable") && typeof sensor.value === "number") {
-      uncorrectable += sensor.value;
-    }
-  }
-
-  // SEL entry count
-  let selCount = 0;
-  const selInfo = await run("ipmitool", ["sel", "info"]);
-  if (selInfo) {
-    const match = selInfo.match(/Entries\s*:\s*(\d+)/i);
-    if (match) selCount = parseInt(match[1], 10);
-  }
-
-  // SEL recent events
-  const selEvents = await collectSelEvents();
-
-  // Fan status
-  const fans = await collectFanStatus();
-
-  return {
-    available: true,
-    sensors,
-    ecc_errors: { correctable, uncorrectable },
-    sel_entries_count: selCount,
-    sel_events_recent: selEvents,
-    fans,
-  };
-}
-
-async function collectSelEvents(): Promise<SelEvent[]> {
-  const output = await run("ipmitool", ["sel", "elist"]);
-  if (!output) return [];
-
-  const events: SelEvent[] = [];
-  const lines = output.trim().split("\n");
-  const fiveMinAgo = new Date(Date.now() - 5 * 60 * 1000);
-
-  for (const line of lines) {
-    const parts = line.split("|").map((s) => s.trim());
-    if (parts.length < 5) continue;
-
-    const [idStr, date, time, sensor, event, direction] = parts;
-
-    const timestamp = parseSelTimestamp(date, time);
-    const tsDate = new Date(timestamp);
-
-    // Only include events from the last 5 minutes on subsequent runs
-    // On first run this will include everything (fiveMinAgo is always recent)
-    // We keep last 20 events max regardless
-    const sensorType = classifySensor(sensor);
-    const severity = deriveSelSeverity(event, sensorType);
-
-    events.push({
-      id: parseInt(idStr) || 0,
-      timestamp,
-      sensor,
-      sensor_type: sensorType,
-      event,
-      direction: direction || "Asserted",
-      severity,
-    });
-  }
-
-  // Return last 20 events, most recent first
-  return events.slice(-20).reverse();
-}
-
-export function parseSelTimestamp(date: string, time: string): string {
-  if (!date || !time) return new Date().toISOString();
-  // Format: "04/05/2026" and "14:23:05"
-  const parts = date.split("/");
-  if (parts.length !== 3) return new Date().toISOString();
-  const [month, day, year] = parts;
-  return `${year}-${month.padStart(2, "0")}-${day.padStart(2, "0")}T${time}Z`;
-}
-
-export function classifySensor(sensor: string): string {
-  const lower = sensor.toLowerCase();
-  if (lower.includes("memory") || lower.includes("dimm")) return "memory";
-  if (lower.includes("power supply") || lower.includes("psu")) return "power";
-  if (lower.includes("fan")) return "fan";
-  if (lower.includes("watchdog")) return "watchdog";
-  if (lower.includes("processor") || lower.includes("cpu")) return "processor";
-  if (lower.includes("temperature") || lower.includes("temp")) return "temperature";
-  if (lower.includes("voltage")) return "voltage";
-  if (lower.includes("drive") || lower.includes("disk")) return "storage";
-  if (lower.includes("chassis") || lower.includes("intrusion")) return "chassis";
-  return "other";
-}
-
-export function deriveSelSeverity(event: string, sensorType: string): string {
-  const lower = event.toLowerCase();
-
-  // Critical events
-  if (lower.includes("uncorrectable")) return "critical";
-  if (lower.includes("failure detected")) return "critical";
-  if (lower.includes("ac lost")) return "critical";
-  if (lower.includes("hard reset")) return "critical";
-  if (lower.includes("power off")) return "critical";
-  if (lower.includes("critical")) return "critical";
-  if (lower.includes("non-recoverable")) return "critical";
-  if (lower.includes("thermal trip")) return "critical";
-  if (lower.includes("processor disabled")) return "critical";
-  if (lower.includes("machine check")) return "critical";
-
-  // Warning events
-  if (lower.includes("correctable ecc")) return "warning";
-  if (lower.includes("logging limit")) return "warning";
-  if (lower.includes("lower critical going low")) return "warning";
-  if (lower.includes("upper critical going high")) return "warning";
-  if (lower.includes("redundancy lost")) return "warning";
-  if (lower.includes("predictive failure")) return "warning";
-  if (lower.includes("degraded")) return "warning";
-
-  // Info events
-  if (lower.includes("presence detected")) return "info";
-  if (lower.includes("power cycle")) return "info";
-  if (lower.includes("oem")) return "info";
-
-  if (["memory", "power", "fan", "processor"].includes(sensorType)) return "warning";
-  return "info";
-}
-
-async function collectFanStatus(): Promise<FanStatus[]> {
-  const output = await run("ipmitool", ["sdr", "type", "Fan"]);
-  if (!output) return [];
-  return parseFanStatus(output);
-}
-
-export function parseFanStatus(output: string): FanStatus[] {
-  const fans: FanStatus[] = [];
-  const lines = output.trim().split("\n");
-
-  for (const line of lines) {
-    const parts = line.split("|").map((s) => s.trim());
-    if (parts.length < 3) continue;
-
-    const name = parts[0];
-    const fullLine = parts.join(" ");
-
-    let rpm = 0;
-    let status = "ok";
-
-    // Search all fields for RPM value (format varies by BMC)
-    const rpmMatch = fullLine.match(/(\d+)\s*RPM/i);
-    if (rpmMatch) {
-      rpm = parseInt(rpmMatch[1]);
-    }
-
-    // Check status codes across all fields
-    const hasNoReading = fullLine.toLowerCase().includes("no reading");
-    const statusCodes = parts.slice(1).map((p) => p.toLowerCase());
-    const hasCritical = statusCodes.some((s) => s === "cr" || s === "nr");
-    const hasWarning = statusCodes.some((s) => s === "nc");
-    const hasAbsent = statusCodes.some((s) => s === "ns") || hasNoReading;
-    const hasOk = statusCodes.some((s) => s === "ok");
-
-    if (hasCritical) status = "critical";
-    else if (hasWarning) status = "warning";
-    else if (hasAbsent) status = "absent";
-    else if (hasOk) status = "ok";
-    else if (rpm === 0 && !hasNoReading) status = "critical";
-
-    fans.push({ name, rpm, status });
-  }
-
-  return fans;
-}

package/src/collect/memory.ts

@@ -1,30 +0,0 @@
-import { readProcFile, parseKb } from "../lib/parse.js";
-import type { MemoryInfo } from "../lib/types.js";
-
-export async function collectMemory(): Promise<MemoryInfo> {
-  const raw = readProcFile("/proc/meminfo") || "";
-  const kv: Record<string, string> = {};
-  for (const line of raw.split("\n")) {
-    const match = line.match(/^(\w+):\s+(.+)/);
-    if (match) kv[match[1]] = match[2];
-  }
-
-  const totalKb = parseKb(kv["MemTotal"]);
-  const availableKb = parseKb(kv["MemAvailable"]);
-  const swapTotalKb = parseKb(kv["SwapTotal"]);
-  const swapFreeKb = parseKb(kv["SwapFree"]);
-
-  const totalMb = Math.round(totalKb / 1024);
-  const availableMb = Math.round(availableKb / 1024);
-  const usedMb = totalMb - availableMb;
-  const swapTotalMb = Math.round(swapTotalKb / 1024);
-  const swapUsedMb = Math.round((swapTotalKb - swapFreeKb) / 1024);
-
-  return {
-    total_mb: totalMb,
-    used_mb: usedMb,
-    available_mb: availableMb,
-    swap_total_mb: swapTotalMb,
-    swap_used_mb: swapUsedMb,
-  };
-}

package/src/collect/network.ts

@@ -1,193 +0,0 @@
-import { readProcFile, sleep } from "../lib/parse.js";
-import { readFileSync, readdirSync } from "fs";
-import type { NetworkInfo } from "../lib/types.js";
-
-interface IfaceStats {
-  rx_bytes: number; rx_packets: number; rx_errors: number; rx_drops: number;
-  tx_bytes: number; tx_packets: number; tx_errors: number; tx_drops: number;
-}
-
-// Previous cumulative counters for delta computation (persists in process memory across cycles)
-interface PreviousCounters {
-  rx_errors: number;
-  tx_errors: number;
-  rx_drops: number;
-  tx_drops: number;
-  rx_packets: number;
-  tx_packets: number;
-  rx_crc_errors?: number;
-  rx_frame_errors?: number;
-  rx_length_errors?: number;
-  tx_carrier_errors?: number;
-}
-
-const previousCounters = new Map<string, PreviousCounters>();
-
-function readStatCounter(iface: string, name: string): number | undefined {
-  try {
-    const raw = readFileSync(`/sys/class/net/${iface}/statistics/${name}`, "utf-8").trim();
-    const val = parseInt(raw, 10);
-    return Number.isFinite(val) ? val : undefined;
-  } catch {
-    return undefined;
-  }
-}
-
-function parseNetDev(): Record<string, IfaceStats> {
-  const raw = readProcFile("/proc/net/dev") || "";
-  const result: Record<string, IfaceStats> = {};
-  for (const line of raw.split("\n").slice(2)) {
-    const match = line.match(/^\s*(\S+):\s+(.*)/);
-    if (!match) continue;
-    const name = match[1];
-    // Skip virtual interfaces
-    if (name === "lo" || name.startsWith("veth") || name.startsWith("docker") || name.startsWith("br-") || name.startsWith("virbr")) continue;
-    const parts = match[2].trim().split(/\s+/).map(Number);
-    result[name] = {
-      rx_bytes: parts[0] || 0, rx_packets: parts[1] || 0, rx_errors: parts[2] || 0, rx_drops: parts[3] || 0,
-      tx_bytes: parts[8] || 0, tx_packets: parts[9] || 0, tx_errors: parts[10] || 0, tx_drops: parts[11] || 0,
-    };
-  }
-  return result;
-}
-
-function getSpeed(iface: string): number {
-  try {
-    const speed = readFileSync(`/sys/class/net/${iface}/speed`, "utf-8").trim();
-    const val = parseInt(speed, 10);
-    return isNaN(val) || val <= 0 ? 0 : val;
-  } catch {
-    return 0;
-  }
-}
-
-function getOperstate(iface: string): string {
-  try {
-    return readFileSync(`/sys/class/net/${iface}/operstate`, "utf-8").trim();
-  } catch {
-    return "unknown";
-  }
-}
-
-function getBondMaster(iface: string): string | undefined {
-  try {
-    const bonds = readdirSync("/proc/net/bonding/");
-    for (const bond of bonds) {
-      const content = readFileSync(`/proc/net/bonding/${bond}`, "utf-8");
-      if (content.includes(`Slave Interface: ${iface}`)) return bond;
-    }
-  } catch {
-    // No bonds or /proc/net/bonding doesn't exist
-  }
-  return undefined;
-}
-
-function isBondMaster(iface: string): boolean {
-  try {
-    return readdirSync("/proc/net/bonding/").includes(iface);
-  } catch {
-    return false;
-  }
-}
-
-// Compute delta, handling counter wraps (current < previous means reset, use current as delta)
-function delta(current: number, previous: number): number {
-  if (current >= previous) return current - previous;
-  return current; // counter wrapped or reset
-}
-
-export async function collectNetwork(): Promise<NetworkInfo[]> {
-  const stats1 = parseNetDev();
-  await sleep(1000);
-  const stats2 = parseNetDev();
-
-  const currentIfaces = new Set<string>();
-  const results: NetworkInfo[] = [];
-
-  for (const [name, s2] of Object.entries(stats2)) {
-    const s1 = stats1[name];
-    if (!s1) continue;
-    currentIfaces.add(name);
-
-    const prev = previousCounters.get(name);
-
-    // /sys/class/net/*/statistics/ exposes finer-grained RX/TX subtype
-    // counters than /proc/net/dev. Read cumulative values here; delta is
-    // derived below against the previous cycle's snapshot.
-    const rxCrcCum = readStatCounter(name, "rx_crc_errors");
-    const rxFrameCum = readStatCounter(name, "rx_frame_errors");
-    const rxLenCum = readStatCounter(name, "rx_length_errors");
-    const txCarrierCum = readStatCounter(name, "tx_carrier_errors");
-
-    // Compute error/drop deltas (0 on first cycle after start or new interface)
-    let rxErrorsDelta = 0;
-    let txErrorsDelta = 0;
-    let rxDropsDelta = 0;
-    let txDropsDelta = 0;
-    let rxPacketsDelta = 0;
-    let txPacketsDelta = 0;
-    let rxCrcDelta: number | undefined;
-    let rxFrameDelta: number | undefined;
-    let rxLenDelta: number | undefined;
-    let txCarrierDelta: number | undefined;
-
-    if (prev) {
-      rxErrorsDelta = delta(s2.rx_errors, prev.rx_errors);
-      txErrorsDelta = delta(s2.tx_errors, prev.tx_errors);
-      rxDropsDelta = delta(s2.rx_drops, prev.rx_drops);
-      txDropsDelta = delta(s2.tx_drops, prev.tx_drops);
-      rxPacketsDelta = delta(s2.rx_packets, prev.rx_packets);
-      txPacketsDelta = delta(s2.tx_packets, prev.tx_packets);
-      if (rxCrcCum != null && prev.rx_crc_errors != null) rxCrcDelta = delta(rxCrcCum, prev.rx_crc_errors);
-      if (rxFrameCum != null && prev.rx_frame_errors != null) rxFrameDelta = delta(rxFrameCum, prev.rx_frame_errors);
-      if (rxLenCum != null && prev.rx_length_errors != null) rxLenDelta = delta(rxLenCum, prev.rx_length_errors);
-      if (txCarrierCum != null && prev.tx_carrier_errors != null) txCarrierDelta = delta(txCarrierCum, prev.tx_carrier_errors);
-    }
-
-    // Store current cumulative values for next cycle
-    previousCounters.set(name, {
-      rx_errors: s2.rx_errors,
-      tx_errors: s2.tx_errors,
-      rx_drops: s2.rx_drops,
-      tx_drops: s2.tx_drops,
-      rx_packets: s2.rx_packets,
-      tx_packets: s2.tx_packets,
-      rx_crc_errors: rxCrcCum,
-      rx_frame_errors: rxFrameCum,
-      rx_length_errors: rxLenCum,
-      tx_carrier_errors: txCarrierCum,
-    });
-
-    const entry: NetworkInfo = {
-      interface: name,
-      speed_mbps: getSpeed(name),
-      rx_bytes_sec: s2.rx_bytes - s1.rx_bytes, // already a 1-second delta
-      tx_bytes_sec: s2.tx_bytes - s1.tx_bytes,
-      rx_errors: rxErrorsDelta,
-      tx_errors: txErrorsDelta,
-      rx_drops: rxDropsDelta,
-      tx_drops: txDropsDelta,
-      rx_packets: rxPacketsDelta,
-      tx_packets: txPacketsDelta,
-      operstate: getOperstate(name),
-    };
-    if (rxCrcDelta !== undefined) entry.rx_crc_errors = rxCrcDelta;
-    if (rxFrameDelta !== undefined) entry.rx_frame_errors = rxFrameDelta;
-    if (rxLenDelta !== undefined) entry.rx_length_errors = rxLenDelta;
-    if (txCarrierDelta !== undefined) entry.tx_carrier_errors = txCarrierDelta;
-    const master = getBondMaster(name);
-    if (master) entry.bond_master = master;
-    // Identify bond masters (have at least one slave pointing at them).
-    if (isBondMaster(name)) entry.is_bond_master = true;
-    results.push(entry);
-  }
-
-  // Remove stale interfaces that disappeared
-  for (const name of previousCounters.keys()) {
-    if (!currentIfaces.has(name)) {
-      previousCounters.delete(name);
-    }
-  }
-
-  return results;
-}