@appkit/llamacpp-cli 1.4.1 → 1.6.0

package/dist/lib/history-manager.js

@@ -0,0 +1,157 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.HistoryManager = void 0;
+const promises_1 = require("fs/promises");
+const path_1 = require("path");
+const os_1 = require("os");
+const history_types_js_1 = require("../types/history-types.js");
+class HistoryManager {
+    constructor(serverId) {
+        this.MAX_AGE_MS = 24 * 60 * 60 * 1000; // 24 hours
+        this.serverId = serverId;
+        this.historyDir = (0, path_1.join)((0, os_1.homedir)(), '.llamacpp', 'history');
+        this.historyPath = (0, path_1.join)(this.historyDir, `${serverId}.json`);
+    }
+    /**
+     * Append a new snapshot to history (with auto-pruning)
+     */
+    async appendSnapshot(serverMetrics, systemMetrics) {
+        try {
+            // Ensure history directory exists
+            await (0, promises_1.mkdir)(this.historyDir, { recursive: true });
+            // Load existing history
+            const historyData = await this.loadHistoryData();
+            // Create new snapshot
+            const snapshot = {
+                timestamp: Date.now(),
+                server: {
+                    healthy: serverMetrics.healthy,
+                    uptime: serverMetrics.uptime,
+                    activeSlots: serverMetrics.activeSlots,
+                    idleSlots: serverMetrics.idleSlots,
+                    totalSlots: serverMetrics.totalSlots,
+                    avgPromptSpeed: serverMetrics.avgPromptSpeed,
+                    avgGenerateSpeed: serverMetrics.avgGenerateSpeed,
+                    processMemory: serverMetrics.processMemory,
+                    processCpuUsage: serverMetrics.processCpuUsage,
+                },
+                system: systemMetrics ? {
+                    gpuUsage: systemMetrics.gpuUsage,
+                    cpuUsage: systemMetrics.cpuUsage,
+                    aneUsage: systemMetrics.aneUsage,
+                    temperature: systemMetrics.temperature,
+                    memoryUsed: systemMetrics.memoryUsed,
+                    memoryTotal: systemMetrics.memoryTotal,
+                } : undefined,
+            };
+            // Append new snapshot
+            historyData.snapshots.push(snapshot);
+            // Prune old snapshots (keep only last 24h)
+            historyData.snapshots = this.pruneOldSnapshots(historyData.snapshots, this.MAX_AGE_MS);
+            // Atomic write: write to temp file in same directory, then rename
+            // This prevents read collisions during concurrent access
+            // IMPORTANT: temp file MUST be in same directory as destination for rename to work across filesystems
+            const tempPath = (0, path_1.join)(this.historyDir, `.${this.serverId}-${Date.now()}.tmp`);
+            await (0, promises_1.writeFile)(tempPath, JSON.stringify(historyData, null, 2), 'utf-8');
+            await (0, promises_1.rename)(tempPath, this.historyPath);
+        }
+        catch (error) {
+            // Silent failure - don't interrupt monitoring
+            // Don't throw - just return silently to avoid polluting console
+            return;
+        }
+    }
+    /**
+     * Load all snapshots within specified time window
+     */
+    async loadHistory(windowHours) {
+        // Retry logic for file I/O collisions during concurrent read/write
+        const maxRetries = 3;
+        let lastError = null;
+        for (let attempt = 0; attempt < maxRetries; attempt++) {
+            try {
+                const historyData = await this.loadHistoryData();
+                return this.filterByTimeWindow(historyData.snapshots, windowHours);
+            }
+            catch (error) {
+                lastError = error;
+                // Wait briefly before retry (exponential backoff)
+                if (attempt < maxRetries - 1) {
+                    await new Promise(resolve => setTimeout(resolve, 50 * Math.pow(2, attempt)));
+                }
+            }
+        }
+        // All retries failed - throw error so it can be handled upstream
+        throw new Error(`Failed to load history after ${maxRetries} attempts: ${lastError?.message || 'Unknown error'}`);
+    }
+    /**
+     * Load history for specific time window type
+     */
+    async loadHistoryByWindow(window) {
+        return this.loadHistory(history_types_js_1.TIME_WINDOW_HOURS[window]);
+    }
+    /**
+     * Get file path for server history
+     */
+    getHistoryPath() {
+        return this.historyPath;
+    }
+    /**
+     * Check if history file exists
+     */
+    async hasHistory() {
+        try {
+            await (0, promises_1.access)(this.historyPath);
+            return true;
+        }
+        catch {
+            return false;
+        }
+    }
+    /**
+     * Clear all history for server
+     */
+    async clearHistory() {
+        const emptyHistory = {
+            serverId: this.serverId,
+            snapshots: [],
+        };
+        await (0, promises_1.mkdir)(this.historyDir, { recursive: true });
+        // Atomic write - temp file in same directory as destination
+        const tempPath = (0, path_1.join)(this.historyDir, `.${this.serverId}-${Date.now()}.tmp`);
+        await (0, promises_1.writeFile)(tempPath, JSON.stringify(emptyHistory, null, 2), 'utf-8');
+        await (0, promises_1.rename)(tempPath, this.historyPath);
+    }
+    /**
+     * Load full history data from file
+     */
+    async loadHistoryData() {
+        try {
+            const content = await (0, promises_1.readFile)(this.historyPath, 'utf-8');
+            return JSON.parse(content);
+        }
+        catch (error) {
+            // File doesn't exist or is corrupted, return empty history
+            return {
+                serverId: this.serverId,
+                snapshots: [],
+            };
+        }
+    }
+    /**
+     * Prune snapshots older than maxAge
+     */
+    pruneOldSnapshots(snapshots, maxAgeMs) {
+        const cutoff = Date.now() - maxAgeMs;
+        return snapshots.filter(s => s.timestamp >= cutoff);
+    }
+    /**
+     * Filter snapshots by time window
+     */
+    filterByTimeWindow(snapshots, windowHours) {
+        const cutoff = Date.now() - (windowHours * 60 * 60 * 1000);
+        return snapshots.filter(s => s.timestamp >= cutoff);
+    }
+}
+exports.HistoryManager = HistoryManager;
+//# sourceMappingURL=history-manager.js.map

package/dist/lib/history-manager.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"history-manager.js","sourceRoot":"","sources":["../../src/lib/history-manager.ts"],"names":[],"mappings":";;;AAAA,0CAAyE;AACzE,+BAA4B;AAC5B,2BAA6B;AAE7B,gEAAwG;AAExG,MAAa,cAAc;IAMzB,YAAY,QAAgB;QAFX,eAAU,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,WAAW;QAG5D,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,UAAU,GAAG,IAAA,WAAI,EAAC,IAAA,YAAO,GAAE,EAAE,WAAW,EAAE,SAAS,CAAC,CAAC;QAC1D,IAAI,CAAC,WAAW,GAAG,IAAA,WAAI,EAAC,IAAI,CAAC,UAAU,EAAE,GAAG,QAAQ,OAAO,CAAC,CAAC;IAC/D,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,cAAc,CAAC,aAA4B,EAAE,aAA6B;QAC9E,IAAI,CAAC;YACH,kCAAkC;YAClC,MAAM,IAAA,gBAAK,EAAC,IAAI,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAElD,wBAAwB;YACxB,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,eAAe,EAAE,CAAC;YAEjD,sBAAsB;YACtB,MAAM,QAAQ,GAAoB;gBAChC,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;gBACrB,MAAM,EAAE;oBACN,OAAO,EAAE,aAAa,CAAC,OAAO;oBAC9B,MAAM,EAAE,aAAa,CAAC,MAAM;oBAC5B,WAAW,EAAE,aAAa,CAAC,WAAW;oBACtC,SAAS,EAAE,aAAa,CAAC,SAAS;oBAClC,UAAU,EAAE,aAAa,CAAC,UAAU;oBACpC,cAAc,EAAE,aAAa,CAAC,cAAc;oBAC5C,gBAAgB,EAAE,aAAa,CAAC,gBAAgB;oBAChD,aAAa,EAAE,aAAa,CAAC,aAAa;oBAC1C,eAAe,EAAE,aAAa,CAAC,eAAe;iBAC/C;gBACD,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;oBACtB,QAAQ,EAAE,aAAa,CAAC,QAAQ;oBAChC,QAAQ,EAAE,aAAa,CAAC,QAAQ;oBAChC,QAAQ,EAAE,aAAa,CAAC,QAAQ;oBAChC,WAAW,EAAE,aAAa,CAAC,WAAW;oBACtC,UAAU,EAAE,aAAa,CAAC,UAAU;oBACpC,WAAW,EAAE,aAAa,CAAC,WAAW;iBACvC,CAAC,CAAC,CAAC,SAAS;aACd,CAAC;YAEF,sBAAsB;YACtB,WAAW,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAErC,2CAA2C;YAC3C,WAAW,CAAC,SAAS,GAAG,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,SAAS,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC;YAEvF,kEAAkE;YAClE,yDAAyD;YACzD,sGAAsG;YACtG,MAAM,QAAQ,GAAG,IAAA,WAAI,EAAC,IAAI,CAAC,UAAU,EAAE,IAAI,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YAC9E,MAAM,IAAA,oBAAS,EAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;YACzE,MAAM,IAAA,iBAAM,EAAC,QAAQ,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC;QAC3C,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,8CAA8C;YAC9C,gEAAgE;YAChE,OAAO;QACT,CAAC;IACH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,WAAW,CAAC,WAAmB;QACnC,mEAAmE;QACnE,MAAM,UAAU,GAAG,CAAC,CAAC;QACrB,IAAI,SAAS,GAAiB,IAAI,CAAC;QAEnC,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;YACtD,IAAI,CAAC;gBACH,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,eAAe,EAAE,CAAC;gBACjD,OAAO,IAAI,CAAC,kBAAkB,CAAC,WAAW,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC;YACrE,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,SAAS,GAAG,KAAc,CAAC;gBAC3B,kDAAkD;gBAClD,IAAI,OAAO,GAAG,UAAU,GAAG,CAAC,EAAE,CAAC;oBAC7B,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;gBAC/E,CAAC;YACH,CAAC;QACH,CAAC;QAED,iEAAiE;QACjE,MAAM,IAAI,KAAK,CAAC,gCAAgC,UAAU,cAAc,SAAS,EAAE,OAAO,IAAI,eAAe,EAAE,CAAC,CAAC;IACnH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,mBAAmB,CAAC,MAAkB;QAC1C,OAAO,IAAI,CAAC,WAAW,CAAC,oCAAiB,CAAC,MAAM,CAAC,CAAC,CAAC;IACrD,CAAC;IAED;;OAEG;IACH,cAAc;QACZ,OAAO,IAAI,CAAC,WAAW,CAAC;IAC1B,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,UAAU;QACd,IAAI,CAAC;YACH,MAAM,IAAA,iBAAM,EAAC,IAAI,CAAC,WAAW,CAAC,CAAC;YAC/B,OAAO,IAAI,CAAC;QACd,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,YAAY;QAChB,MAAM,YAAY,GAAgB;YAChC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,SAAS,EAAE,EAAE;SACd,CAAC;QAEF,MAAM,IAAA,gBAAK,EAAC,IAAI,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAElD,4DAA4D;QAC5D,MAAM,QAAQ,GAAG,IAAA,WAAI,EAAC,IAAI,CAAC,UAAU,EAAE,IAAI,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC9E,MAAM,IAAA,oBAAS,EAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;QAC1E,MAAM,IAAA,iBAAM,EAAC,QAAQ,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC;IAC3C,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,eAAe;QAC3B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,IAAA,mBAAQ,EAAC,IAAI,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;YAC1D,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAgB,CAAC;QAC5C,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,2DAA2D;YAC3D,OAAO;gBACL,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,SAAS,EAAE,EAAE;aACd,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;OAEG;IACK,iBAAiB,CAAC,SAA4B,EAAE,QAAgB;QACtE,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,CAAC;QACrC,OAAO,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,IAAI,MAAM,CAAC,CAAC;IACtD,CAAC;IAED;;OAEG;IACK,kBAAkB,CAAC,SAA4B,EAAE,WAAmB;QAC1E,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,WAAW,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC;QAC3D,OAAO,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,IAAI,MAAM,CAAC,CAAC;IACtD,CAAC;CACF;AArKD,wCAqKC"}

package/dist/lib/metrics-aggregator.d.ts

@@ -0,0 +1,40 @@
+import { ServerConfig } from '../types/server-config.js';
+import { ServerMetrics, MonitorData } from '../types/monitor-types.js';
+/**
+ * Aggregates metrics from llama.cpp server API endpoints
+ * Combines server health, slot status, and model properties
+ */
+export declare class MetricsAggregator {
+    private serverUrl;
+    private timeout;
+    private previousSlots;
+    constructor(server: ServerConfig, timeout?: number);
+    /**
+     * Fetch data from llama.cpp API with timeout
+     */
+    private fetchWithTimeout;
+    /**
+     * Get server health status
+     */
+    private getHealth;
+    /**
+     * Get server properties (model info, context size, etc.)
+     */
+    private getProps;
+    /**
+     * Get active slots information with calculated tok/s
+     */
+    private getSlots;
+    /**
+     * Aggregate all server metrics
+     * @param server - Server configuration
+     * @param processMemory - Optional pre-fetched process memory (for batch collection)
+     * @param processCpuUsage - Optional pre-fetched process CPU usage (for batch collection)
+     */
+    collectServerMetrics(server: ServerConfig, processMemory?: number | null, processCpuUsage?: number | null): Promise<ServerMetrics>;
+    /**
+     * Collect complete monitoring data (server + system metrics)
+     */
+    collectMonitorData(server: ServerConfig, updateInterval?: number): Promise<MonitorData>;
+}
+//# sourceMappingURL=metrics-aggregator.d.ts.map

package/dist/lib/metrics-aggregator.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"metrics-aggregator.d.ts","sourceRoot":"","sources":["../../src/lib/metrics-aggregator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAC;AACzD,OAAO,EAAE,aAAa,EAAY,WAAW,EAAE,MAAM,2BAA2B,CAAC;AAKjF;;;GAGG;AACH,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,aAAa,CAAoE;gBAE7E,MAAM,EAAE,YAAY,EAAE,OAAO,GAAE,MAAa;IAOxD;;OAEG;YACW,gBAAgB;IA0B9B;;OAEG;YACW,SAAS;IAKvB;;OAEG;YACW,QAAQ;IAItB;;OAEG;YACW,QAAQ;IAyDtB;;;;;OAKG;IACG,oBAAoB,CACxB,MAAM,EAAE,YAAY,EACpB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,EAC7B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,GAC9B,OAAO,CAAC,aAAa,CAAC;IAmGzB;;OAEG;IACG,kBAAkB,CACtB,MAAM,EAAE,YAAY,EACpB,cAAc,GAAE,MAAa,GAC5B,OAAO,CAAC,WAAW,CAAC;CAexB"}

package/dist/lib/metrics-aggregator.js

@@ -0,0 +1,211 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.MetricsAggregator = void 0;
+const status_checker_js_1 = require("./status-checker.js");
+const system_collector_js_1 = require("./system-collector.js");
+const process_utils_js_1 = require("../utils/process-utils.js");
+/**
+ * Aggregates metrics from llama.cpp server API endpoints
+ * Combines server health, slot status, and model properties
+ */
+class MetricsAggregator {
+    constructor(server, timeout = 5000) {
+        this.previousSlots = new Map();
+        // Handle null host (legacy configs) by defaulting to 127.0.0.1
+        const host = server.host || '127.0.0.1';
+        this.serverUrl = `http://${host}:${server.port}`;
+        this.timeout = timeout;
+    }
+    /**
+     * Fetch data from llama.cpp API with timeout
+     */
+    async fetchWithTimeout(endpoint, customTimeout) {
+        try {
+            const controller = new AbortController();
+            const timeoutMs = customTimeout ?? this.timeout;
+            const timeoutId = setTimeout(() => controller.abort(), timeoutMs);
+            const response = await fetch(`${this.serverUrl}${endpoint}`, {
+                signal: controller.signal,
+            });
+            clearTimeout(timeoutId);
+            if (!response.ok) {
+                return null;
+            }
+            return await response.json();
+        }
+        catch (err) {
+            // Network error, timeout, or parse error
+            return null;
+        }
+    }
+    /**
+     * Get server health status
+     */
+    async getHealth() {
+        const health = await this.fetchWithTimeout('/health');
+        return health !== null && health.status === 'ok';
+    }
+    /**
+     * Get server properties (model info, context size, etc.)
+     */
+    async getProps() {
+        return await this.fetchWithTimeout('/props');
+    }
+    /**
+     * Get active slots information with calculated tok/s
+     */
+    async getSlots() {
+        const data = await this.fetchWithTimeout('/slots');
+        if (!data || !Array.isArray(data)) {
+            return [];
+        }
+        const now = Date.now();
+        return data.map((slot) => {
+            const slotId = slot.id;
+            const n_decoded = slot.next_token?.[0]?.n_decoded || 0;
+            const isProcessing = slot.is_processing;
+            // Calculate tokens per second by comparing with previous poll
+            let predicted_per_second;
+            if (isProcessing && n_decoded > 0) {
+                const previous = this.previousSlots.get(slotId);
+                if (previous && previous.n_decoded < n_decoded) {
+                    const tokensGenerated = n_decoded - previous.n_decoded;
+                    const timeElapsed = (now - previous.timestamp) / 1000; // Convert to seconds
+                    if (timeElapsed > 0) {
+                        predicted_per_second = tokensGenerated / timeElapsed;
+                    }
+                }
+                // Store current state for next comparison
+                this.previousSlots.set(slotId, { n_decoded, timestamp: now });
+            }
+            else if (!isProcessing) {
+                // Clear history when slot becomes idle
+                this.previousSlots.delete(slotId);
+            }
+            return {
+                id: slotId,
+                state: isProcessing ? 'processing' : 'idle',
+                n_prompt_tokens: slot.n_prompt_tokens,
+                n_decoded,
+                n_ctx: slot.n_ctx || 0,
+                timings: predicted_per_second
+                    ? {
+                        prompt_n: 0,
+                        prompt_ms: 0,
+                        prompt_per_token_ms: 0,
+                        prompt_per_second: 0,
+                        predicted_n: n_decoded,
+                        predicted_ms: 0,
+                        predicted_per_token_ms: 0,
+                        predicted_per_second,
+                    }
+                    : undefined,
+            };
+        });
+    }
+    /**
+     * Aggregate all server metrics
+     * @param server - Server configuration
+     * @param processMemory - Optional pre-fetched process memory (for batch collection)
+     * @param processCpuUsage - Optional pre-fetched process CPU usage (for batch collection)
+     */
+    async collectServerMetrics(server, processMemory, processCpuUsage) {
+        const now = Date.now();
+        // Check basic server status first
+        const status = await status_checker_js_1.statusChecker.checkServer(server);
+        // Calculate uptime if server is running and has lastStarted
+        let uptime;
+        if (status.isRunning && server.lastStarted) {
+            const startTime = new Date(server.lastStarted).getTime();
+            const uptimeSeconds = Math.floor((now - startTime) / 1000);
+            const hours = Math.floor(uptimeSeconds / 3600);
+            const minutes = Math.floor((uptimeSeconds % 3600) / 60);
+            const seconds = uptimeSeconds % 60;
+            uptime = `${hours}h ${minutes}m ${seconds}s`;
+        }
+        // If server not running, return minimal data
+        if (!status.isRunning) {
+            return {
+                server,
+                healthy: false,
+                modelLoaded: false,
+                modelName: server.modelName,
+                contextSize: server.ctxSize,
+                totalSlots: 0,
+                activeSlots: 0,
+                idleSlots: 0,
+                slots: [],
+                timestamp: now,
+                stale: false,
+            };
+        }
+        // Fetch detailed metrics in parallel
+        // If processMemory/CPU were pre-fetched (batch mode), use them; otherwise fetch individually
+        const [healthy, props, slots, fetchedMemory, fetchedCpu] = await Promise.all([
+            this.getHealth(),
+            this.getProps(),
+            this.getSlots(),
+            processMemory !== undefined
+                ? Promise.resolve(processMemory)
+                : (server.pid ? (0, process_utils_js_1.getProcessMemory)(server.pid) : Promise.resolve(null)),
+            processCpuUsage !== undefined
+                ? Promise.resolve(processCpuUsage)
+                : (server.pid ? (0, process_utils_js_1.getProcessCpu)(server.pid) : Promise.resolve(null)),
+        ]);
+        // Calculate slot statistics
+        const activeSlots = slots.filter((s) => s.state === 'processing').length;
+        const idleSlots = slots.filter((s) => s.state === 'idle').length;
+        const totalSlots = props?.total_slots || slots.length;
+        // Calculate average speeds (only from processing slots)
+        const processingSlots = slots.filter((s) => s.state === 'processing' && s.timings);
+        const avgPromptSpeed = processingSlots.length > 0
+            ? processingSlots.reduce((sum, s) => sum + (s.timings?.prompt_per_second || 0), 0) / processingSlots.length
+            : undefined;
+        const avgGenerateSpeed = processingSlots.length > 0
+            ? processingSlots.reduce((sum, s) => sum + (s.timings?.predicted_per_second || 0), 0) / processingSlots.length
+            : undefined;
+        // Calculate total memory (CPU + Metal GPU memory if available)
+        let totalMemory = fetchedMemory ?? undefined;
+        if (totalMemory !== undefined && server.metalMemoryMB) {
+            // Add Metal memory (convert MB to bytes)
+            totalMemory += server.metalMemoryMB * 1024 * 1024;
+        }
+        return {
+            server,
+            healthy,
+            uptime,
+            modelLoaded: props !== null,
+            modelName: server.modelName,
+            contextSize: props?.default_generation_settings?.n_ctx || server.ctxSize,
+            totalSlots,
+            activeSlots,
+            idleSlots,
+            slots,
+            avgPromptSpeed,
+            avgGenerateSpeed,
+            processMemory: totalMemory,
+            processCpuUsage: fetchedCpu ?? undefined,
+            timestamp: now,
+            stale: false,
+        };
+    }
+    /**
+     * Collect complete monitoring data (server + system metrics)
+     */
+    async collectMonitorData(server, updateInterval = 2000) {
+        // Collect server and system metrics in parallel
+        const [serverMetrics, systemMetrics] = await Promise.all([
+            this.collectServerMetrics(server),
+            system_collector_js_1.systemCollector.collectSystemMetrics(),
+        ]);
+        return {
+            server: serverMetrics,
+            system: systemMetrics,
+            lastUpdated: new Date(),
+            updateInterval,
+            consecutiveFailures: 0,
+        };
+    }
+}
+exports.MetricsAggregator = MetricsAggregator;
+//# sourceMappingURL=metrics-aggregator.js.map

package/dist/lib/metrics-aggregator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"metrics-aggregator.js","sourceRoot":"","sources":["../../src/lib/metrics-aggregator.ts"],"names":[],"mappings":";;;AAEA,2DAAoD;AACpD,+DAAwD;AACxD,gEAA4E;AAE5E;;;GAGG;AACH,MAAa,iBAAiB;IAK5B,YAAY,MAAoB,EAAE,UAAkB,IAAI;QAFhD,kBAAa,GAA0D,IAAI,GAAG,EAAE,CAAC;QAGvF,+DAA+D;QAC/D,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,IAAI,WAAW,CAAC;QACxC,IAAI,CAAC,SAAS,GAAG,UAAU,IAAI,IAAI,MAAM,CAAC,IAAI,EAAE,CAAC;QACjD,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IACzB,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,gBAAgB,CAC5B,QAAgB,EAChB,aAAsB;QAEtB,IAAI,CAAC;YACH,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;YACzC,MAAM,SAAS,GAAG,aAAa,IAAI,IAAI,CAAC,OAAO,CAAC;YAChD,MAAM,SAAS,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;YAElE,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,GAAG,QAAQ,EAAE,EAAE;gBAC3D,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YAEH,YAAY,CAAC,SAAS,CAAC,CAAC;YAExB,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,OAAO,IAAI,CAAC;YACd,CAAC;YAED,OAAO,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QAC/B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,yCAAyC;YACzC,OAAO,IAAI,CAAC;QACd,CAAC;IACH,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,SAAS;QACrB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,SAAS,CAAC,CAAC;QACtD,OAAO,MAAM,KAAK,IAAI,IAAI,MAAM,CAAC,MAAM,KAAK,IAAI,CAAC;IACnD,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,QAAQ;QACpB,OAAO,MAAM,IAAI,CAAC,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IAC/C,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,QAAQ;QACpB,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,QAAQ,CAAC,CAAC;QACnD,IAAI,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC;YAClC,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAEvB,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,IAAS,EAAE,EAAE;YAC5B,MAAM,MAAM,GAAG,IAAI,CAAC,EAAE,CAAC;YACvB,MAAM,SAAS,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,SAAS,IAAI,CAAC,CAAC;YACvD,MAAM,YAAY,GAAG,IAAI,CAAC,aAAa,CAAC;YAExC,8DAA8D;YAC9D,IAAI,oBAAwC,CAAC;YAE7C,IAAI,YAAY,IAAI,SAAS,GAAG,CAAC,EAAE,CAAC;gBAClC,MAAM,QAAQ,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;gBAEhD,IAAI,QAAQ,IAAI,QAAQ,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;oBAC/C,MAAM,eAAe,GAAG,SAAS,GAAG,QAAQ,CAAC,SAAS,CAAC;oBACvD,MAAM,WAAW,GAAG,CAAC,GAAG,GAAG,QAAQ,CAAC,SAAS,CAAC,GAAG,IAAI,CAAC,CAAC,qBAAqB;oBAE5E,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;wBACpB,oBAAoB,GAAG,eAAe,GAAG,WAAW,CAAC;oBACvD,CAAC;gBACH,CAAC;gBAED,0CAA0C;gBAC1C,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;YAChE,CAAC;iBAAM,IAAI,CAAC,YAAY,EAAE,CAAC;gBACzB,uCAAuC;gBACvC,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YACpC,CAAC;YAED,OAAO;gBACL,EAAE,EAAE,MAAM;gBACV,KAAK,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,MAAM;gBAC3C,eAAe,EAAE,IAAI,CAAC,eAAe;gBACrC,SAAS;gBACT,KAAK,EAAE,IAAI,CAAC,KAAK,IAAI,CAAC;gBACtB,OAAO,EAAE,oBAAoB;oBAC3B,CAAC,CAAC;wBACE,QAAQ,EAAE,CAAC;wBACX,SAAS,EAAE,CAAC;wBACZ,mBAAmB,EAAE,CAAC;wBACtB,iBAAiB,EAAE,CAAC;wBACpB,WAAW,EAAE,SAAS;wBACtB,YAAY,EAAE,CAAC;wBACf,sBAAsB,EAAE,CAAC;wBACzB,oBAAoB;qBACrB;oBACH,CAAC,CAAC,SAAS;aACd,CAAC;QACJ,CAAC,CAAC,CAAC;IACL,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,oBAAoB,CACxB,MAAoB,EACpB,aAA6B,EAC7B,eAA+B;QAE/B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAEvB,kCAAkC;QAClC,MAAM,MAAM,GAAG,MAAM,iCAAa,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;QAEvD,4DAA4D;QAC5D,IAAI,MAA0B,CAAC;QAC/B,IAAI,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,WAAW,EAAE,CAAC;YAC3C,MAAM,SAAS,GAAG,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,OAAO,EAAE,CAAC;YACzD,MAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,GAAG,SAAS,CAAC,GAAG,IAAI,CAAC,CAAC;YAC3D,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,IAAI,CAAC,CAAC;YAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,aAAa,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;YACxD,MAAM,OAAO,GAAG,aAAa,GAAG,EAAE,CAAC;YACnC,MAAM,GAAG,GAAG,KAAK,KAAK,OAAO,KAAK,OAAO,GAAG,CAAC;QAC/C,CAAC;QAED,6CAA6C;QAC7C,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;YACtB,OAAO;gBACL,MAAM;gBACN,OAAO,EAAE,KAAK;gBACd,WAAW,EAAE,KAAK;gBAClB,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,WAAW,EAAE,MAAM,CAAC,OAAO;gBAC3B,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,SAAS,EAAE,CAAC;gBACZ,KAAK,EAAE,EAAE;gBACT,SAAS,EAAE,GAAG;gBACd,KAAK,EAAE,KAAK;aACb,CAAC;QACJ,CAAC;QAED,qCAAqC;QACrC,6FAA6F;QAC7F,MAAM,CAAC,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,aAAa,EAAE,UAAU,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YAC3E,IAAI,CAAC,SAAS,EAAE;YAChB,IAAI,CAAC,QAAQ,EAAE;YACf,IAAI,CAAC,QAAQ,EAAE;YACf,aAAa,KAAK,SAAS;gBACzB,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC;gBAChC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,IAAA,mCAAgB,EAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YACvE,eAAe,KAAK,SAAS;gBAC3B,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,eAAe,CAAC;gBAClC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,IAAA,gCAAa,EAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;SACrE,CAAC,CAAC;QAEH,4BAA4B;QAC5B,MAAM,WAAW,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,YAAY,CAAC,CAAC,MAAM,CAAC;QACzE,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;QACjE,MAAM,UAAU,GAAG,KAAK,EAAE,WAAW,IAAI,KAAK,CAAC,MAAM,CAAC;QAEtD,wDAAwD;QACxD,MAAM,eAAe,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,YAAY,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC;QAEnF,MAAM,cAAc,GAClB,eAAe,CAAC,MAAM,GAAG,CAAC;YACxB,CAAC,CAAC,eAAe,CAAC,MAAM,CACpB,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,iBAAiB,IAAI,CAAC,CAAC,EACrD,CAAC,CACF,GAAG,eAAe,CAAC,MAAM;YAC5B,CAAC,CAAC,SAAS,CAAC;QAEhB,MAAM,gBAAgB,GACpB,eAAe,CAAC,MAAM,GAAG,CAAC;YACxB,CAAC,CAAC,eAAe,CAAC,MAAM,CACpB,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,oBAAoB,IAAI,CAAC,CAAC,EACxD,CAAC,CACF,GAAG,eAAe,CAAC,MAAM;YAC5B,CAAC,CAAC,SAAS,CAAC;QAEhB,+DAA+D;QAC/D,IAAI,WAAW,GAAG,aAAa,IAAI,SAAS,CAAC;QAC7C,IAAI,WAAW,KAAK,SAAS,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;YACtD,yCAAyC;YACzC,WAAW,IAAI,MAAM,CAAC,aAAa,GAAG,IAAI,GAAG,IAAI,CAAC;QACpD,CAAC;QAED,OAAO;YACL,MAAM;YACN,OAAO;YACP,MAAM;YACN,WAAW,EAAE,KAAK,KAAK,IAAI;YAC3B,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,WAAW,EAAE,KAAK,EAAE,2BAA2B,EAAE,KAAK,IAAI,MAAM,CAAC,OAAO;YACxE,UAAU;YACV,WAAW;YACX,SAAS;YACT,KAAK;YACL,cAAc;YACd,gBAAgB;YAChB,aAAa,EAAE,WAAW;YAC1B,eAAe,EAAE,UAAU,IAAI,SAAS;YACxC,SAAS,EAAE,GAAG;YACd,KAAK,EAAE,KAAK;SACb,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,kBAAkB,CACtB,MAAoB,EACpB,iBAAyB,IAAI;QAE7B,gDAAgD;QAChD,MAAM,CAAC,aAAa,EAAE,aAAa,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YACvD,IAAI,CAAC,oBAAoB,CAAC,MAAM,CAAC;YACjC,qCAAe,CAAC,oBAAoB,EAAE;SACvC,CAAC,CAAC;QAEH,OAAO;YACL,MAAM,EAAE,aAAa;YACrB,MAAM,EAAE,aAAa;YACrB,WAAW,EAAE,IAAI,IAAI,EAAE;YACvB,cAAc;YACd,mBAAmB,EAAE,CAAC;SACvB,CAAC;IACJ,CAAC;CACF;AAtPD,8CAsPC"}

package/dist/lib/system-collector.d.ts

@@ -0,0 +1,80 @@
+import { SystemMetrics } from '../types/monitor-types.js';
+/**
+ * System metrics collector using macmon (optional) and vm_stat (fallback)
+ * Provides GPU, CPU, ANE, and memory metrics on macOS
+ */
+export declare class SystemCollector {
+    private macmonPath;
+    private macmonAvailable;
+    private lastSystemMetrics;
+    private lastCollectionTime;
+    private readonly CACHE_TTL_MS;
+    private collectingLock;
+    private pCoreCount;
+    private eCoreCount;
+    private totalCores;
+    constructor(macmonPath?: string);
+    /**
+     * Get CPU core counts for weighted average calculation
+     */
+    private initializeCoreCount;
+    /**
+     * Check if macmon is available
+     */
+    private checkMacmonAvailability;
+    /**
+     * Parse macmon JSON output
+     * Expected format from 'macmon pipe':
+     * {
+     *   "gpu_usage": [count, percentage],
+     *   "pcpu_usage": [count, percentage],
+     *   "ecpu_usage": [count, percentage],
+     *   "ane_power": number,
+     *   "temp": {"cpu_temp_avg": number, "gpu_temp_avg": number}
+     * }
+     */
+    private parseMacmonJson;
+    /**
+     * Collect macmon metrics (GPU, CPU, ANE)
+     * Uses 'macmon pipe' which outputs one JSON line per update
+     * Spawns macmon, reads one line, and kills it to prevent process leaks
+     */
+    private getMacmonMetrics;
+    /**
+     * Parse vm_stat output for memory metrics
+     * Expected format:
+     * Pages free:                               123456.
+     * Pages active:                             234567.
+     * Pages inactive:                           345678.
+     * Pages speculative:                        45678.
+     * Pages throttled:                          0.
+     * Pages wired down:                         123456.
+     * Pages purgeable count:                    0.
+     * "Translation faults":                     12345678.
+     * Pages copy-on-write:                      123456.
+     * ...
+     */
+    private parseVmStatOutput;
+    /**
+     * Get total system memory from sysctl
+     * Returns installed RAM size in bytes
+     */
+    private getTotalMemory;
+    /**
+     * Collect vm_stat memory metrics + total system memory from sysctl
+     */
+    private getMemoryMetrics;
+    /**
+     * Collect all system metrics
+     * Attempts macmon first (GPU/CPU/ANE), always gets memory from vm_stat + sysctl
+     * Caches results for 4s to prevent spawning multiple macmon processes
+     */
+    collectSystemMetrics(): Promise<SystemMetrics>;
+    /**
+     * Internal method to actually collect system metrics
+     * Called by collectSystemMetrics with caching/locking
+     */
+    private doCollectSystemMetrics;
+}
+export declare const systemCollector: SystemCollector;
+//# sourceMappingURL=system-collector.d.ts.map

package/dist/lib/system-collector.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"system-collector.d.ts","sourceRoot":"","sources":["../../src/lib/system-collector.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,aAAa,EAAE,MAAM,2BAA2B,CAAC;AAE1D;;;GAGG;AACH,qBAAa,eAAe;IAC1B,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,eAAe,CAAwB;IAC/C,OAAO,CAAC,iBAAiB,CAA8B;IACvD,OAAO,CAAC,kBAAkB,CAAa;IACvC,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAQ;IACrC,OAAO,CAAC,cAAc,CAAuC;IAC7D,OAAO,CAAC,UAAU,CAAa;IAC/B,OAAO,CAAC,UAAU,CAAa;IAC/B,OAAO,CAAC,UAAU,CAAa;gBAEnB,UAAU,GAAE,MAAmC;IAK3D;;OAEG;YACW,mBAAmB;IA4BjC;;OAEG;YACW,uBAAuB;IAerC;;;;;;;;;;OAUG;IACH,OAAO,CAAC,eAAe;IAiDvB;;;;OAIG;YACW,gBAAgB;IA2B9B;;;;;;;;;;;;;OAaG;IACH,OAAO,CAAC,iBAAiB;IA6BzB;;;OAGG;YACW,cAAc;IAS5B;;OAEG;YACW,gBAAgB;IAmB9B;;;;OAIG;IACG,oBAAoB,IAAI,OAAO,CAAC,aAAa,CAAC;IA0BpD;;;OAGG;YACW,sBAAsB;CAkCrC;AAGD,eAAO,MAAM,eAAe,iBAAwB,CAAC"}