@sheepbun/yips 0.1.1 → 0.1.46

package/dist/llama-server.js

@@ -1,620 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.isLocalLlamaEndpoint = isLocalLlamaEndpoint;
-exports.checkLlamaHealth = checkLlamaHealth;
-exports.stopLlamaServer = stopLlamaServer;
-exports.resetLlamaForFreshSession = resetLlamaForFreshSession;
-exports.startLlamaServer = startLlamaServer;
-exports.ensureLlamaReady = ensureLlamaReady;
-exports.formatLlamaStartupFailure = formatLlamaStartupFailure;
-const node_child_process_1 = require("node:child_process");
-const node_fs_1 = require("node:fs");
-const promises_1 = require("node:fs/promises");
-const node_path_1 = require("node:path");
-const hardware_1 = require("./hardware");
-const HEALTH_PATH = "/health";
-const STARTUP_TIMEOUT_MS = 60_000;
-const HEALTH_TIMEOUT_MS = 2_000;
-const HEALTH_RETRY_INTERVAL_MS = 500;
-const STDERR_TAIL_LIMIT = 120;
-let runningState = null;
-const LOCAL_ENDPOINT_HOSTS = new Set([
-    "localhost",
-    "127.0.0.1",
-    "0.0.0.0",
-    "::1"
-]);
-function sleep(ms) {
-    return new Promise((resolveSleep) => {
-        setTimeout(resolveSleep, ms);
-    });
-}
-async function exists(path) {
-    try {
-        await (0, promises_1.access)(path, node_fs_1.constants.R_OK);
-        return true;
-    }
-    catch {
-        return false;
-    }
-}
-function toErrorMessage(error) {
-    return error instanceof Error ? error.message : String(error);
-}
-function resolveHost(config) {
-    const host = config.llamaHost.trim();
-    return host.length > 0 ? host : "127.0.0.1";
-}
-function resolvePort(config) {
-    return Number.isInteger(config.llamaPort) && config.llamaPort > 0 ? config.llamaPort : 8080;
-}
-function buildBaseUrl(config) {
-    const host = resolveHost(config);
-    const port = resolvePort(config);
-    const fallback = `http://${host}:${port}`;
-    const trimmed = config.llamaBaseUrl?.trim();
-    return trimmed.length > 0 ? trimmed.replace(/\/+$/, "") : fallback;
-}
-function isLocalHostname(hostname) {
-    const normalized = hostname.trim().toLowerCase().replace(/^\[(.*)\]$/u, "$1");
-    return LOCAL_ENDPOINT_HOSTS.has(normalized) || normalized.startsWith("127.");
-}
-function isLocalLlamaEndpoint(config) {
-    const baseUrl = buildBaseUrl(config);
-    try {
-        const parsed = new URL(baseUrl);
-        return isLocalHostname(parsed.hostname);
-    }
-    catch {
-        return false;
-    }
-}
-function buildDefaultDeps(overrides) {
-    return {
-        checkHealth: checkLlamaHealth,
-        inspectPortOwner: inspectPortOwner,
-        sleep,
-        spawnProcess: node_child_process_1.spawn,
-        sendSignal: (pid, signal) => {
-            process.kill(pid, signal);
-        },
-        isPidRunning: (pid) => {
-            try {
-                process.kill(pid, 0);
-                return true;
-            }
-            catch {
-                return false;
-            }
-        },
-        currentUid: () => {
-            return typeof process.getuid === "function" ? process.getuid() : null;
-        },
-        ...overrides
-    };
-}
-async function resolveServerBinaryPath(config) {
-    const envPath = process.env["LLAMA_SERVER_PATH"]?.trim();
-    if (envPath && (await exists(envPath))) {
-        return envPath;
-    }
-    const configPath = config.llamaServerPath.trim();
-    if (configPath && (await exists(configPath))) {
-        return configPath;
-    }
-    const whichResult = (0, node_child_process_1.spawnSync)("which", ["llama-server"], {
-        encoding: "utf8",
-        stdio: ["ignore", "pipe", "ignore"]
-    });
-    if (whichResult.status === 0) {
-        const whichPath = String(whichResult.stdout ?? "").trim();
-        if (whichPath.length > 0 && (await exists(whichPath))) {
-            return whichPath;
-        }
-    }
-    const fallback = (0, node_path_1.resolve)(process.env["HOME"] ?? "~", "llama.cpp", "build", "bin", "llama-server");
-    if (await exists(fallback)) {
-        return fallback;
-    }
-    return null;
-}
-async function collectGgufPaths(root) {
-    const discovered = [];
-    async function walk(path) {
-        const entries = await (0, promises_1.readdir)(path, { withFileTypes: true });
-        for (const entry of entries) {
-            const fullPath = (0, node_path_1.join)(path, entry.name);
-            if (entry.isDirectory()) {
-                await walk(fullPath);
-                continue;
-            }
-            if (entry.isFile() && entry.name.toLowerCase().endsWith(".gguf")) {
-                discovered.push(fullPath);
-            }
-        }
-    }
-    try {
-        await walk(root);
-    }
-    catch {
-        return [];
-    }
-    return discovered;
-}
-async function resolveModelPath(config) {
-    const model = config.model.trim();
-    if (model.length === 0 || model.toLowerCase() === "default") {
-        return null;
-    }
-    if (await exists(model)) {
-        return (0, node_path_1.resolve)(model);
-    }
-    const fromModelsDir = (0, node_path_1.resolve)(config.llamaModelsDir, model);
-    if (await exists(fromModelsDir)) {
-        return fromModelsDir;
-    }
-    const candidates = await collectGgufPaths((0, node_path_1.resolve)(config.llamaModelsDir));
-    const found = candidates.find((candidate) => candidate.includes(model));
-    return found ?? null;
-}
-function getOptimalContextSize() {
-    const specs = (0, hardware_1.getSystemSpecs)();
-    const rawContext = Math.floor(specs.totalMemoryGb * 512);
-    const rounded = Math.floor(rawContext / 1024) * 1024;
-    return Math.max(2048, rounded);
-}
-function startFailure(kind, message, details, metadata) {
-    return {
-        started: false,
-        failure: {
-            kind,
-            message,
-            details,
-            ...metadata
-        }
-    };
-}
-function parseProcNetListeners(procNetText, port) {
-    const inodes = [];
-    for (const line of procNetText.split(/\r?\n/).slice(1)) {
-        const parts = line.trim().split(/\s+/);
-        if (parts.length < 10) {
-            continue;
-        }
-        const localAddress = parts[1] ?? "";
-        const state = parts[3] ?? "";
-        const inode = parts[9] ?? "";
-        if (state !== "0A" || inode.length === 0) {
-            continue;
-        }
-        const addressParts = localAddress.split(":");
-        if (addressParts.length !== 2) {
-            continue;
-        }
-        const portHex = addressParts[1] ?? "";
-        const entryPort = Number.parseInt(portHex, 16);
-        if (entryPort === port) {
-            inodes.push(inode);
-        }
-    }
-    return inodes;
-}
-async function readProcessUid(pid) {
-    try {
-        const statusText = await (0, promises_1.readFile)(`/proc/${pid}/status`, "utf8");
-        const uidLine = statusText
-            .split(/\r?\n/)
-            .find((line) => line.startsWith("Uid:") || line.startsWith("Uid\t"));
-        if (!uidLine) {
-            return null;
-        }
-        const fields = uidLine.trim().split(/\s+/);
-        const uid = Number.parseInt(fields[1] ?? "", 10);
-        return Number.isInteger(uid) ? uid : null;
-    }
-    catch {
-        return null;
-    }
-}
-async function readProcessCommand(pid) {
-    try {
-        const cmdline = await (0, promises_1.readFile)(`/proc/${pid}/cmdline`, "utf8");
-        const text = cmdline.replace(/\0+/g, " ").trim();
-        if (text.length > 0) {
-            return text;
-        }
-    }
-    catch {
-        // noop
-    }
-    try {
-        return (await (0, promises_1.readFile)(`/proc/${pid}/comm`, "utf8")).trim();
-    }
-    catch {
-        return "unknown";
-    }
-}
-async function findPortOwnerByInode(inode) {
-    let processEntries;
-    try {
-        processEntries = await (0, promises_1.readdir)("/proc", { withFileTypes: true });
-    }
-    catch {
-        return null;
-    }
-    for (const entry of processEntries) {
-        if (!entry.isDirectory() || !/^\d+$/.test(entry.name)) {
-            continue;
-        }
-        const pid = Number.parseInt(entry.name, 10);
-        if (!Number.isInteger(pid) || pid <= 0) {
-            continue;
-        }
-        let fdEntries;
-        try {
-            fdEntries = await (0, promises_1.readdir)(`/proc/${pid}/fd`, { withFileTypes: true });
-        }
-        catch {
-            continue;
-        }
-        for (const fdEntry of fdEntries) {
-            const fdPath = `/proc/${pid}/fd/${fdEntry.name}`;
-            let target;
-            try {
-                target = await (0, promises_1.readlink)(fdPath);
-            }
-            catch {
-                continue;
-            }
-            if (target === `socket:[${inode}]`) {
-                return {
-                    pid,
-                    uid: await readProcessUid(pid),
-                    command: await readProcessCommand(pid)
-                };
-            }
-        }
-    }
-    return null;
-}
-async function inspectPortOwner(_host, port) {
-    if (process.platform !== "linux") {
-        return null;
-    }
-    const inodeSet = new Set();
-    for (const path of ["/proc/net/tcp", "/proc/net/tcp6"]) {
-        try {
-            const content = await (0, promises_1.readFile)(path, "utf8");
-            for (const inode of parseProcNetListeners(content, port)) {
-                inodeSet.add(inode);
-            }
-        }
-        catch {
-            // noop
-        }
-    }
-    for (const inode of inodeSet) {
-        const owner = await findPortOwnerByInode(inode);
-        if (owner) {
-            return owner;
-        }
-    }
-    return null;
-}
-function isLlamaServerCommand(command) {
-    const normalized = command.toLowerCase();
-    return normalized.includes("llama-server") || normalized.includes("llama.cpp/build/bin/server");
-}
-async function tryTerminatePid(pid, deps) {
-    try {
-        deps.sendSignal(pid, "SIGTERM");
-    }
-    catch {
-        return !deps.isPidRunning(pid);
-    }
-    for (let i = 0; i < 6; i += 1) {
-        if (!deps.isPidRunning(pid)) {
-            return true;
-        }
-        await deps.sleep(500);
-    }
-    try {
-        deps.sendSignal(pid, "SIGKILL");
-    }
-    catch {
-        return !deps.isPidRunning(pid);
-    }
-    for (let i = 0; i < 4; i += 1) {
-        if (!deps.isPidRunning(pid)) {
-            return true;
-        }
-        await deps.sleep(250);
-    }
-    return !deps.isPidRunning(pid);
-}
-async function resolvePortConflict(config, context, deps) {
-    const owner = await deps.inspectPortOwner(context.host, context.port);
-    if (!owner) {
-        return null;
-    }
-    const ownerText = `PID ${owner.pid} (${owner.command})`;
-    const metadata = {
-        host: context.host,
-        port: context.port,
-        conflictPid: owner.pid,
-        conflictCommand: owner.command
-    };
-    const policy = config.llamaPortConflictPolicy;
-    if (policy === "fail") {
-        return startFailure("port-unavailable", `Configured llama.cpp port ${context.host}:${context.port} is already in use by ${ownerText}.`, ["Change llamaPort or stop the conflicting process."], metadata);
-    }
-    if (policy === "kill-llama" && !isLlamaServerCommand(owner.command)) {
-        return startFailure("port-unavailable", `Configured llama.cpp port ${context.host}:${context.port} is already in use by ${ownerText}.`, ["Port conflict policy is kill-llama, but the owner is not llama-server."], metadata);
-    }
-    if (policy === "kill-user") {
-        const currentUid = deps.currentUid();
-        if (currentUid === null || owner.uid === null || owner.uid !== currentUid) {
-            return startFailure("port-unavailable", `Configured llama.cpp port ${context.host}:${context.port} is already in use by ${ownerText}.`, ["Port conflict policy is kill-user, but the process is not owned by current user."], metadata);
-        }
-    }
-    const terminated = await tryTerminatePid(owner.pid, deps);
-    if (!terminated) {
-        return startFailure("port-unavailable", `Failed to free llama.cpp port ${context.host}:${context.port} from ${ownerText}.`, ["Tried SIGTERM then SIGKILL."], metadata);
-    }
-    const remainingOwner = await deps.inspectPortOwner(context.host, context.port);
-    if (remainingOwner) {
-        return startFailure("port-unavailable", `Configured llama.cpp port ${context.host}:${context.port} is still in use after terminating ${ownerText}.`, [
-            `Current owner: PID ${remainingOwner.pid} (${remainingOwner.command})`,
-            "Change llamaPort or stop the conflicting process manually."
-        ], {
-            host: context.host,
-            port: context.port,
-            conflictPid: remainingOwner.pid,
-            conflictCommand: remainingOwner.command
-        });
-    }
-    return null;
-}
-function pushStderrLines(buffer, chunk, partial) {
-    partial.value += chunk;
-    const lines = partial.value.split(/\r?\n/);
-    partial.value = lines.pop() ?? "";
-    for (const line of lines) {
-        if (line.length === 0) {
-            continue;
-        }
-        buffer.push(line);
-        if (buffer.length > STDERR_TAIL_LIMIT) {
-            buffer.shift();
-        }
-    }
-}
-function flushStderrPartial(buffer, partial) {
-    const tail = partial.value.trim();
-    if (tail.length > 0) {
-        buffer.push(tail);
-        if (buffer.length > STDERR_TAIL_LIMIT) {
-            buffer.shift();
-        }
-    }
-    partial.value = "";
-}
-function isBindError(stderrLines) {
-    const text = stderrLines.join("\n").toLowerCase();
-    return (text.includes("couldn't bind") ||
-        text.includes("address already in use") ||
-        text.includes("http server socket") ||
-        text.includes("exiting due to http server error"));
-}
-function stderrTailDetails(stderrLines) {
-    if (stderrLines.length === 0) {
-        return [];
-    }
-    const tail = stderrLines.slice(-6);
-    return ["llama-server stderr (tail):", ...tail.map((line) => `  ${line}`)];
-}
-async function checkLlamaHealth(baseUrl, fetchImpl = fetch) {
-    const controller = new AbortController();
-    const timeout = setTimeout(() => controller.abort(), HEALTH_TIMEOUT_MS);
-    const endpoint = `${baseUrl.replace(/\/+$/, "")}${HEALTH_PATH}`;
-    try {
-        const response = await fetchImpl(endpoint, {
-            method: "GET",
-            signal: controller.signal
-        });
-        return response.ok;
-    }
-    catch {
-        return false;
-    }
-    finally {
-        clearTimeout(timeout);
-    }
-}
-async function killProcess(process) {
-    if (process.exitCode !== null) {
-        return;
-    }
-    process.kill("SIGTERM");
-    const started = Date.now();
-    while (process.exitCode === null && Date.now() - started < 5_000) {
-        await sleep(100);
-    }
-    if (process.exitCode === null) {
-        process.kill("SIGKILL");
-    }
-}
-async function stopLlamaServer() {
-    if (!runningState) {
-        return;
-    }
-    const state = runningState;
-    runningState = null;
-    await killProcess(state.process);
-}
-async function resetLlamaForFreshSession(config, overrides) {
-    if (!isLocalLlamaEndpoint(config)) {
-        return { started: false };
-    }
-    await stopLlamaServer();
-    return startLlamaServer(config, overrides);
-}
-async function createStartContext(config) {
-    const binaryPath = await resolveServerBinaryPath(config);
-    const baseUrl = buildBaseUrl(config);
-    const modelPath = await resolveModelPath(config);
-    const host = resolveHost(config);
-    const port = resolvePort(config);
-    if (!binaryPath) {
-        return startFailure("binary-not-found", "Could not locate llama-server binary.", [
-            "Set LLAMA_SERVER_PATH or config.llamaServerPath to your llama-server executable.",
-            "Try: which llama-server"
-        ]);
-    }
-    if (!modelPath) {
-        return startFailure("model-not-found", `Could not resolve configured model '${config.model}'.`, [
-            `Checked models dir: ${config.llamaModelsDir}`,
-            "Pick a model with /model or update config.model."
-        ]);
-    }
-    return {
-        binaryPath,
-        modelPath,
-        baseUrl,
-        host,
-        port
-    };
-}
-async function startLlamaServer(config, overrides) {
-    const deps = buildDefaultDeps(overrides);
-    const contextOrFailure = await createStartContext(config);
-    if ("started" in contextOrFailure) {
-        return contextOrFailure;
-    }
-    const context = contextOrFailure;
-    if (runningState && runningState.baseUrl === context.baseUrl) {
-        if (await deps.checkHealth(context.baseUrl)) {
-            return { started: false };
-        }
-        await stopLlamaServer();
-    }
-    const preStartConflict = await resolvePortConflict(config, context, deps);
-    if (preStartConflict) {
-        return preStartConflict;
-    }
-    const contextSize = config.llamaContextSize > 0 ? config.llamaContextSize : getOptimalContextSize();
-    const gpuLayers = config.llamaGpuLayers > 0 ? config.llamaGpuLayers : 999;
-    const args = [
-        "-m",
-        context.modelPath,
-        "-c",
-        String(contextSize),
-        "--host",
-        context.host,
-        "--port",
-        String(context.port),
-        "--embedding",
-        "-ngl",
-        String(gpuLayers)
-    ];
-    let process;
-    const stderrLines = [];
-    const partial = { value: "" };
-    try {
-        process = deps.spawnProcess(context.binaryPath, args, {
-            stdio: ["ignore", "ignore", "pipe"]
-        });
-        process.stderr?.setEncoding("utf8");
-        process.stderr?.on("data", (chunk) => {
-            pushStderrLines(stderrLines, String(chunk), partial);
-        });
-    }
-    catch (error) {
-        return startFailure("start-failed", `Failed to start llama-server: ${toErrorMessage(error)}`, [
-            `Binary: ${context.binaryPath}`,
-            `Model: ${context.modelPath}`
-        ]);
-    }
-    runningState = {
-        process,
-        baseUrl: context.baseUrl,
-        modelPath: context.modelPath
-    };
-    const started = Date.now();
-    while (Date.now() - started < STARTUP_TIMEOUT_MS) {
-        if (process.exitCode !== null) {
-            runningState = null;
-            flushStderrPartial(stderrLines, partial);
-            const bindFailure = isBindError(stderrLines);
-            const details = [
-                `Binary: ${context.binaryPath}`,
-                `Model: ${context.modelPath}`,
-                `Endpoint: ${context.host}:${context.port}`,
-                ...stderrTailDetails(stderrLines)
-            ];
-            if (bindFailure || process.exitCode === 98) {
-                return startFailure("port-unavailable", `llama-server could not bind ${context.host}:${context.port}.`, details, { host: context.host, port: context.port });
-            }
-            return startFailure("process-exited", `llama-server exited before becoming healthy (exit code ${String(process.exitCode)}).`, details, { host: context.host, port: context.port });
-        }
-        if (await deps.checkHealth(context.baseUrl)) {
-            return { started: true };
-        }
-        await deps.sleep(HEALTH_RETRY_INTERVAL_MS);
-    }
-    await stopLlamaServer();
-    return startFailure("health-timeout", "llama-server did not report healthy status before timeout.", [
-        `Endpoint: ${context.baseUrl}${HEALTH_PATH}`,
-        `Model: ${context.modelPath}`
-    ]);
-}
-function failureFromResult(result) {
-    return (result.failure ?? {
-        kind: "start-failed",
-        message: "Failed to start llama-server.",
-        details: []
-    });
-}
-async function ensureLlamaReady(config, overrides) {
-    const deps = buildDefaultDeps(overrides);
-    const baseUrl = buildBaseUrl(config);
-    if (await deps.checkHealth(baseUrl)) {
-        return { ready: true, started: false };
-    }
-    if (!config.llamaAutoStart) {
-        return {
-            ready: false,
-            started: false,
-            failure: {
-                kind: "start-failed",
-                message: "llama.cpp is not reachable and auto-start is disabled.",
-                details: [`Expected endpoint: ${baseUrl}${HEALTH_PATH}`]
-            }
-        };
-    }
-    const started = await startLlamaServer(config, overrides);
-    if (started.failure) {
-        return {
-            ready: false,
-            started: false,
-            failure: failureFromResult(started)
-        };
-    }
-    return { ready: true, started: started.started };
-}
-function formatLlamaStartupFailure(failure, config) {
-    const lines = [`${failure.message}`];
-    if (failure.host && failure.port) {
-        lines.push(`- Endpoint: ${failure.host}:${failure.port}`);
-    }
-    if (failure.conflictPid && failure.conflictCommand) {
-        lines.push(`- Conflict: PID ${failure.conflictPid} (${failure.conflictCommand})`);
-    }
-    for (const detail of failure.details) {
-        lines.push(`- ${detail}`);
-    }
-    lines.push(`- Base URL: ${buildBaseUrl(config)}`);
-    lines.push("- Verify llama-server is installed: which llama-server");
-    lines.push(`- Verify model exists: ls ${(0, node_path_1.resolve)(config.llamaModelsDir, config.model)}`);
-    return lines.join("\n");
-}