"use strict";
var __importDefault = (this && this.__importDefault) || function (mod) {
    return (mod && mod.__esModule) ? mod : { "default": mod };
};
Object.defineProperty(exports, "__esModule", { value: true });
exports.ensureLlamaServerRouterRunning = ensureLlamaServerRouterRunning;
exports.stopLlamaServerRouter = stopLlamaServerRouter;
/**
 * Lifecycle manager for a standalone llama.cpp router-mode `llama-server` process, used by the
 * VISION_API="llama-server" strand (see VisionAnalyzer.ts). Router mode auto-discovers GGUF
 * models from the local HF cache and loads/unloads them on demand per request, entirely isolated
 * from any agent-model process (LM Studio Bionic, Unsloth Studio) -- live-verified 2026-10-01 that
 * loading/using a vision model through a separate router-mode process never evicts an agent model
 * loaded elsewhere, which is the whole reason this strand exists. See
 * made-for-bionic/planning/made-for-preset-rollout-plan.md for the root-cause writeup of the
 * eviction behavior this works around.
 */
const fs_1 = __importDefault(require("fs"));
const os_1 = __importDefault(require("os"));
const path_1 = __importDefault(require("path"));
const child_process_1 = require("child_process");
const core_bundle_mjs_1 = require("../core-bundle.mjs");
const HEALTH_POLL_INTERVAL_MS = 1_000;
const HEALTH_POLL_TIMEOUT_MS = 30_000;
const HEALTH_FETCH_TIMEOUT_MS = 2_000;
let activePort = null;
let idleStopTimer = null;
function cancelIdleStop() {
    if (idleStopTimer !== null) {
        clearTimeout(idleStopTimer);
        idleStopTimer = null;
    }
}
function normalizePath(value) {
    const trimmed = value.trim();
    if (!trimmed)
        return "";
    return trimmed.startsWith("~") ? path_1.default.join(os_1.default.homedir(), trimmed.slice(1)) : trimmed;
}
function pidFilePath() {
    return path_1.default.join((0, core_bundle_mjs_1.getLogsDir)(), "llama-server-router.pid");
}
function logFilePath() {
    return path_1.default.join((0, core_bundle_mjs_1.getLogsDir)(), "llama-server-router.log");
}
function presetFilePath() {
    return path_1.default.join((0, core_bundle_mjs_1.getLogsDir)(), "llama-server-router-preset.ini");
}
function isProcessAlive(pid) {
    try {
        process.kill(pid, 0);
        return true;
    }
    catch {
        return false;
    }
}
function readPid() {
    try {
        const raw = fs_1.default.readFileSync(pidFilePath(), "utf-8").trim();
        const pid = parseInt(raw, 10);
        return Number.isFinite(pid) && pid > 0 ? pid : null;
    }
    catch {
        return null;
    }
}
function fetchWithTimeout(url, timeoutMs) {
    const controller = new AbortController();
    const timer = setTimeout(() => controller.abort(), timeoutMs);
    return fetch(url, { signal: controller.signal }).finally(() => clearTimeout(timer));
}
async function isHealthy(host, port) {
    try {
        const res = await fetchWithTimeout(`http://${host}:${port}/v1/models`, HEALTH_FETCH_TIMEOUT_MS);
        return res.ok;
    }
    catch {
        return false;
    }
}
async function pollHealth(host, port) {
    const deadline = Date.now() + HEALTH_POLL_TIMEOUT_MS;
    while (Date.now() < deadline) {
        if (await isHealthy(host, port))
            return true;
        await new Promise((resolve) => setTimeout(resolve, HEALTH_POLL_INTERVAL_MS));
    }
    return false;
}
function buildPresetIni(ctxSize, localModel) {
    let ini = `[*]\nctx-size = ${ctxSize}\n`;
    if (localModel) {
        ini += `\n[${localModel.name}]\nmodel = ${localModel.modelPath}\nmmproj = ${localModel.mmprojPath}\n`;
    }
    return ini;
}
async function listRouterModelIds(host, port) {
    try {
        const res = await fetchWithTimeout(`http://${host}:${port}/v1/models`, HEALTH_FETCH_TIMEOUT_MS);
        if (!res.ok)
            return null;
        const data = (await res.json());
        const models = Array.isArray(data?.data) ? data.data : [];
        return models.map((m) => String(m?.id ?? "").trim()).filter(Boolean);
    }
    catch {
        return null;
    }
}
/**
 * Ensures a router-mode llama-server is listening at host:port, spawning it if nothing answers
 * there yet. Idempotent and safe to call before every request -- a already-healthy endpoint
 * (ours or adopted) is trusted as-is, since router mode's own model management is independent of
 * which process started the server. Reschedules this process's own idle-shutdown timer on every
 * call (see releaseIdleLlamaServerRouter()).
 */
async function ensureLlamaServerRouterRunning(config) {
    const { host, port, ctxSize, modelsMax, idleTtlMinutes, localModel } = config;
    cancelIdleStop();
    activePort = port;
    const binaryPath = normalizePath(config.binaryPath);
    if (!binaryPath) {
        throw new Error("LLAMA_SERVER_BINARY is not configured.");
    }
    if (!fs_1.default.existsSync(binaryPath)) {
        throw new Error(`llama-server binary does not exist: ${binaryPath}`);
    }
    if (await isHealthy(host, port)) {
        if (localModel) {
            const registeredIds = await listRouterModelIds(host, port);
            const alreadyRegistered = registeredIds !== null && registeredIds.includes(localModel.name);
            if (!alreadyRegistered) {
                // The running router's preset doesn't know this local model yet -- router mode has no
                // hot-reload endpoint for --models-preset, so the only way to register a new local model
                // is to rewrite the ini and restart the process we ourselves own.
                await stopLlamaServerRouter(port);
            }
        }
        if (await isHealthy(host, port)) {
            scheduleIdleStop(port, idleTtlMinutes);
            return;
        }
    }
    const stalePid = readPid();
    if (stalePid !== null) {
        if (isProcessAlive(stalePid)) {
            throw new Error(`Port ${port} is not responding, but a previously spawned llama-server router (PID ${stalePid}) is still alive. Check logs: ${logFilePath()}`);
        }
        try {
            fs_1.default.unlinkSync(pidFilePath());
        }
        catch { }
    }
    fs_1.default.mkdirSync((0, core_bundle_mjs_1.getLogsDir)(), { recursive: true });
    fs_1.default.writeFileSync(presetFilePath(), buildPresetIni(ctxSize, localModel));
    const args = [
        "--host", host,
        "--port", String(port),
        "--models-max", String(modelsMax),
        "--models-preset", presetFilePath(),
    ];
    const logStream = fs_1.default.createWriteStream(logFilePath(), { flags: "a" });
    const child = (0, child_process_1.spawn)(binaryPath, args, {
        detached: true,
        stdio: ["ignore", "pipe", "pipe"],
        env: { ...process.env, HOME: os_1.default.homedir() },
    });
    child.stdout?.pipe(logStream);
    child.stderr?.pipe(logStream);
    child.once("error", (error) => {
        try {
            fs_1.default.appendFileSync(logFilePath(), `[llama-server-router] process error: ${error.message}\n`);
        }
        catch { }
    });
    child.unref();
    if (!child.pid) {
        throw new Error(`Failed to spawn llama-server router process. Check logs: ${logFilePath()}`);
    }
    fs_1.default.writeFileSync(pidFilePath(), String(child.pid));
    const ready = await pollHealth(host, port);
    if (!ready) {
        throw new Error(`llama-server router did not become healthy within ${HEALTH_POLL_TIMEOUT_MS / 1000}s on ${host}:${port}. Check logs: ${logFilePath()}`);
    }
    scheduleIdleStop(port, idleTtlMinutes);
}
/** 0 disables idle shutdown entirely (process stays until the host process itself exits). */
function scheduleIdleStop(port, idleTtlMinutes) {
    cancelIdleStop();
    if (idleTtlMinutes <= 0)
        return;
    idleStopTimer = setTimeout(() => {
        idleStopTimer = null;
        if (activePort !== port)
            return;
        void stopLlamaServerRouter(port).catch((error) => {
            try {
                fs_1.default.appendFileSync(logFilePath(), `[llama-server-router] failed to stop idle router: ${error instanceof Error ? error.message : String(error)}\n`);
            }
            catch { }
        });
    }, idleTtlMinutes * 60_000);
    idleStopTimer.unref();
}
/** Stops the router process this plugin itself spawned (tracked via the PID file); a no-op if nothing is running. */
async function stopLlamaServerRouter(port) {
    cancelIdleStop();
    const pid = readPid();
    if (pid === null)
        return;
    if (isProcessAlive(pid)) {
        try {
            process.kill(pid, "SIGTERM");
        }
        catch { }
    }
    try {
        fs_1.default.unlinkSync(pidFilePath());
    }
    catch { }
    if (activePort === port)
        activePort = null;
}

"use strict";
var __importDefault = (this && this.__importDefault) || function (mod) {
    return (mod && mod.__esModule) ? mod : { "default": mod };
};
Object.defineProperty(exports, "__esModule", { value: true });
exports.ensureLlamaServerRouterRunning = ensureLlamaServerRouterRunning;
exports.stopLlamaServerRouter = stopLlamaServerRouter;
/**
 * Lifecycle manager for a standalone llama.cpp router-mode `llama-server` process, used by the
 * VISION_API="llama-server" strand (see VisionAnalyzer.ts). Router mode auto-discovers GGUF
 * models from the local HF cache and loads/unloads them on demand per request, entirely isolated
 * from any agent-model process (LM Studio Bionic, Unsloth Studio) -- live-verified 2026-10-01 that
 * loading/using a vision model through a separate router-mode process never evicts an agent model
 * loaded elsewhere, which is the whole reason this strand exists. See
 * made-for-bionic/planning/made-for-preset-rollout-plan.md for the root-cause writeup of the
 * eviction behavior this works around.
 */
const fs_1 = __importDefault(require("fs"));
const os_1 = __importDefault(require("os"));
const path_1 = __importDefault(require("path"));
const child_process_1 = require("child_process");
const core_bundle_mjs_1 = require("../core-bundle.mjs");
const HEALTH_POLL_INTERVAL_MS = 1_000;
const HEALTH_POLL_TIMEOUT_MS = 30_000;
const HEALTH_FETCH_TIMEOUT_MS = 2_000;
let activePort = null;
let idleStopTimer = null;
function cancelIdleStop() {
    if (idleStopTimer !== null) {
        clearTimeout(idleStopTimer);
        idleStopTimer = null;
    }
}
function normalizePath(value) {
    const trimmed = value.trim();
    if (!trimmed)
        return "";
    return trimmed.startsWith("~") ? path_1.default.join(os_1.default.homedir(), trimmed.slice(1)) : trimmed;
}
function pidFilePath() {
    return path_1.default.join((0, core_bundle_mjs_1.getLogsDir)(), "llama-server-router.pid");
}
function logFilePath() {
    return path_1.default.join((0, core_bundle_mjs_1.getLogsDir)(), "llama-server-router.log");
}
function presetFilePath() {
    return path_1.default.join((0, core_bundle_mjs_1.getLogsDir)(), "llama-server-router-preset.ini");
}
function isProcessAlive(pid) {
    try {
        process.kill(pid, 0);
        return true;
    }
    catch {
        return false;
    }
}
function readPid() {
    try {
        const raw = fs_1.default.readFileSync(pidFilePath(), "utf-8").trim();
        const pid = parseInt(raw, 10);
        return Number.isFinite(pid) && pid > 0 ? pid : null;
    }
    catch {
        return null;
    }
}
function fetchWithTimeout(url, timeoutMs) {
    const controller = new AbortController();
    const timer = setTimeout(() => controller.abort(), timeoutMs);
    return fetch(url, { signal: controller.signal }).finally(() => clearTimeout(timer));
}
async function isHealthy(host, port) {
    try {
        const res = await fetchWithTimeout(`http://${host}:${port}/v1/models`, HEALTH_FETCH_TIMEOUT_MS);
        return res.ok;
    }
    catch {
        return false;
    }
}
async function pollHealth(host, port) {
    const deadline = Date.now() + HEALTH_POLL_TIMEOUT_MS;
    while (Date.now() < deadline) {
        if (await isHealthy(host, port))
            return true;
        await new Promise((resolve) => setTimeout(resolve, HEALTH_POLL_INTERVAL_MS));
    }
    return false;
}
function buildPresetIni(ctxSize, localModel) {
    let ini = `[*]\nctx-size = ${ctxSize}\n`;
    if (localModel) {
        ini += `\n[${localModel.name}]\nmodel = ${localModel.modelPath}\nmmproj = ${localModel.mmprojPath}\n`;
    }
    return ini;
}
async function listRouterModelIds(host, port) {
    try {
        const res = await fetchWithTimeout(`http://${host}:${port}/v1/models`, HEALTH_FETCH_TIMEOUT_MS);
        if (!res.ok)
            return null;
        const data = (await res.json());
        const models = Array.isArray(data?.data) ? data.data : [];
        return models.map((m) => String(m?.id ?? "").trim()).filter(Boolean);
    }
    catch {
        return null;
    }
}
/**
 * Ensures a router-mode llama-server is listening at host:port, spawning it if nothing answers
 * there yet. Idempotent and safe to call before every request -- a already-healthy endpoint
 * (ours or adopted) is trusted as-is, since router mode's own model management is independent of
 * which process started the server. Reschedules this process's own idle-shutdown timer on every
 * call (see releaseIdleLlamaServerRouter()).
 */
async function ensureLlamaServerRouterRunning(config) {
    const { host, port, ctxSize, modelsMax, idleTtlMinutes, localModel } = config;
    cancelIdleStop();
    activePort = port;
    const binaryPath = normalizePath(config.binaryPath);
    if (!binaryPath) {
        throw new Error("LLAMA_SERVER_BINARY is not configured.");
    }
    if (!fs_1.default.existsSync(binaryPath)) {
        throw new Error(`llama-server binary does not exist: ${binaryPath}`);
    }
    if (await isHealthy(host, port)) {
        if (localModel) {
            const registeredIds = await listRouterModelIds(host, port);
            const alreadyRegistered = registeredIds !== null && registeredIds.includes(localModel.name);
            if (!alreadyRegistered) {
                // The running router's preset doesn't know this local model yet -- router mode has no
                // hot-reload endpoint for --models-preset, so the only way to register a new local model
                // is to rewrite the ini and restart the process we ourselves own.
                await stopLlamaServerRouter(port);
            }
        }
        if (await isHealthy(host, port)) {
            scheduleIdleStop(port, idleTtlMinutes);
            return;
        }
    }
    const stalePid = readPid();
    if (stalePid !== null) {
        if (isProcessAlive(stalePid)) {
            throw new Error(`Port ${port} is not responding, but a previously spawned llama-server router (PID ${stalePid}) is still alive. Check logs: ${logFilePath()}`);
        }
        try {
            fs_1.default.unlinkSync(pidFilePath());
        }
        catch { }
    }
    fs_1.default.mkdirSync((0, core_bundle_mjs_1.getLogsDir)(), { recursive: true });
    fs_1.default.writeFileSync(presetFilePath(), buildPresetIni(ctxSize, localModel));
    const args = [
        "--host", host,
        "--port", String(port),
        "--models-max", String(modelsMax),
        "--models-preset", presetFilePath(),
    ];
    const logStream = fs_1.default.createWriteStream(logFilePath(), { flags: "a" });
    const child = (0, child_process_1.spawn)(binaryPath, args, {
        detached: true,
        stdio: ["ignore", "pipe", "pipe"],
        env: { ...process.env, HOME: os_1.default.homedir() },
    });
    child.stdout?.pipe(logStream);
    child.stderr?.pipe(logStream);
    child.once("error", (error) => {
        try {
            fs_1.default.appendFileSync(logFilePath(), `[llama-server-router] process error: ${error.message}\n`);
        }
        catch { }
    });
    child.unref();
    if (!child.pid) {
        throw new Error(`Failed to spawn llama-server router process. Check logs: ${logFilePath()}`);
    }
    fs_1.default.writeFileSync(pidFilePath(), String(child.pid));
    const ready = await pollHealth(host, port);
    if (!ready) {
        throw new Error(`llama-server router did not become healthy within ${HEALTH_POLL_TIMEOUT_MS / 1000}s on ${host}:${port}. Check logs: ${logFilePath()}`);
    }
    scheduleIdleStop(port, idleTtlMinutes);
}
/** 0 disables idle shutdown entirely (process stays until the host process itself exits). */
function scheduleIdleStop(port, idleTtlMinutes) {
    cancelIdleStop();
    if (idleTtlMinutes <= 0)
        return;
    idleStopTimer = setTimeout(() => {
        idleStopTimer = null;
        if (activePort !== port)
            return;
        void stopLlamaServerRouter(port).catch((error) => {
            try {
                fs_1.default.appendFileSync(logFilePath(), `[llama-server-router] failed to stop idle router: ${error instanceof Error ? error.message : String(error)}\n`);
            }
            catch { }
        });
    }, idleTtlMinutes * 60_000);
    idleStopTimer.unref();
}
/** Stops the router process this plugin itself spawned (tracked via the PID file); a no-op if nothing is running. */
async function stopLlamaServerRouter(port) {
    cancelIdleStop();
    const pid = readPid();
    if (pid === null)
        return;
    if (isProcessAlive(pid)) {
        try {
            process.kill(pid, "SIGTERM");
        }
        catch { }
    }
    try {
        fs_1.default.unlinkSync(pidFilePath());
    }
    catch { }
    if (activePort === port)
        activePort = null;
}