@cartanova/qgrid-cli 1.5.6 → 1.6.0

package/bundle/src/application/qgrid/pool.ts

@@ -1,165 +0,0 @@
-/**
- * ClaudePool — 멀티 토큰 프로세스 풀.
- *
- * 2-layer 라우팅: 토큰 선택 (least-total-depth + round-robin) → 워커 선택
- * - Map<token, Worker[]>로 토큰별 워커 관리
- * - 토큰 간 quota 균등 소진
- */
-import { type CliResult, type PoolConfig, type QueryInput, type TokenStats } from "./qgrid.types";
-import { QuotaError } from "./qgrid.types";
-import { Worker, type WorkerConfig } from "./worker";
-
-export class ClaudePool {
-  workers = new Map<string, Worker[]>();
-  tokenNames = new Map<string, string>();
-  quotaExhausted = new Set<string>();
-  requestCounts = new Map<string, number>();
-  lastUsedToken = "";
-  lastUsedWorkerIndex = 0;
-  size: number;
-  model: string;
-  timeout: number;
-  cwd: string;
-  maxCalls: number;
-
-  constructor(config: PoolConfig) {
-    this.size = config.size ?? 3;
-    this.model = config.model ?? "sonnet";
-    this.timeout = config.timeout ?? 600_000;
-    this.cwd = config.cwd ?? "/tmp/qgrid";
-    this.maxCalls = config.maxCalls ?? 500;
-
-    config.tokens.forEach(({ token, name }) => {
-      this.createWorkers(token, name);
-    });
-  }
-
-  getLastUsedWorkerName(): string {
-    const baseName = this.tokenNames.get(this.lastUsedToken) ?? "Unknown";
-    return `${baseName}-${this.lastUsedWorkerIndex}`;
-  }
-
-  getStats(): TokenStats[] {
-    return [...this.workers.keys()].map((token) => ({
-      token,
-      name: this.tokenNames.get(token) ?? "Unknown",
-      requests: this.requestCounts.get(token) ?? 0,
-      active: !this.quotaExhausted.has(token),
-    }));
-  }
-
-  private tokenRrIndex = 0;
-  private workerRrIndexes = new Map<string, number>();
-
-  selectWorker(): Worker | null {
-    const availableTokens = [...this.workers.entries()].filter(
-      ([token]) => !this.quotaExhausted.has(token),
-    );
-
-    if (availableTokens.length === 0) return null;
-
-    // 1단계: 토큰별 총 큐 depth가 가장 낮은 토큰 선택
-    const tokenDepths = availableTokens.map(([token, workers]) => ({
-      token,
-      workers,
-      totalDepth: workers.reduce((sum, w) => sum + w.getQueueDepth(), 0),
-    }));
-    const minTokenDepth = Math.min(...tokenDepths.map((t) => t.totalDepth));
-    const idleTokens = tokenDepths.filter((t) => t.totalDepth === minTokenDepth);
-    const selected = idleTokens[this.tokenRrIndex % idleTokens.length]!;
-    this.tokenRrIndex++;
-
-    // 2단계: 토큰 내에서 least-depth 워커, 동률이면 round-robin
-    const minDepth = Math.min(...selected.workers.map((w) => w.getQueueDepth()));
-    const idleWorkers = selected.workers.filter((w) => w.getQueueDepth() === minDepth);
-    const wrIdx = this.workerRrIndexes.get(selected.token) ?? 0;
-    const picked = idleWorkers[wrIdx % idleWorkers.length]!;
-    this.workerRrIndexes.set(selected.token, wrIdx + 1);
-
-    return picked;
-  }
-
-  async query(input: QueryInput, timeoutMs?: number): Promise<CliResult> {
-    const triedTokens = new Set<string>();
-
-    while (true) {
-      const worker = this.selectWorker();
-      if (!worker) {
-        throw new QuotaError("All tokens exhausted");
-      }
-
-      if (triedTokens.has(worker.tokenId)) {
-        throw new QuotaError("All tokens exhausted");
-      }
-
-      const workerName = this.tokenNames.get(worker.tokenId) ?? "Unknown";
-      const tokenWorkers = this.workers.get(worker.tokenId);
-      const workerIdx = tokenWorkers ? tokenWorkers.indexOf(worker) + 1 : 0;
-      console.log(`[qgrid] → ${workerName}-${workerIdx} (queue: ${worker.getQueueDepth()})`);
-
-      try {
-        const result = await worker.query(input, timeoutMs);
-        this.lastUsedToken = worker.tokenId;
-        this.lastUsedWorkerIndex = workerIdx;
-        this.requestCounts.set(worker.tokenId, (this.requestCounts.get(worker.tokenId) ?? 0) + 1);
-
-        return result;
-      } catch (err) {
-        if (err instanceof QuotaError) {
-          this.quotaExhausted.add(worker.tokenId);
-          triedTokens.add(worker.tokenId);
-          continue;
-        }
-        throw err;
-      }
-    }
-  }
-
-  kill(): void {
-    [...this.workers.values()].flat().forEach((w) => {
-      w.kill();
-    });
-  }
-
-  createWorkers(token: string, name: string): void {
-    if (this.workers.has(token)) return;
-
-    this.tokenNames.set(token, name);
-    const workerConfig: WorkerConfig = {
-      token,
-      model: this.model,
-      timeout: this.timeout,
-      cwd: this.cwd,
-      maxCalls: this.maxCalls,
-    };
-    const workers = Array.from({ length: this.size }, () => new Worker(workerConfig));
-    this.workers.set(token, workers);
-    this.requestCounts.set(token, 0);
-  }
-
-  destroyWorkers(token: string): void {
-    const workers = this.workers.get(token);
-    if (!workers) return;
-
-    workers.forEach((w) => {
-      w.kill();
-    });
-    this.workers.delete(token);
-    this.tokenNames.delete(token);
-    this.quotaExhausted.delete(token);
-    this.requestCounts.delete(token);
-  }
-}
-
-// Pool 싱글턴 관리
-let pool: ClaudePool | null = null;
-
-export function initPool(tokens: { token: string; name: string }[]): ClaudePool {
-  pool = new ClaudePool({ tokens });
-  return pool;
-}
-
-export function getPool(): ClaudePool {
-  if (!pool) throw new Error("Pool not initialized. Call initPool() first.");
-  return pool;
-}

package/bundle/src/application/qgrid/worker.ts

@@ -1,240 +0,0 @@
-/**
- * Worker — Claude CLI stream-json 프로세스 하나를 관리.
- *
- * 기존 SocratsAI ClaudeCliService에서 추출 + 리팩터:
- * - 토큰/모델을 생성자에서 주입
- * - env allowlist (PATH, TMPDIR, CLAUDE_CODE_OAUTH_TOKEN + CLAUDE_CODE_DISABLE_* )
- * - callCount 추적 + maxCalls 도달 시 프로세스 재활용
- * - QuotaError/TimeoutError/ProcessError 구분
- */
-import { type ChildProcess, spawn } from "node:child_process";
-import { mkdirSync } from "node:fs";
-
-import { type CliResult, type QueryInput } from "./qgrid.types";
-import { maskToken, ProcessError, QuotaError, TimeoutError } from "./qgrid.types";
-
-type PendingRequest = {
-  resolve: (value: CliResult) => void;
-  reject: (reason: Error) => void;
-  timer: ReturnType<typeof setTimeout>;
-};
-
-export type WorkerConfig = {
-  token: string;
-  model: string;
-  timeout: number;
-  cwd: string;
-  maxCalls: number;
-};
-
-export class Worker {
-  tokenId: string;
-  config: WorkerConfig;
-  process: ChildProcess | null = null;
-  buffer = "";
-  pending: PendingRequest | null = null;
-  sessionId: string | null = null;
-  queue: Array<() => void> = [];
-  cwdCreated = false;
-  callCount = 0;
-  needsRecycle = false;
-
-  constructor(config: WorkerConfig) {
-    this.config = config;
-    this.tokenId = config.token;
-  }
-
-  getQueueDepth(): number {
-    return this.queue.length + (this.pending ? 1 : 0);
-  }
-
-  rejectPending(err: Error): void {
-    if (!this.pending) return;
-    this.pending.reject(err);
-    clearTimeout(this.pending.timer);
-    this.pending = null;
-  }
-
-  ensureProcess(): ChildProcess {
-    if (this.process && this.process.exitCode === null && !this.needsRecycle) {
-      return this.process;
-    }
-
-    if (this.process && this.process.exitCode === null) {
-      this.process.kill();
-      this.process = null;
-    }
-
-    const env: NodeJS.ProcessEnv = {
-      PATH: process.env.PATH,
-      TMPDIR: process.env.TMPDIR,
-      CLAUDE_CODE_OAUTH_TOKEN: this.config.token,
-      CLAUDE_CODE_DISABLE_AUTO_MEMORY: "1",
-      CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING: "1",
-      CLAUDE_CODE_DISABLE_1M_CONTEXT: "1",
-    };
-
-    const { cwd } = this.config;
-    if (!this.cwdCreated) {
-      mkdirSync(cwd, { recursive: true });
-      this.cwdCreated = true;
-    }
-
-    const child = spawn(
-      "claude",
-      [
-        "-p",
-        "--input-format",
-        "stream-json",
-        "--output-format",
-        "stream-json",
-        "--verbose",
-        "--max-turns",
-        "1",
-        "--permission-mode",
-        "bypassPermissions",
-        "--setting-sources",
-        "project",
-        "--model",
-        this.config.model,
-      ],
-      { stdio: ["pipe", "pipe", "ignore"], env, cwd },
-    );
-
-    child.stdout?.on("data", (d: Buffer) => {
-      this.buffer += d.toString();
-      this.processBuffer();
-    });
-
-    child.on("close", () => {
-      this.process = null;
-      this.rejectPending(
-        new ProcessError(`CLI process closed (token: ${maskToken(this.tokenId)})`),
-      );
-      this.drainQueue();
-    });
-
-    child.on("error", (err) => {
-      this.process = null;
-      this.rejectPending(
-        new ProcessError(`CLI process error: ${err.message} (token: ${maskToken(this.tokenId)})`),
-      );
-      this.drainQueue();
-    });
-
-    this.process = child;
-    this.buffer = "";
-    this.sessionId = null;
-    this.callCount = 0;
-    this.needsRecycle = false;
-    return child;
-  }
-
-  processBuffer(): void {
-    const lines = this.buffer.split("\n");
-    this.buffer = lines.pop() ?? "";
-
-    lines
-      .filter((line) => line.trim() !== "")
-      .forEach((line) => {
-        try {
-          const j = JSON.parse(line);
-
-          if (j.type === "system" && j.subtype === "init" && j.session_id) {
-            this.sessionId = j.session_id;
-          }
-
-          if (j.type === "result" && this.pending) {
-            let text: string = j.result ?? "";
-            text = text.replace(/^```(?:json)?\s*\n?/i, "").replace(/\n?```\s*$/i, "");
-
-            if (text.startsWith("You've hit")) {
-              this.pending.reject(
-                new QuotaError(`Quota exhausted (token: ${maskToken(this.tokenId)})`),
-              );
-              clearTimeout(this.pending.timer);
-              this.pending = null;
-              this.drainQueue();
-              return;
-            }
-
-            const u = j.usage ?? {};
-            this.pending.resolve({
-              text,
-              usage: {
-                input_tokens: u.input_tokens ?? 0,
-                output_tokens: u.output_tokens ?? 0,
-                cache_creation_input_tokens: u.cache_creation_input_tokens ?? 0,
-                cache_read_input_tokens: u.cache_read_input_tokens ?? 0,
-              },
-              durationMs: j.duration_ms ?? 0,
-              costUsd: j.total_cost_usd ?? 0,
-            });
-            clearTimeout(this.pending.timer);
-            this.pending = null;
-            this.drainQueue();
-          }
-        } catch {
-          // skip non-json lines
-        }
-      });
-  }
-
-  drainQueue(): void {
-    if (this.pending) return;
-    const next = this.queue.shift();
-    if (next) next();
-  }
-
-  async query(input: QueryInput, timeoutMs?: number): Promise<CliResult> {
-    if (this.pending) {
-      await new Promise<void>((resolve) => {
-        this.queue.push(resolve);
-      });
-    }
-
-    const child = this.ensureProcess();
-    const timeout = timeoutMs ?? this.config.timeout;
-
-    this.callCount++;
-    if (this.callCount >= this.config.maxCalls) {
-      this.needsRecycle = true;
-    }
-
-    const prompt = input.system ? `${input.system}\n\n${input.prompt}` : input.prompt;
-
-    return new Promise<CliResult>((resolve, reject) => {
-      const timer = setTimeout(() => {
-        this.pending = null;
-        reject(
-          new TimeoutError(`Timeout after ${timeout / 1000}s (token: ${maskToken(this.tokenId)})`),
-        );
-        this.kill();
-      }, timeout);
-
-      this.pending = { resolve, reject, timer };
-
-      const msg = JSON.stringify({
-        type: "user",
-        message: { role: "user", content: prompt },
-        session_id: this.sessionId ?? "default",
-        parent_tool_use_id: null,
-      });
-
-      child.stdin?.write(`${msg}\n`);
-    });
-  }
-
-  kill(): void {
-    if (this.process && this.process.exitCode === null) {
-      this.process.kill();
-    }
-    this.process = null;
-    this.buffer = "";
-    this.sessionId = null;
-    this.callCount = 0;
-    this.needsRecycle = false;
-    this.rejectPending(new ProcessError(`Worker killed (token: ${maskToken(this.tokenId)})`));
-    this.drainQueue();
-  }
-}