Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

3.3 KB
Raw Blame
/**
 * Cached server-state probe for the statusline.
 *
 * The statusline command runs once per conversation update, fresh process,
 * across potentially many concurrent CC sessions. Each call could hit the
 * server; with N tabs that's N stampedes per turn. We cache the response
 * in a shared file under STATE_DIR and let any session within `ttlMs` reuse
 * it.
 *
 * Hard request timeout (see REQUEST_TIMEOUT_MS below) — the statusline has
 * its own outer budget and we never want a slow server to make the user's
 * terminal feel laggy.
 */

import { readFileSync, writeFileSync } from "node:fs";
import { join } from "node:path";
import { buildOvHeaders } from "../shared/ov-http.mjs";
import { STATE_DIR } from "./state.mjs";

const CACHE_FILE = join(STATE_DIR, "server-probe.json");
const DEFAULT_TTL_MS = 5000;
// Per-request timeout. 1 s is loose enough that ordinary remote-server
// hiccups (TLS resumption, occasional GC pause) don't show up as a flapping
// "offline" badge, while still bounded so the statusline never blocks the
// terminal. Combined with the 5 s cache it amortises to one network round-
// trip per session per 5 s window.
const REQUEST_TIMEOUT_MS = 1000;

function readCache() {
  try {
    const raw = readFileSync(CACHE_FILE, "utf-8");
    return JSON.parse(raw);
  } catch {
    return null;
  }
}

function writeCache(payload) {
  try {
    writeFileSync(CACHE_FILE, JSON.stringify(payload));
  } catch { /* best effort */ }
}

async function fetchWithTimeout(url, headers, ms) {
  const controller = new AbortController();
  const timer = setTimeout(() => controller.abort(), ms);
  try {
    const res = await fetch(url, { headers, signal: controller.signal });
    return { ok: res.ok, status: res.status };
  } finally {
    clearTimeout(timer);
  }
}

/**
 * Probe the server for liveness and (best-effort) queue health. Result is
 * cached for `ttlMs`. Always resolves; never throws — failures map to
 * `{healthy: false, error: ...}`.
 */
export async function probeServer(cfg, { ttlMs = DEFAULT_TTL_MS } = {}) {
  const cached = readCache();
  if (cached && typeof cached.ts === "number" && Date.now() - cached.ts < ttlMs) {
    if (cached.base_url === cfg.baseUrl) return cached;
  }

  // No actor peer: /health is not scoped to one, and the statusline probe is
  // shared across every session on this machine.
  const headers = buildOvHeaders(cfg);

  const t0 = Date.now();
  let healthy = false;
  let error;

  try {
    const health = await fetchWithTimeout(`${cfg.baseUrl}/health`, headers, REQUEST_TIMEOUT_MS);
    healthy = health.ok;
    if (!healthy) error = `health_${health.status}`;
  } catch (err) {
    error = err?.name === "AbortError" ? "timeout" : (err?.message || "unreachable");
  }

  // Note: /api/v1/observer/queue was probed here in earlier iterations to
  // surface a "queue unhealthy" badge, but its `is_healthy` boolean is
  // derived from `QueueManager.has_errors()` — a lifetime cumulative counter
  // that flips to true on the first error and never resets. In practice
  // this means a server with 95%+ success rate still reports unhealthy
  // forever, producing a permanent false-alarm badge. Removed.

  const payload = {
    healthy,
    latency_ms: Date.now() - t0,
    base_url: cfg.baseUrl,
    error,
    ts: Date.now(),
  };
  writeCache(payload);
  return payload;
}