Isolate CTP in worker process and improve strategy roll UX.

Split vn.py into qihuo-ctp worker with IPC client bridge, keep CTP connected during breaks with cached account fallback, speed up strategy page loads, and allow off-session breakout roll submissions. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-01 12:35:47 +08:00
parent 08d55411aa
commit 9cd81a3ea7
17 changed files with 2214 additions and 227 deletions
@@ -0,0 +1,112 @@
+"""Verify qihuo web process survives an isolated qihuo-ctp restart."""
+from __future__ import annotations
+
+import json
+import sys
+import time
+
+import paramiko
+
+sys.stdout.reconfigure(encoding="utf-8", errors="replace")
+
+
+def _connect() -> paramiko.SSHClient:
+    c = paramiko.SSHClient()
+    c.set_missing_host_key_policy(paramiko.AutoAddPolicy())
+    c.connect("192.168.8.21", username="root", password="woaini88", timeout=15)
+    return c
+
+
+def _run(c: paramiko.SSHClient, cmd: str, timeout: int = 60) -> str:
+    _, o, e = c.exec_command(cmd, timeout=timeout)
+    out = o.read().decode("utf-8", "replace")
+    err = e.read().decode("utf-8", "replace")
+    return out + (("\nERR:\n" + err) if err.strip() else "")
+
+
+def _pm2(c: paramiko.SSHClient) -> dict[str, dict]:
+    raw = _run(c, "pm2 jlist", timeout=30)
+    rows = json.loads(raw)
+    return {r.get("name"): r for r in rows}
+
+
+def _restart_count(row: dict) -> int:
+    env = row.get("pm2_env") or {}
+    return int(env.get("restart_time") or 0)
+
+
+def main() -> int:
+    c = _connect()
+    try:
+        before = _pm2(c)
+        for name in ("qihuo", "qihuo-ctp"):
+            row = before.get(name) or {}
+            env = row.get("pm2_env") or {}
+            print(
+                "before",
+                name,
+                "status",
+                env.get("status"),
+                "restarts",
+                _restart_count(row),
+                "pid",
+                row.get("pid"),
+            )
+
+        health = _run(
+            c,
+            'curl -s -H "X-Qihuo-CTP-Token: qihuo-local-ctp" '
+            "http://127.0.0.1:6601/health",
+            timeout=30,
+        )
+        print("worker_health", health[:1000])
+        web = _run(
+            c,
+            "curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:6600/login",
+            timeout=30,
+        )
+        print("web_login_before", web.strip())
+
+        print("restarting qihuo-ctp only")
+        print(_run(c, "pm2 restart qihuo-ctp --update-env", timeout=60))
+        time.sleep(8)
+
+        after = _pm2(c)
+        for name in ("qihuo", "qihuo-ctp"):
+            row = after.get(name) or {}
+            env = row.get("pm2_env") or {}
+            print(
+                "after",
+                name,
+                "status",
+                env.get("status"),
+                "restarts",
+                _restart_count(row),
+                "pid",
+                row.get("pid"),
+            )
+
+        web_after = _run(
+            c,
+            "curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:6600/login",
+            timeout=30,
+        )
+        print("web_login_after", web_after.strip())
+
+        qihuo_before = _restart_count(before.get("qihuo") or {})
+        qihuo_after = _restart_count(after.get("qihuo") or {})
+        ctp_before = _restart_count(before.get("qihuo-ctp") or {})
+        ctp_after = _restart_count(after.get("qihuo-ctp") or {})
+        ok = (
+            qihuo_after == qihuo_before
+            and ctp_after >= ctp_before + 1
+            and web_after.strip() == "200"
+        )
+        print("isolation_ok", ok)
+        return 0 if ok else 1
+    finally:
+        c.close()
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())