feat(relay): реле через общую папку — запросы агента из файлов в свой HTTP и обратно (#5)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Andrey Epifancev 2026-10-13 00:39:36 +04:00
parent e025577d05
commit 787543ea89
2 changed files with 466 additions and 0 deletions

223
src/jira_hub/relay.py Normal file
View File

@ -0,0 +1,223 @@
"""Реле через общую папку (specs/hub-api, «Реле через общую папку»): файлы-запросы агента из <relay_dir>/req/
пересылаются в HTTP этого же хаба, ответы кладутся в <relay_dir>/resp/. HTTP API, обработчики и токены не меняются;
реле для них — ещё один клиент. Нужно, когда VPN на виртуалке закрывает сеть до хаба (design.md folder-transport)."""
import json
import os
import threading
import time
import traceback
import urllib.error
import urllib.request
from datetime import datetime, timezone
from pathlib import Path
from . import settings
CLAIM_PATH = "/agent/claim"
CLAIM_MIN_LEFT = 35 # с: claim пересылается, только если до deadline осталось не меньше (хаб держит claim ≤ 30 с)
SETTINGS_EVERY = 5 # с: как часто перечитывать relay_dir
CLEANUP_EVERY = 60 # с: как часто убирать мусор
RESP_TTL = 600 # с: ответ, за которым агент не пришёл
REQ_GRACE = 60 # с после deadline: запрос, который никто не взял
TMP_TTL = 600 # с: брошенные временные файлы (имя на точку)
BAD_JSON_WAIT = 2 # с: сколько перечитывать нечитаемый запрос
FORWARD_TIMEOUT = 90 # с: ожидание ответа своего HTTP (claim держит до 30 с)
METHODS = ("GET", "POST", "PUT")
def say(msg: str) -> None:
print(f"[{time.strftime('%d/%b/%Y %H:%M:%S')}] {msg}", flush=True)
def parse_deadline(value) -> float | None:
"""ISO 8601 → секунды эпохи; без зоны считается UTC; мусор → None."""
if not isinstance(value, str):
return None
try:
dt = datetime.fromisoformat(value)
except ValueError:
return None
if dt.tzinfo is None:
dt = dt.replace(tzinfo=timezone.utc)
return dt.timestamp()
def write_atomic(directory: Path, name: str, payload: dict) -> None:
"""Записать JSON под временным именем на точку и переименовать: читатель видит файл только целиком."""
tmp = Path(directory) / f".{name}.tmp"
with open(tmp, "w", encoding="utf-8") as f:
json.dump(payload, f, ensure_ascii=False)
f.flush()
os.fsync(f.fileno())
os.replace(tmp, Path(directory) / name)
class Relay:
def __init__(self, hub, port: int, host: str = "127.0.0.1", poll: float = 0.1):
self.hub, self.port, self.host, self.poll = hub, port, host, poll
self.root: Path | None = None
self._settings_at: float | None = None
self._cleanup_at = 0.0
self._ready = False
self._warned = False
self.workers: list[threading.Thread] = []
# --- цикл ---------------------------------------------------------------------
def run(self, stop: threading.Event) -> None:
while not stop.is_set():
try:
self.step()
except Exception: # noqa: BLE001 — сбой одного шага не останавливает реле
traceback.print_exc()
stop.wait(self.poll)
def step(self) -> None:
now = time.time()
if self._settings_at is None or now - self._settings_at >= SETTINGS_EVERY:
self._settings_at = now
self._reload()
if self.root is None:
return
if not self.root.is_dir():
if not self._warned:
say(f"relay: папки {self.root} нет — жду")
self._warned = True
self._ready = False
return
if not self._ready:
self._prepare()
for path in sorted((self.root / "req").iterdir()):
if path.name.startswith(".") or not path.name.endswith(".json"):
continue
self._take(path)
if now - self._cleanup_at >= CLEANUP_EVERY:
self._cleanup_at = now
self.cleanup(now)
def join(self, timeout: float = 5) -> None:
for w in list(self.workers):
w.join(timeout)
# --- папка ----------------------------------------------------------------------
def _reload(self) -> None:
conn = self.hub.conn()
try:
value = settings.get(conn, "relay_dir")
finally:
conn.close()
root = Path(value) if value else None
if root != self.root:
self.root, self._ready, self._warned = root, False, False
say(f"relay: папка {root}" if root else "relay: выключено")
def _prepare(self) -> None:
for sub in ("req", "resp", "work"):
(self.root / sub).mkdir(exist_ok=True)
for stale in (self.root / "work").iterdir():
stale.unlink(missing_ok=True)
say(f"relay: удалён незавершённый запрос {stale.name} (перезапуск реле)")
self._ready, self._warned = True, False
def _take(self, path: Path) -> None:
target = self.root / "work" / path.name
try:
os.rename(path, target)
except FileNotFoundError: # агент успел удалить свой просроченный запрос
return
worker = threading.Thread(target=self.process, args=(self.root, target), daemon=True)
worker.start()
self.workers = [w for w in self.workers if w.is_alive()] + [worker]
# --- обработка запроса -------------------------------------------------------------
def process(self, root: Path, work: Path) -> None:
started = time.monotonic()
try:
req = self._read(work)
if req is None:
say(f"relay: нечитаемый запрос {work.name} — выброшен")
return
method, path = req.get("method"), req.get("path")
deadline = parse_deadline(req.get("deadline"))
left = deadline - time.time() if deadline is not None else -1.0
if left <= 0 or (method == "POST" and path == CLAIM_PATH and left < CLAIM_MIN_LEFT):
say(f"relay {method} {path} просрочен (осталось {left:.0f} с) — выброшен")
return
status, body = self.forward(method, path, req.get("body"), req.get("authorization"))
write_atomic(root / "resp", work.name, {"id": req.get("id"), "status": status, "body": body})
say(f"relay {method} {path} {status} {time.monotonic() - started:.1f} с")
except OSError as e: # папка пропала на ходу — агент дождётся своего срока
say(f"relay: сбой файла {work.name}: {e}")
finally:
work.unlink(missing_ok=True)
def _read(self, work: Path) -> dict | None:
end = time.monotonic() + BAD_JSON_WAIT
while True:
try:
with open(work, encoding="utf-8") as f:
data = json.load(f)
return data if isinstance(data, dict) else None
except (ValueError, OSError):
if time.monotonic() >= end:
return None
time.sleep(0.1)
def forward(self, method, path, body, authorization) -> tuple[int, str]:
if method not in METHODS or not isinstance(path, str) or not path.startswith("/"):
return 400, _error("bad_request", f"неверный метод или путь: {method} {path}")
data = json.dumps(body, ensure_ascii=False).encode("utf-8") if body is not None else None
try:
req = urllib.request.Request(f"http://{self.host}:{self.port}{path}", data=data, method=method)
if authorization:
req.add_header("Authorization", str(authorization))
if data is not None:
req.add_header("Content-Type", "application/json; charset=utf-8")
with urllib.request.urlopen(req, timeout=FORWARD_TIMEOUT) as resp:
return resp.status, resp.read().decode("utf-8", "replace")
except urllib.error.HTTPError as e:
try:
return e.code, e.read().decode("utf-8", "replace")
finally:
e.close()
except ValueError as e: # недопустимый заголовок (перевод строки) или URL
return 400, _error("bad_request", str(e))
except (urllib.error.URLError, OSError) as e:
return 502, _error("relay", f"хаб недоступен изнутри: {e}")
# --- уборка ------------------------------------------------------------------------
def cleanup(self, now: float) -> None:
if self.root is None or not self.root.is_dir():
return
for p in _entries(self.root / "resp"):
ttl = TMP_TTL if p.name.startswith(".") else RESP_TTL
if _age(p, now) > ttl:
p.unlink(missing_ok=True)
for p in _entries(self.root / "req"):
if p.name.startswith("."):
if _age(p, now) > TMP_TTL:
p.unlink(missing_ok=True)
continue
try:
deadline = parse_deadline(json.loads(p.read_text(encoding="utf-8")).get("deadline"))
except (OSError, ValueError, AttributeError):
deadline = None
if deadline is None or now - deadline > REQ_GRACE:
p.unlink(missing_ok=True)
def _error(error: str, detail: str) -> str:
return json.dumps({"error": error, "detail": detail}, ensure_ascii=False)
def _entries(directory: Path) -> list[Path]:
try:
return list(directory.iterdir())
except FileNotFoundError:
return []
def _age(p: Path, now: float) -> float:
try:
return now - p.stat().st_mtime
except FileNotFoundError:
return 0.0

243
tests/test_relay.py Normal file
View File

@ -0,0 +1,243 @@
import json
import os
import threading
import time
from datetime import datetime, timedelta, timezone
import pytest
from jira_hub import api, relay, settings
from jira_hub.client import HubClient
@pytest.fixture
def env(tmp_path):
h = api.Hub(tmp_path / "hub.sqlite", backups_dir=tmp_path / "backups")
h.claim_wait = 0
conn = h.conn()
root = tmp_path / "bridge"
root.mkdir()
settings.set_(conn, "relay_dir", str(root))
tok = settings.tokens(conn)
conn.close()
srv = api.make_server(h, "127.0.0.1", 0)
th = threading.Thread(target=srv.serve_forever, daemon=True)
th.start()
r = relay.Relay(h, srv.server_port)
client = HubClient(f"http://127.0.0.1:{srv.server_port}", tok["client"])
yield {"hub": h, "root": root, "relay": r, "agent": "Bearer " + tok["agent"], "client": client,
"port": srv.server_port}
srv.shutdown()
srv.server_close()
th.join(5)
def deadline(seconds):
return (datetime.now(timezone.utc) + timedelta(seconds=seconds)).isoformat()
def put_request(root, rid, method, path, auth, body=None, seconds=60, **extra):
req = {"id": rid, "method": method, "path": path, "body": body, "authorization": auth,
"deadline": deadline(seconds), **extra}
relay.write_atomic(root / "req", f"{rid}.json", req)
def response(root, rid):
p = root / "resp" / f"{rid}.json"
return json.loads(p.read_text(encoding="utf-8")) if p.exists() else None
def run_once(r):
r.step()
r.join()
def test_request_roundtrip(env):
r, root = env["relay"], env["root"]
r.step() # подготовка: req/, resp/, work/
put_request(root, "a1", "GET", "/agent/settings", env["agent"])
run_once(r)
resp = response(root, "a1")
assert resp["status"] == 200 and resp["id"] == "a1"
assert json.loads(resp["body"])["project_key"] == "AURORA"
assert not list((root / "work").iterdir()) and not list((root / "req").iterdir())
def test_status_passes_through(env):
r, root = env["relay"], env["root"]
r.step()
put_request(root, "u1", "GET", "/issues", env["agent"]) # агентский токен на клиентский маршрут
put_request(root, "u2", "GET", "/agent/commands/999", env["agent"])
run_once(r)
assert response(root, "u1")["status"] == 401
assert response(root, "u2")["status"] == 404
def test_expired_and_short_claim_dropped(env):
r, root, c = env["relay"], env["root"], env["client"]
b = c.post("/batches", {"title": "t", "author": "test",
"commands": [{"type": "ensure_fields", "params": {"key": "A-1", "duedate": "2026-10-23"}}]})
r.step()
put_request(root, "c1", "POST", "/agent/claim", env["agent"], body={}, seconds=20)
put_request(root, "s1", "GET", "/agent/settings", env["agent"], seconds=-1)
run_once(r)
assert response(root, "c1") is None and response(root, "s1") is None
statuses = [x["status"] for x in c.get(f"/batches/{b['id']}")["commands"]]
assert statuses[0] == "queued"
assert not list((root / "work").iterdir())
def test_claim_with_enough_time_is_forwarded(env):
r, root = env["relay"], env["root"]
r.step()
put_request(root, "c2", "POST", "/agent/claim", env["agent"], body={"agent_version": 2}, seconds=50)
run_once(r)
assert response(root, "c2")["status"] == 200
def test_long_claim_does_not_block_other_requests(env):
env["hub"].claim_wait = 3
r, root = env["relay"], env["root"]
r.step()
put_request(root, "c3", "POST", "/agent/claim", env["agent"], body={}, seconds=60)
r.step()
put_request(root, "s3", "GET", "/agent/settings", env["agent"])
t0 = time.monotonic()
while response(root, "s3") is None and time.monotonic() - t0 < 2.5:
r.step()
time.sleep(0.05)
assert response(root, "s3")["status"] == 200
assert response(root, "c3") is None # claim ещё ждёт
r.join()
assert response(root, "c3")["status"] == 200
def test_unreachable_hub_gives_502(env):
r = relay.Relay(env["hub"], 1) # на порту 1 никто не слушает
root = env["root"]
r.step()
put_request(root, "p1", "GET", "/agent/settings", env["agent"])
run_once(r)
resp = response(root, "p1")
assert resp["status"] == 502 and "хаб недоступен" in resp["body"]
def test_dotfiles_ignored_and_bad_json_dropped(env, monkeypatch, capsys):
monkeypatch.setattr(relay, "BAD_JSON_WAIT", 0.2)
r, root = env["relay"], env["root"]
r.step()
(root / "req" / ".x1.json.tmp").write_text("{}", encoding="utf-8")
(root / "req" / "bad.json").write_text("{не json", encoding="utf-8")
run_once(r)
assert (root / "req" / ".x1.json.tmp").exists()
assert response(root, "bad") is None and not list((root / "work").iterdir())
assert "нечитаемый запрос bad.json" in capsys.readouterr().out
def test_bad_request_fields_get_400(env):
r, root = env["relay"], env["root"]
r.step()
put_request(root, "b1", "GET", "agent/settings", env["agent"]) # путь без /
put_request(root, "b2", "DELETE", "/agent/settings", env["agent"]) # метод не из списка
put_request(root, "b3", "GET", "/agent/settings", "Bearer x\r\nX-Evil: 1") # перевод строки в заголовке
run_once(r)
for rid in ("b1", "b2", "b3"):
assert response(root, rid)["status"] == 400, rid
def test_missing_deadline_is_dropped(env):
r, root = env["relay"], env["root"]
r.step()
relay.write_atomic(root / "req", "d1.json", {"id": "d1", "method": "GET", "path": "/agent/settings",
"authorization": env["agent"]})
relay.write_atomic(root / "req", "d2.json", {"id": "d2", "method": "GET", "path": "/agent/settings",
"authorization": env["agent"], "deadline": "завтра"})
run_once(r)
assert response(root, "d1") is None and response(root, "d2") is None
def test_vanished_request_is_skipped(env, capsys):
r, root = env["relay"], env["root"]
r.step()
put_request(root, "v1", "GET", "/agent/settings", env["agent"])
orig = os.rename
def vanish(src, dst):
os.remove(src) # агент удалил свой запрос раньше реле
return orig(src, dst)
os.rename = vanish
try:
run_once(r)
finally:
os.rename = orig
assert response(root, "v1") is None
assert "Traceback" not in capsys.readouterr().err
def test_work_cleared_on_prepare(env, capsys):
r, root = env["relay"], env["root"]
(root / "work").mkdir()
(root / "work" / "old.json").write_text("{}", encoding="utf-8")
r.step()
assert not (root / "work" / "old.json").exists()
assert "old.json" in capsys.readouterr().out
def test_missing_dir_warns_once_then_works(env, capsys):
r, root = env["relay"], env["root"]
gone = root.parent / "bridge-later"
conn = env["hub"].conn()
settings.set_(conn, "relay_dir", str(gone))
conn.close()
for _ in range(5):
r.step()
out = capsys.readouterr().out
assert out.count("нет — жду") == 1
gone.mkdir()
r.step()
assert (gone / "req").is_dir() and (gone / "resp").is_dir() and (gone / "work").is_dir()
put_request(gone, "m1", "GET", "/agent/settings", env["agent"])
run_once(r)
assert response(gone, "m1")["status"] == 200
def test_disabled_and_reenabled_without_restart(env, monkeypatch):
r, root = env["relay"], env["root"]
conn = env["hub"].conn()
settings.set_(conn, "relay_dir", None)
conn.close()
r.step()
assert r.root is None and not (root / "req").exists()
monkeypatch.setattr(relay, "SETTINGS_EVERY", 0)
conn = env["hub"].conn()
settings.set_(conn, "relay_dir", str(root))
conn.close()
r.step()
assert r.root == root and (root / "req").is_dir()
def test_cleanup_by_age(env):
r, root = env["relay"], env["root"]
r.step()
old, fresh = root / "resp" / "old.json", root / "resp" / "fresh.json"
tmp_old = root / "resp" / ".t.json.tmp"
for p in (old, fresh, tmp_old):
p.write_text("{}", encoding="utf-8")
now = time.time()
os.utime(old, (now - 700, now - 700))
os.utime(tmp_old, (now - 700, now - 700))
stale_req = root / "req" / ".q.json.tmp"
stale_req.write_text("{}", encoding="utf-8")
os.utime(stale_req, (now - 700, now - 700))
r.cleanup(now)
assert not old.exists() and not tmp_old.exists() and not stale_req.exists()
assert fresh.exists()
def test_log_line(env, capsys):
r, root = env["relay"], env["root"]
r.step()
put_request(root, "l1", "GET", "/agent/settings", env["agent"])
run_once(r)
assert "relay GET /agent/settings 200" in capsys.readouterr().out