M-4 (hub#12), T5.6 и M-12 (hub#4). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
332 lines
19 KiB
Python
332 lines
19 KiB
Python
"""Очередь команд: пакеты, выдача агенту по одной, отчёты, отмена, повтор, сверка с зеркалом, автообновление.
|
||
Новых команд хаб сам не создаёт: повтор после failed — новый пакет по запросу человека. Команду, выданную агенту и
|
||
оставшуюся без отчёта, хаб довыдаёт при следующем claim — это доставка той же выдачи (bridge-reliability, решение 1)."""
|
||
import json
|
||
import logging
|
||
from datetime import timedelta
|
||
|
||
from . import backup as _backup # noqa: F401 (имя модуля занято параметром backup в claim)
|
||
from . import commands as C
|
||
from . import db, mirror, settings
|
||
|
||
log = logging.getLogger(__name__)
|
||
|
||
TERMINAL = {"done", "failed", "skipped", "cancelled", "confirmed_by_state"}
|
||
MAX_DELIVERIES = 3 # выдач без отчёта, после которых команда — failed (bridge-reliability, решение 2)
|
||
ERROR_LIMIT = 1500 # символов ошибки в command.error (spec commands)
|
||
ONLINE_MINUTES = 5
|
||
|
||
|
||
class ClaimedError(Exception):
|
||
def __init__(self, claimed_at):
|
||
super().__init__(f"команда взята агентом {claimed_at}; отмена — с force")
|
||
self.claimed_at = claimed_at
|
||
|
||
|
||
class NotClaimedError(Exception):
|
||
"""Отчёт по команде, которую агенту не выдавали (status = queued) — API отвечает 409."""
|
||
|
||
|
||
def _cmd(row) -> dict:
|
||
d = dict(row)
|
||
d["params"] = json.loads(d["params"])
|
||
d["result"] = json.loads(d["result"]) if d.get("result") else None
|
||
return d
|
||
|
||
|
||
def get_command(conn, command_id) -> dict | None:
|
||
row = conn.execute("SELECT * FROM command WHERE id = ?", (command_id,)).fetchone()
|
||
return _cmd(row) if row else None
|
||
|
||
|
||
def _dump(value) -> str:
|
||
return json.dumps(value, ensure_ascii=False)
|
||
|
||
|
||
# --- пакеты -----------------------------------------------------------------
|
||
|
||
def create_batch(conn, title, author, cmds) -> dict:
|
||
cmds = C.validate(cmds)
|
||
keys = []
|
||
for c in cmds:
|
||
keys += [k for k in C.referenced_keys(c) if k not in keys]
|
||
if any(c["type"] != "refresh" for c in cmds) and keys:
|
||
cmds.append({"type": "refresh", "params": {"keys": keys}})
|
||
est = C.estimate([c["type"] for c in cmds], settings.pacing(conn))
|
||
with db.transaction(conn):
|
||
cur = conn.execute("INSERT INTO batch(title, author, created_at, estimate_seconds) VALUES (?, ?, ?, ?)",
|
||
(title, author or "api", db.now(), est))
|
||
batch_id = cur.lastrowid
|
||
defined = {}
|
||
for seq, c in enumerate(cmds, 1):
|
||
deps = [defined[k] for k in C.temp_keys_used(c) if k in defined]
|
||
cur = conn.execute("INSERT INTO command(batch_id, seq, type, params, depends_on) VALUES (?, ?, ?, ?, ?)",
|
||
(batch_id, seq, c["type"], _dump(c["params"]), deps[0] if deps else None))
|
||
if c["type"] == "ensure_issue":
|
||
defined[c["params"]["temp_key"]] = cur.lastrowid
|
||
return {"id": batch_id, "estimate_seconds": est, "count": len(cmds)}
|
||
|
||
|
||
def _status_of(cmds, cancelled_at=None) -> str:
|
||
sts = [c["status"] for c in cmds]
|
||
if not sts:
|
||
return "done"
|
||
if all(s in TERMINAL for s in sts):
|
||
return "cancelled" if cancelled_at else "done" # отменённый пакет — отменён, даже с частью done (T5.6)
|
||
if any(s == "claimed" for s in sts) or any(s in TERMINAL for s in sts):
|
||
return "running"
|
||
return "queued"
|
||
|
||
|
||
def batch_view(conn, batch_id, with_commands=True) -> dict | None:
|
||
b = conn.execute("SELECT * FROM batch WHERE id = ?", (batch_id,)).fetchone()
|
||
if b is None:
|
||
return None
|
||
cmds = [_cmd(r) for r in conn.execute("SELECT * FROM command WHERE batch_id = ? ORDER BY seq", (batch_id,))]
|
||
pacing = settings.pacing(conn)
|
||
d = dict(b)
|
||
d["status"] = _status_of(cmds, b["cancelled_at"])
|
||
d["counts"] = {s: sum(1 for c in cmds if c["status"] == s) for s in sorted({c["status"] for c in cmds})}
|
||
d["remaining_seconds"] = C.estimate([c["type"] for c in cmds if c["status"] in ("queued", "claimed")], pacing)
|
||
if with_commands:
|
||
d["commands"] = cmds
|
||
return d
|
||
|
||
|
||
def list_batches(conn, limit=20) -> list[dict]:
|
||
ids = [r["id"] for r in conn.execute("SELECT id FROM batch ORDER BY id DESC LIMIT ?", (int(limit),))]
|
||
return [batch_view(conn, i, with_commands=False) for i in ids]
|
||
|
||
|
||
# --- выдача -----------------------------------------------------------------
|
||
|
||
def _mapping(conn, batch_id) -> dict:
|
||
"""temp_key → настоящий ключ по завершённым ensure_issue пакета."""
|
||
rows = conn.execute("SELECT json_extract(params, '$.temp_key') AS t, json_extract(result, '$.key') AS k FROM command "
|
||
"WHERE batch_id = ? AND type = 'ensure_issue' AND status IN ('done', 'confirmed_by_state')",
|
||
(batch_id,)).fetchall()
|
||
return {r["t"]: r["k"] for r in rows if r["k"]}
|
||
|
||
|
||
def _definer_status(conn, batch_id, temp_key):
|
||
r = conn.execute("SELECT status FROM command WHERE batch_id = ? AND type = 'ensure_issue' "
|
||
"AND json_extract(params, '$.temp_key') = ?", (batch_id, temp_key)).fetchone()
|
||
return r["status"] if r else "failed"
|
||
|
||
|
||
def _next(conn) -> dict | None:
|
||
rows = conn.execute("SELECT c.* FROM command c JOIN batch b ON b.id = c.batch_id "
|
||
"WHERE c.status = 'queued' AND b.cancelled_at IS NULL ORDER BY b.created_at, b.id, c.seq").fetchall()
|
||
for row in rows:
|
||
cmd = _cmd(row)
|
||
needed = C.temp_keys_used(cmd)
|
||
if needed:
|
||
mapping = _mapping(conn, cmd["batch_id"])
|
||
unresolved = [k for k in needed if k not in mapping]
|
||
if any(_definer_status(conn, cmd["batch_id"], k) not in TERMINAL for k in unresolved):
|
||
return None # создание ещё не завершено — ждём (одна в полёте)
|
||
cmd["params"] = C.substitute(cmd["type"], cmd["params"], mapping)
|
||
if cmd["type"] == "refresh":
|
||
cmd["params"]["keys"] = [k for k in cmd["params"].get("keys", []) if not C.is_temp(k)]
|
||
unresolved = [] if cmd["params"]["keys"] else unresolved
|
||
if unresolved:
|
||
conn.execute("UPDATE command SET status = 'skipped', finished_at = ?, error = ? WHERE id = ?",
|
||
(db.now(), f"задача {unresolved[0]} не создана", cmd["id"]))
|
||
continue
|
||
return cmd
|
||
return None
|
||
|
||
|
||
def _redeliver(conn, now) -> dict | None:
|
||
"""Самая старая взятая команда без отчёта — выдать снова. Агент приходит за новой командой, только когда своей
|
||
незавершённой нет, значит выданную он не получил или потерял. После MAX_DELIVERIES выдач — failed."""
|
||
while True:
|
||
row = conn.execute("SELECT * FROM command WHERE status = 'claimed' ORDER BY claimed_at, id LIMIT 1").fetchone()
|
||
if row is None:
|
||
return None
|
||
cmd = _cmd(row)
|
||
if cmd["deliveries"] < MAX_DELIVERIES:
|
||
return cmd
|
||
conn.execute("UPDATE command SET status = 'failed', finished_at = ?, error = ? WHERE id = ?",
|
||
(now, f"агент не отчитался после {MAX_DELIVERIES} выдач", cmd["id"]))
|
||
conn.execute("UPDATE agent_status SET current_command_id = NULL WHERE id = 1 AND current_command_id = ?", (cmd["id"],))
|
||
log.warning("команда %s: агент не отчитался после %s выдач — failed", cmd["id"], MAX_DELIVERIES)
|
||
|
||
|
||
def claim(conn, agent_version=None, settings_version=None, last_jira_call_at=None, backup=None) -> dict | None:
|
||
now = db.now()
|
||
to_backup = None
|
||
with db.transaction(conn):
|
||
conn.execute("UPDATE agent_status SET last_seen_at = ?, agent_version = COALESCE(?, agent_version), "
|
||
"settings_version = COALESCE(?, settings_version), last_jira_call_at = COALESCE(?, last_jira_call_at) WHERE id = 1",
|
||
(now, None if agent_version is None else str(agent_version), settings_version, last_jira_call_at))
|
||
cmd = _redeliver(conn, now)
|
||
if cmd is not None:
|
||
pause = int(settings.pacing(conn).get(cmd["type"], 0))
|
||
conn.execute("UPDATE command SET claimed_at = ?, pause_seconds = ?, deliveries = deliveries + 1 WHERE id = ?",
|
||
(now, pause, cmd["id"]))
|
||
conn.execute("UPDATE agent_status SET current_command_id = ? WHERE id = 1", (cmd["id"],))
|
||
log.info("команда %s выдана снова (выдача %s)", cmd["id"], cmd["deliveries"] + 1)
|
||
else:
|
||
cmd = _next(conn)
|
||
if cmd is None:
|
||
conn.execute("UPDATE agent_status SET current_command_id = NULL WHERE id = 1")
|
||
return None
|
||
pause = int(settings.pacing(conn).get(cmd["type"], 0))
|
||
first = conn.execute("SELECT COUNT(*) AS n FROM command WHERE batch_id = ? AND status != 'queued'",
|
||
(cmd["batch_id"],)).fetchone()["n"] == 0
|
||
has_writes = conn.execute("SELECT 1 FROM command WHERE batch_id = ? AND type != 'refresh' LIMIT 1",
|
||
(cmd["batch_id"],)).fetchone() is not None
|
||
conn.execute("UPDATE command SET status = 'claimed', claimed_at = ?, pause_seconds = ?, params = ?, deliveries = 1 "
|
||
"WHERE id = ?", (now, pause, _dump(cmd["params"]), cmd["id"]))
|
||
conn.execute("UPDATE agent_status SET current_command_id = ? WHERE id = 1", (cmd["id"],))
|
||
if cmd["type"] == "refresh":
|
||
p = cmd["params"]
|
||
conn.execute("INSERT INTO sync_run(started_at, kind, query, command_id) VALUES (?, ?, ?, ?)",
|
||
(now, "jql" if p.get("jql") else "keys", p.get("jql") or ",".join(p.get("keys", [])), cmd["id"]))
|
||
if first and has_writes:
|
||
to_backup = f"batch-{cmd['batch_id']}"
|
||
if to_backup and backup:
|
||
try:
|
||
backup(to_backup)
|
||
except Exception: # сбой копии не должен держать очередь: команда уже взята
|
||
log.warning("резервная копия %s не создана, команда %s выдаётся без неё", to_backup, cmd["id"], exc_info=True)
|
||
return {"id": cmd["id"], "type": cmd["type"], "params": cmd["params"], "pause_seconds": pause, "batch_id": cmd["batch_id"]}
|
||
|
||
|
||
# --- отчёты -----------------------------------------------------------------
|
||
|
||
def is_full_refresh(conn, command_id) -> bool:
|
||
"""Команда — refresh по базовой JQL (полный прогон): только он вправе сбрасывать флаг neighbor."""
|
||
if not command_id:
|
||
return False
|
||
cmd = get_command(conn, command_id)
|
||
return bool(cmd and cmd["type"] == "refresh" and cmd["params"].get("jql")
|
||
and cmd["params"]["jql"] == settings.get(conn, "base_jql"))
|
||
|
||
|
||
def record_state(conn, command_id, issues_count, links_count) -> None:
|
||
conn.execute("UPDATE sync_run SET issues_count = issues_count + ?, links_count = links_count + ? WHERE command_id = ?",
|
||
(int(issues_count), int(links_count), command_id))
|
||
|
||
|
||
def report_result(conn, command_id, status, result=None, error=None, jira_calls=0, rate_limited=0) -> dict:
|
||
if status not in ("done", "failed"):
|
||
raise ValueError("status — done или failed")
|
||
error = None if error is None else str(error)[:ERROR_LIMIT]
|
||
with db.transaction(conn):
|
||
cmd = get_command(conn, command_id)
|
||
if cmd is None:
|
||
raise KeyError(command_id)
|
||
if cmd["status"] == "queued":
|
||
raise NotClaimedError(f"команда {command_id} не выдана")
|
||
if cmd["status"] in TERMINAL:
|
||
# поздний отчёт о созданной задаче по команде, которую хаб уже закрыл (force-отмена — M-5, предел выдач —
|
||
# M-4 #12): ключ не теряем, статус не трогаем
|
||
if (cmd["status"] in ("cancelled", "failed") and cmd["type"] == "ensure_issue" and status == "done"
|
||
and result and result.get("key") and not cmd["result"]):
|
||
conn.execute("UPDATE command SET result = ?, jira_calls = ?, rate_limited = ? WHERE id = ?",
|
||
(_dump(result), int(jira_calls or 0), int(rate_limited or 0), command_id))
|
||
mirror.rename_note(conn, cmd["params"]["temp_key"], result["key"])
|
||
log.warning("команда %s уже %s, но агент создал %s — ключ записан в результат",
|
||
command_id, cmd["status"], result["key"])
|
||
return get_command(conn, command_id)
|
||
return cmd # идемпотентно; cancelled не затирается
|
||
now = db.now()
|
||
conn.execute("UPDATE command SET status = ?, finished_at = ?, result = ?, error = ?, jira_calls = ?, rate_limited = ? WHERE id = ?",
|
||
(status, now, _dump(result) if result is not None else None, error, int(jira_calls or 0), int(rate_limited or 0), command_id))
|
||
conn.execute("UPDATE agent_status SET current_command_id = NULL, last_seen_at = ? WHERE id = 1 AND current_command_id = ?",
|
||
(now, command_id))
|
||
if cmd["type"] == "ensure_issue" and status == "done" and result and result.get("key"):
|
||
mirror.rename_note(conn, cmd["params"]["temp_key"], result["key"])
|
||
if cmd["type"] == "refresh":
|
||
conn.execute("UPDATE sync_run SET finished_at = ?, jira_calls = ?, rate_limited = ? WHERE command_id = ?",
|
||
(now, int(jira_calls or 0), int(rate_limited or 0), command_id))
|
||
if status == "done" and is_full_refresh(conn, command_id):
|
||
# полный прогон не обновил задачу — в базовую JQL она больше не входит: сосед. Пустой прогон (0 задач) —
|
||
# почти всегда сбой выборки, а не «все задачи ушли»: соседей не трогаем (M-12)
|
||
run = conn.execute("SELECT started_at, issues_count FROM sync_run WHERE command_id = ?", (command_id,)).fetchone()
|
||
if run and run["issues_count"] > 0:
|
||
conn.execute("UPDATE issue SET neighbor = 1 WHERE fetched_at < ?", (run["started_at"],))
|
||
if cmd["type"] == "refresh" and status == "done":
|
||
reconcile(conn)
|
||
return get_command(conn, command_id)
|
||
|
||
|
||
def reconcile(conn) -> list[int]:
|
||
"""Взятые команды без ответа, чей результат уже виден в зеркале → confirmed_by_state."""
|
||
link_types = settings.get(conn, "link_types") or {}
|
||
changed = []
|
||
for row in conn.execute("SELECT * FROM command WHERE status = 'claimed'").fetchall():
|
||
cmd = _cmd(row)
|
||
ok, result = C.check_confirmed(conn, cmd, link_types)
|
||
if not ok:
|
||
continue
|
||
with db.transaction(conn):
|
||
cur = conn.execute("UPDATE command SET status = 'confirmed_by_state', finished_at = ?, result = ? "
|
||
"WHERE id = ? AND status = 'claimed'",
|
||
(db.now(), _dump(result) if result else None, cmd["id"]))
|
||
if cur.rowcount == 0: # отчёт агента или отмена опередили сверку
|
||
continue
|
||
changed.append(cmd["id"])
|
||
return changed
|
||
|
||
|
||
# --- отмена и повтор ----------------------------------------------------------
|
||
|
||
def cancel_batch(conn, batch_id) -> dict:
|
||
with db.transaction(conn):
|
||
if conn.execute("SELECT 1 FROM batch WHERE id = ?", (batch_id,)).fetchone() is None:
|
||
raise KeyError(batch_id)
|
||
now = db.now()
|
||
conn.execute("UPDATE batch SET cancelled_at = ? WHERE id = ? AND cancelled_at IS NULL", (now, batch_id))
|
||
conn.execute("UPDATE command SET status = 'cancelled', finished_at = ? WHERE batch_id = ? AND status = 'queued'", (now, batch_id))
|
||
return batch_view(conn, batch_id)
|
||
|
||
|
||
def cancel_command(conn, command_id, force=False) -> dict:
|
||
with db.transaction(conn):
|
||
cmd = get_command(conn, command_id)
|
||
if cmd is None:
|
||
raise KeyError(command_id)
|
||
if cmd["status"] == "claimed" and not force:
|
||
raise ClaimedError(cmd["claimed_at"])
|
||
if cmd["status"] in ("queued", "claimed"):
|
||
conn.execute("UPDATE command SET status = 'cancelled', finished_at = ? WHERE id = ?", (db.now(), command_id))
|
||
conn.execute("UPDATE agent_status SET current_command_id = NULL WHERE id = 1 AND current_command_id = ?", (command_id,))
|
||
return get_command(conn, command_id)
|
||
|
||
|
||
def retry_command(conn, command_id, author) -> dict:
|
||
cmd = get_command(conn, command_id)
|
||
if cmd is None:
|
||
raise KeyError(command_id)
|
||
return create_batch(conn, f"Повтор команды {command_id}", author, [{"type": cmd["type"], "params": cmd["params"]}])
|
||
|
||
|
||
# --- агент и автообновление ----------------------------------------------------
|
||
|
||
def agent_view(conn) -> dict:
|
||
d = dict(conn.execute("SELECT * FROM agent_status WHERE id = 1").fetchone())
|
||
now = db.utcnow()
|
||
d["online"] = bool(d["last_seen_at"]) and now - db.parse(d["last_seen_at"]) < timedelta(minutes=ONLINE_MINUTES)
|
||
min_version = int(settings.get(conn, "min_agent_version") or 0)
|
||
d["warning"] = "обновите агента" if d["agent_version"] and int(d["agent_version"]) < min_version else None
|
||
d["rate_limited_24h"] = conn.execute("SELECT COALESCE(SUM(rate_limited), 0) AS n FROM command WHERE finished_at >= ?",
|
||
((now - timedelta(days=1)).isoformat(),)).fetchone()["n"]
|
||
d["redelivered"] = [dict(r) for r in conn.execute(
|
||
"SELECT id, type, claimed_at, deliveries FROM command WHERE status = 'claimed' AND deliveries > 1 ORDER BY id")]
|
||
return d
|
||
|
||
|
||
def maybe_auto_refresh(conn) -> dict | None:
|
||
minutes = int(settings.get(conn, "auto_refresh_minutes") or 0)
|
||
if minutes <= 0 or not agent_view(conn)["online"]:
|
||
return None
|
||
if conn.execute("SELECT 1 FROM command WHERE status IN ('queued', 'claimed') LIMIT 1").fetchone():
|
||
return None
|
||
last = conn.execute("SELECT MAX(started_at) AS t FROM sync_run").fetchone()["t"]
|
||
if last and db.utcnow() - db.parse(last) < timedelta(minutes=minutes):
|
||
return None
|
||
return create_batch(conn, "Автообновление", "hub", [{"type": "refresh", "params": {"jql": settings.get(conn, "base_jql")}}])
|