fix(broker): enforce operator broker truth boundary

This commit is contained in:
Hermes Agent
2026-07-15 08:07:21 +09:00
parent 1b63e79edc
commit 216e462df4
19 changed files with 603 additions and 35 deletions
+33 -8
View File
@@ -18,6 +18,7 @@ import httpx
from trex_core.application.broker_port import (
BalanceItem,
BrokerBalanceError,
FillStatus,
OrderAck,
OrderRequest,
@@ -285,8 +286,9 @@ class KISBroker:
"""KIS 주식잔고조회 (`TTTC8434R`/`VTTC8434R`).
Aggregates `output1[*].pdno` + `hldg_qty`, drops zero-qty rows
(D-2 leftover, Design Decision 7). Network/credential errors map to
an empty list (caller treats as no broker positions).
(D-2 leftover, Design Decision 7). Only a validated `rt_cd=0` response
may return an empty list; token/HTTP/shape/KIS failures raise a sanitized
:class:`BrokerBalanceError` so reconciliation cannot infer false zero.
"""
cano, product = self._account()
tr_id = "VTTC8434R" if self._mock else "TTTC8434R"
@@ -298,21 +300,44 @@ class KISBroker:
"PRCS_DVSN": "01",
"CTX_AREA_FK100": "", "CTX_AREA_NK100": "",
}
try:
headers = self._headers(tr_id)
except (httpx.HTTPError, KeyError, TypeError, ValueError):
# Token issuance/header construction is distinct from a rejected
# balance endpoint response and must never look like an empty book.
raise BrokerBalanceError("token_issuance_failed") from None
try:
resp = httpx.get(
f"{self._base_url}/uapi/domestic-stock/v1/trading/inquire-balance",
headers=self._headers(tr_id), params=params, timeout=self._timeout,
headers=headers, params=params, timeout=self._timeout,
)
resp.raise_for_status()
payload = resp.json()
except (httpx.HTTPError, ValueError):
return []
except httpx.HTTPStatusError as exc:
status = exc.response.status_code if exc.response is not None else None
code = "authentication_failed" if status == 401 else (
"authorization_failed" if status == 403 else "transport_failed"
)
raise BrokerBalanceError(code) from None
except httpx.HTTPError:
raise BrokerBalanceError("transport_failed") from None
except (ValueError, TypeError):
raise BrokerBalanceError("malformed_response") from None
if not isinstance(payload, dict):
raise BrokerBalanceError("malformed_response")
if str(payload.get("rt_cd")) != "0":
return []
rows = payload.get("output1") or []
raise BrokerBalanceError("broker_rejected")
rows = payload.get("output1")
if not isinstance(rows, list):
raise BrokerBalanceError("malformed_response")
out: list[BalanceItem] = []
for row in rows:
if not isinstance(row, dict):
raise BrokerBalanceError("malformed_response")
symbol = str(row.get("pdno") or "").strip()
qty = _safe_int(row.get("hldg_qty")) or 0
qty = _safe_int(row.get("hldg_qty"))
if qty is None:
raise BrokerBalanceError("malformed_response")
if not symbol or qty <= 0:
continue
out.append(BalanceItem(symbol=symbol, hldg_qty=qty))
@@ -134,6 +134,18 @@ class PostgreSQLSlotRepository:
)
return [self._row_to_slot(r) for r in cur.fetchall()]
def find_managed_slot_by_symbol(self, instance_key: str, symbol: str) -> Optional[Slot]:
with self._psycopg.connect(self._dsn) as conn:
with conn.cursor() as cur:
cur.execute(
f"SELECT {_SLOT_COLUMNS} FROM daily_slot "
"WHERE instance_key = %s AND symbol = %s "
"ORDER BY updated_at DESC NULLS LAST, slot_id ASC LIMIT 1",
(instance_key, symbol),
)
row = cur.fetchone()
return self._row_to_slot(row) if row else None
def reconcile_universe(
self, instance_key: str, universe: dict[str, str],
*, allow_retire: bool = True,
+15
View File
@@ -13,6 +13,21 @@ from decimal import Decimal
from typing import Any, Optional, Protocol
class BrokerBalanceError(RuntimeError):
"""Sanitized failure distinct from a validated empty broker balance."""
_CODES = frozenset({
"token_issuance_failed", "authentication_failed", "authorization_failed",
"transport_failed", "malformed_response", "broker_rejected",
})
def __init__(self, code: str) -> None:
if code not in self._CODES:
raise ValueError(f"unsupported broker balance error code: {code}")
self.code = code
super().__init__(f"broker balance unavailable ({code})")
@dataclass(frozen=True)
class OrderRequest:
slot_id: str # originating slot (audit/correlation)
+18 -7
View File
@@ -58,7 +58,8 @@ def reconcile_core_holdings(
→ set by reconcile_positions
warn_store: Optional — when None, WARN-log-only (backward compat, REQ-103 AC-5).
Returns {key: 'error'|'qty_warning'|'ok'|'unmanaged_warning'} for telemetry.
Returns {key: 'error'|'qty_warning'|'ok'|'unmanaged_position'|
'managed_non_holding_divergence'} for telemetry.
"""
broker_qty: dict[str, int] = {
item.symbol: int(item.hldg_qty) for item in broker.get_balance()
@@ -114,13 +115,23 @@ def reconcile_core_holdings(
)
else:
decisions[str(slot.slot_id)] = "ok"
for sym, bq in broker_qty.items(): # unmanaged broker position
for sym, bq in broker_qty.items(): # unmatched broker position
if bq > 0 and sym not in matched:
decisions[f"broker:{sym}"] = "unmanaged_warning"
_log.warning(
"broker holds %s qty=%d with no core holding slot (unmanaged position)",
sym, bq,
)
managed = slot_repo.find_managed_slot_by_symbol(instance_key, sym)
if managed is not None:
decisions[f"broker:{sym}"] = "managed_non_holding_divergence"
_log.warning(
"managed non-holding divergence: slot_id=%s portfolio_id=%s "
"symbol=%s status=%s instance_key=%s",
managed.slot_id, managed.portfolio_id, sym,
managed.status.value, instance_key,
)
else:
decisions[f"broker:{sym}"] = "unmanaged_position"
_log.warning(
"unmanaged position: symbol=%s qty=%d instance_key=%s",
sym, bq, instance_key,
)
return decisions
+14 -1
View File
@@ -51,6 +51,7 @@ class SnapshotScheduler:
warn_store: Optional[ReconcileWarningStore] = None,
credential_fetch_enabled: bool = False,
kis_reinit_callback: Optional[Callable[[BrokerCredentialPayload], None]] = None,
credential_refresh: Optional[Callable[[], object]] = None,
) -> None:
self._read = read_port
self._store = store
@@ -77,6 +78,8 @@ class SnapshotScheduler:
# KISBroker with the fetched day-start credential. None = no reinit (paper
# broker or credential_fetch_enabled=False). Set by daemon.py assembly.
self._kis_reinit_callback: Optional[Callable[[BrokerCredentialPayload], None]] = kis_reinit_callback
# REQ-119: provider-owned refresh supersedes the legacy raw fetch hook.
self._credential_refresh = credential_refresh
# REQ-113: detect break-glass at init time (env var set by operator).
self._break_glass: bool = (
os.environ.get("TREX_KIS_BROKER_BREAK_GLASS", "").lower()
@@ -250,7 +253,17 @@ class SnapshotScheduler:
# Must run before portfolio snapshot so entry_allowed is set before
# the first BuyIntent can arrive from the strategy daemon.
if self._credential_fetch_enabled:
self._maybe_fetch_credential()
if self._credential_refresh is None:
self._maybe_fetch_credential()
else:
try:
self._credential_refresh()
self._entry_allowed = True
except Exception as exc: # provider exposes only sanitized failure
_log.error("DAY_START_LOAD_FAILED: broker credential refresh failed: %s (instance=%s)",
exc.__class__.__name__, self._instance_key)
self._entry_allowed = False
self._emit_day_start_alarm(reason="broker_credential_unavailable")
if self._daemon_id is None:
try:
@@ -89,6 +89,19 @@ class InMemorySlotRepository:
if s.instance_key == instance_key and s.status is SlotState.HOLDING
]
def find_managed_slot_by_symbol(self, instance_key: str, symbol: str) -> Optional[Slot]:
candidates = [
s for s in self._slots.values()
if s.instance_key == instance_key and s.symbol == symbol
]
# PostgreSQL parity: updated_at DESC NULLS LAST, slot_id ASC.
candidates.sort(key=lambda s: (
s.updated_at is None,
-s.updated_at.timestamp() if s.updated_at is not None else 0,
str(s.slot_id),
))
return candidates[0] if candidates else None
def reconcile_universe(
self, instance_key: str, universe: dict[str, str],
*, allow_retire: bool = True,
+7
View File
@@ -72,6 +72,13 @@ class SlotRepository(Protocol):
reconcile; safety-sell / kill-switch sweeps)."""
...
def find_managed_slot_by_symbol(self, instance_key: str, symbol: str) -> Optional[Slot]:
"""Return one deterministic managed slot for exact instance+symbol.
Read-only reconciliation triage; this must not imply remediation.
"""
...
def reconcile_universe(
self, instance_key: str, universe: dict[str, str],
*, allow_retire: bool = True,
+147
View File
@@ -0,0 +1,147 @@
"""Common broker credential resolution and lifecycle boundary (REQ-119/120)."""
from __future__ import annotations
import os
from dataclasses import dataclass
from datetime import date, datetime, timedelta, timezone
from typing import Literal, Protocol
from trex_core.adapters.broker.paper import PaperBroker
from trex_core.application.broker_port import BrokerPort
from trex_core.application.portfolio import PortfolioHubUnavailableError
from trex_core.runtime.config import DaemonConfig
_KST = timezone(timedelta(hours=9))
@dataclass(frozen=True)
class BrokerSource:
"""Non-secret credential-source metadata safe for CLI/log output."""
mode: Literal["env", "fetched", "break_glass"]
identity_label: str
class BrokerCredentialUnavailable(RuntimeError):
_CODES = frozenset({"portfolio_hub_unavailable", "broker_absent", "trading_disabled", "malformed_credential", "broker_initialization_failed"})
def __init__(self, code: str) -> None:
if code not in self._CODES:
raise ValueError(f"unsupported broker credential error code: {code}")
self.code = code
super().__init__("broker credential unavailable")
class BrokerProvider(Protocol):
def get_broker_for_read(self) -> tuple[BrokerPort, BrokerSource]: ...
def refresh_day_start_credential(self) -> BrokerSource: ...
def _break_glass_enabled() -> bool:
return os.environ.get("TREX_KIS_BROKER_BREAK_GLASS", "").lower() in {"1", "true", "yes"}
def _make_env_broker(config: DaemonConfig) -> BrokerPort:
if config.broker != "kis":
seq_store = None
if config.paper.redis_url:
from trex_core.adapters.broker.seq_store import RedisSeqStore
seq_store = RedisSeqStore(config.paper.redis_url, key=f"trex:broker:paper:{config.instance_key}:order-seq")
return PaperBroker(fill_after_polls=config.paper.fill_after_polls, qty_per_poll=config.paper.qty_per_poll,
reject_symbols=set(config.paper.reject_symbols), seq_store=seq_store)
from trex_core.adapters.broker.kis import KISBroker
k = config.kis
if k is None:
raise BrokerCredentialUnavailable("broker_initialization_failed")
token_cache = None
if k.redis_url:
from trex_core.adapters.broker.token_cache import RedisTokenCache
token_cache = RedisTokenCache(k.redis_url)
return KISBroker(k.account_ref, app_key=k.app_key, app_secret=k.app_secret, base_url=k.base_url,
timeout=k.timeout, allow_live=k.allow_live, token_cache=token_cache)
class ConfigBrokerProvider:
"""One provider owns one process/command broker and fetched-source refresh."""
def __init__(self, config: DaemonConfig, *, read_port=None, clock=None) -> None:
self._config = config
self._read = read_port
self._clock = clock or (lambda: datetime.now(tz=timezone.utc))
self._broker: BrokerPort | None = None
self._source: BrokerSource | None = None
self._credential_date: date | None = None
def _read_port(self):
if self._read is None:
if self._config.portfolio_hub is None:
raise BrokerCredentialUnavailable("portfolio_hub_unavailable")
from trex_core.adapters.portfolio.hub_read import PortfolioHubReadAdapter
p = self._config.portfolio_hub
self._read = PortfolioHubReadAdapter(p.base_url, token=p.token or None, timeout=p.timeout)
return self._read
def _fetch(self) -> BrokerSource:
try:
ctx = self._read_port().fetch_broker_credential()
except PortfolioHubUnavailableError:
raise BrokerCredentialUnavailable("portfolio_hub_unavailable") from None
except BrokerCredentialUnavailable:
raise
except Exception:
raise BrokerCredentialUnavailable("portfolio_hub_unavailable") from None
if ctx is None or getattr(ctx, "broker", None) is None:
raise BrokerCredentialUnavailable("broker_absent")
if not getattr(ctx, "trading_enabled", False):
raise BrokerCredentialUnavailable("trading_disabled")
credential = ctx.broker
if not all(isinstance(getattr(credential, x, None), str) and getattr(credential, x).strip()
for x in ("account_ref", "app_key", "app_secret")):
raise BrokerCredentialUnavailable("malformed_credential")
try:
if self._broker is None:
broker = _make_env_broker(self._config)
reinitialize = getattr(broker, "reinitialize", None)
if not callable(reinitialize):
raise RuntimeError("KIS broker lacks credential reload")
reinitialize(credential)
self._broker = broker
else:
self._broker.reinitialize(credential) # type: ignore[attr-defined]
except Exception:
raise BrokerCredentialUnavailable("broker_initialization_failed") from None
self._credential_date = self._clock().astimezone(_KST).date()
# Version is designated non-secret by the portfolio contract; no account/key material.
return BrokerSource("fetched", f"credential-version:{getattr(credential, 'credential_version', 'unknown')}")
def get_broker_for_read(self) -> tuple[BrokerPort, BrokerSource]:
if self._config.broker != "kis":
if self._broker is None:
self._broker = _make_env_broker(self._config)
self._source = BrokerSource("env", "paper-config")
return self._broker, self._source
if not self._config.credential_fetch_enabled:
if self._broker is None:
self._broker = _make_env_broker(self._config)
self._source = BrokerSource("env", "kis-config")
return self._broker, self._source
if _break_glass_enabled():
if self._broker is None:
self._broker = _make_env_broker(self._config)
self._source = BrokerSource("break_glass", "operator-break-glass")
return self._broker, self._source
source = self._fetch()
self._source = source
return self._broker, source # type: ignore[return-value]
def refresh_day_start_credential(self) -> BrokerSource:
# Paper/env/break-glass paths deliberately do no portfolio-hub call.
if self._config.broker != "kis" or not self._config.credential_fetch_enabled or _break_glass_enabled():
return self.get_broker_for_read()[1]
today = self._clock().astimezone(_KST).date()
if self._broker is not None and self._credential_date == today and self._source is not None:
return self._source
_, source = self.get_broker_for_read()
return source
def build_broker_provider(config: DaemonConfig, *, read_port=None) -> ConfigBrokerProvider:
return ConfigBrokerProvider(config, read_port=read_port)
+54 -10
View File
@@ -329,6 +329,29 @@ def _fail(message: str, as_json: bool) -> None:
print(f"error: {message}", file=sys.stderr)
def _broker_read_failure(exc, as_json: bool) -> int:
"""REQ-120 command boundary: expose code only, never upstream payloads."""
code = getattr(exc, "code", "broker_unavailable")
payload = {"error": {"code": code, "message": "broker balance unavailable"}}
if as_json:
print(_json.dumps(payload), file=sys.stderr)
else:
print(f"broker balance unavailable ({code})", file=sys.stderr)
return 1
class _CachedBalanceBroker:
"""One validated balance read per operator invocation; delegates everything else."""
def __init__(self, broker, balance) -> None:
self._broker, self._balance = broker, balance
def get_balance(self):
return self._balance
def __getattr__(self, name):
return getattr(self._broker, name)
# --------------------------------------------------------------------------- #
# arg parsing
# --------------------------------------------------------------------------- #
@@ -446,7 +469,6 @@ def _build_parser() -> argparse.ArgumentParser:
def _cmd_sync(args, cfg, confirm) -> int:
"""REQ-105 (dcy-034, BAT-051): sync slot qty to broker ground truth."""
import psycopg
from trex_core.adapters.portfolio.reconcile_warning_store import (
PostgreSQLReconcileWarningStore,
@@ -476,11 +498,18 @@ def _cmd_sync(args, cfg, confirm) -> int:
print("aborted")
return 1
warn_store = PostgreSQLReconcileWarningStore(cfg.dsn)
# REQ-119/120: resolve and validate broker truth before opening/enumerating DB.
from trex_core.runtime.broker_provider import (
BrokerCredentialUnavailable, build_broker_provider,
)
from trex_core.application.broker_port import BrokerBalanceError
try:
broker_port, _source = build_broker_provider(cfg).get_broker_for_read()
broker_port = _CachedBalanceBroker(broker_port, broker_port.get_balance())
except (BrokerCredentialUnavailable, BrokerBalanceError) as exc:
return _broker_read_failure(exc, args.json)
# Build broker port
from trex_core.runtime.daemon import _build_broker
broker_port = _build_broker(cfg)
warn_store = PostgreSQLReconcileWarningStore(cfg.dsn)
# Build slot_repo + slot_service
from trex_core.adapters.persistence.postgresql import PostgreSQLSlotRepository
@@ -498,6 +527,7 @@ def _cmd_sync(args, cfg, confirm) -> int:
)
# Determine target slot_ids
import psycopg
if args.portfolio_id:
with psycopg.connect(cfg.dsn) as conn:
with conn.cursor() as cur:
@@ -564,7 +594,7 @@ def _cmd_sync(args, cfg, confirm) -> int:
print(f" after: status={after.get('status')} "
f"qty={after.get('qty')} avg_price={after.get('avg_price')}")
if cleared:
print(f" reconcile-warning CLEARED")
print(" reconcile-warning CLEARED")
return exit_code
@@ -579,12 +609,20 @@ def _cmd_verify(args, cfg) -> int:
from trex_core.application.reconciliation import verify_slot_reconcile
instance = _resolve_instance(args)
# REQ-120 no-write/no-enumeration barrier also applies to verify.
from trex_core.runtime.broker_provider import (
BrokerCredentialUnavailable, build_broker_provider,
)
from trex_core.application.broker_port import BrokerBalanceError
try:
broker_port, _source = build_broker_provider(cfg).get_broker_for_read()
broker_port = _CachedBalanceBroker(broker_port, broker_port.get_balance())
except (BrokerCredentialUnavailable, BrokerBalanceError) as exc:
return _broker_read_failure(exc, args.json)
warn_store = PostgreSQLReconcileWarningStore(cfg.dsn)
slot_repo = PostgreSQLSlotRepository(cfg.dsn)
from trex_core.runtime.daemon import _build_broker
broker_port = _build_broker(cfg)
# Build a minimal snapshot store (we only need asset_quantity for C-case check).
# For verify we use a fresh SnapshotStore populated from warn_store metadata.
# The hub_qty is read from the snapshot; if snapshot is empty we fall back to 0
@@ -787,7 +825,13 @@ def main(argv: list[str] | None = None, *, rpc=None, confirm=None) -> int:
else None
)
_load_dotenv(_toml, getattr(args, "env_file", None))
cfg = DaemonConfig.from_env()
cfg = (
DaemonConfig.from_config(
resolve_config_toml(args.config, args.config_path), env_file=args.env_file
)
if args.config
else DaemonConfig.from_env()
)
confirm = confirm or _default_confirm
# REQ-088 (dcy-029): operator report — direct-DB point-in-time aggregate.
+8 -1
View File
@@ -74,6 +74,7 @@ _log = logging.getLogger(__name__)
def _build_broker(config: DaemonConfig) -> BrokerPort:
"""Backward-compatible env-only factory; new callers use broker_provider."""
if config.broker == "kis":
from trex_core.adapters.broker.kis import KISBroker
@@ -180,7 +181,7 @@ class Daemon:
instance_key=config.instance_key,
strategy_state_store=self.strategy_state_store,
)
self.broker = _build_broker(config)
self.outbox = PostgreSQLLedgerOutbox(config.dsn)
# BAT-008: portfolio-hub read + snapshot store + durable pause + scheduler.
# (Created before the ledger writer so option-D write-back can target it.)
@@ -193,6 +194,11 @@ class Daemon:
# BAT-012: day-scoped kill-switch exclusion store + service.
self.killswitch_store = PostgreSQLKillSwitchStore(config.dsn)
self.portfolio_read = _build_portfolio_read(config)
# REQ-119: one provider owns source resolution and the long-lived broker.
# Fetched mode never starts from an implicit env broker.
from trex_core.runtime.broker_provider import build_broker_provider
self.broker_provider = build_broker_provider(config, read_port=self.portfolio_read)
self.broker, self.broker_source = self.broker_provider.get_broker_for_read()
# Event publisher transport — one dedicated instance for the
# execution-loop thread (pika is not thread-safe).
@@ -237,6 +243,7 @@ class Daemon:
alarm_emitter=self.alarm_emitter, # dcy-033 / REQ-100
warn_store=self.warn_store, # BAT-050 / REQ-103: durable reconcile-warning
credential_fetch_enabled=config.credential_fetch_enabled, # dcy-036 / REQ-113
credential_refresh=self.broker_provider.refresh_day_start_credential,
# dcy-036 / REQ-113 AC-2 / REQ-115 AC-1: KISBroker reinit callback.
# None when broker!=kis or credential_fetch_enabled=False (backward compat).
kis_reinit_callback=_build_kis_reinit_callback(config, self.broker),