fix: fence ledger outbox claims by instance

.proofline/criteria/ac-0029.md

@@ -0,0 +1,21 @@
+---
+id: "ac-0029"
+status: active
+---
+
+# Ledger outbox instance ownership and lease fencing
+
+## Criterion
+
+`trex-core`의 durable ledger outbox row는 enqueue 시점의 required non-empty `instance_key`를 immutable owner로 저장해야 하며, daemon·operator backfill·targeted processing을 포함한 모든 delivery 경로는 자기 instance의 row만 claim하고 처리해야 한다. Shared PostgreSQL에서 claim은 pending·retry-due·unclaimed 또는 lease-expired row를 원자적으로 선택해 worker별 bounded lease를 설정해야 하고, 성공·중복·retryable·conflict·permanent·dead-letter outcome은 해당 claim의 fencing 정보가 일치할 때만 반영해야 한다. 다른 instance, legacy `instance_key IS NULL`, 아직 유효한 foreign lease 또는 이미 새 worker가 reclaim한 row는 처리·상태 변경·성공 통계·ledger-skip resolution 없이 fail closed 해야 한다.
+
+Claim owner가 crash하면 lease 만료 후 같은 instance의 다른 worker가 row를 reclaim할 수 있어야 한다. Outcome CAS가 0건이면 claim 상실로 명시적으로 처리하고 성공 outcome으로 보고해서는 안 된다. InMemory와 PostgreSQL outbox는 instance filtering, active lease exclusion, expiry recovery와 stale-claim rejection에서 같은 의미를 가져야 한다. Claim·outcome 관측 정보에는 `instance_key`, sanitized result/reason과 correlation identity만 허용하며 credential, account identifier, token, password, DSN, provider payload 또는 exception 원문을 로그·metric·alert에 포함해서는 안 된다.
+
+## Verification
+
+- 격리 조건: 서로 다른 `instance_key`의 daemon/writer 두 개가 한 PostgreSQL을 공유할 때 각 writer는 자기 row만 claim·전송·집계한다. `list_pending`, `pending_cash_delta`, batch claim과 correlation-targeted claim 모두 foreign row와 `instance_key IS NULL` legacy row를 제외한다. 누락·빈 instance key로 outbox 또는 writer를 생성하면 외부 연결이나 row 생성 전에 거부된다.
+- lease 조건: worker A가 claim한 row는 lease 동안 worker B가 claim하지 못한다. A가 outcome을 기록하면 lease가 해제되고 terminal 또는 retry 상태가 정확히 저장된다. A가 crash한 경우 expiry 전에는 B가 claim하지 못하고 expiry 뒤에는 B가 reclaim할 수 있다. Reclaim 뒤 A의 stale outcome은 typed claim-lost 결과로 거부되며 B의 row 상태, outcome, attempt count, next retry와 lease를 변경하지 않는다.
+- outcome 조건: adapter의 신규 성공과 `DUPLICATE_OK`, retryable, conflict, permanent 및 dead-letter 모든 분기는 claim fencing을 전달한다. CAS 실패 시 completion/retry counter, 성공 로그, snapshot write-back 이후의 ledger-skip resolution을 실행하지 않는다. Adapter duplicate는 row를 `completed/duplicate_completed`로 만들고 lease를 즉시 해제하여 재전송되지 않는다.
+- InMemory parity: 두 번의 동시 claim에서 첫 claim만 성공하고 active lease row는 pending scan에서 제외된다. clock을 이동하면 expiry recovery가 가능하며 이전 claim token으로 outcome을 기록할 수 없다. Instance filter와 targeted processing의 fail-closed 동작이 PostgreSQL과 같다.
+- 보안 조건: exception에 credential·account·token·password·DSN·provider payload를 식별하는 sentinel을 넣어 claim 실패, retry, permanent, dead-letter와 write-back 오류를 실행해도 rendered log, metric key/value와 alert에 sentinel 또는 exception 원문이 나타나지 않는다.
+- migration 및 회귀: migration 번호는 저장소에서 유일하고 기존 row를 자동 귀속시키지 않으며 `instance_key IS NULL`로 격리한다. 빈 DB와 기존 schema 모두에서 migration apply/replay, PostgreSQL claim round-trip, restart/expiry 회수, daemon·CLI wiring을 검증한다. 전체 repository test, Ruff, compileall, ProofLine validation과 `git diff --check`를 실행한다.

.proofline/identities.json

@@ -1,5 +1,5 @@
 {
   "schema_version": 1,
   "next_line_number": 10,
-  "next_ac_number": 29
+  "next_ac_number": 30
 }

scripts/virtual_execution_final_state.py

@@ -32,7 +32,7 @@ def main() -> int:
     )
     repo.reconcile_universe(instance, {portfolio_id: symbol})
     broker = PaperBroker(fill_after_polls=0)
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger()
     execution = ExecutionService(
         slots,

tests/adapters/persistence/test_migration_and_drift.py

@@ -118,7 +118,7 @@ def test_migration_replay_converts_existing_auto_venue_row_to_sor(pg_clean):
 
 
 def test_migration_024_separates_legacy_sor_and_is_replay_safe(pg_clean):
-    outbox = PostgreSQLLedgerOutbox(pg_clean)
+    outbox = PostgreSQLLedgerOutbox(pg_clean, instance_key="test-instance")
     normal = LedgerTrade(
         asset_id="PF-NORMAL",
         order_id="O-NORMAL",
@@ -177,7 +177,7 @@ def test_migration_024_separates_legacy_sor_and_is_replay_safe(pg_clean):
             )
 
 
-def test_migration_028_fail_closes_unowned_legacy_outbox_rows(pg_clean):
+def test_migration_029_fail_closes_unowned_legacy_outbox_rows(pg_clean):
     outbox = PostgreSQLLedgerOutbox(pg_clean, instance_key="inst-a")
     trade = LedgerTrade(
         asset_id="PF-LEGACY",

tests/adapters/persistence/test_postgresql_ledger_outbox.py

@@ -4,12 +4,16 @@ Skipped without TREX_POSTGRESQL_TEST_DSN (same gate as other PG suites).
 """
 from __future__ import annotations
 
+from concurrent.futures import ThreadPoolExecutor
 from datetime import UTC, datetime
 
+import pytest
+
 from trex_core.adapters.ledger.in_memory import InMemoryLedger
 from trex_core.adapters.persistence.postgresql import PostgreSQLLedgerOutbox
 from trex_core.application.ledger_outbox import (
     LedgerWriter,
+    OutboxClaimLostError,
     OutboxOutcome,
     WorkItemStatus,
 )
@@ -196,18 +200,27 @@ def test_stale_claim_cannot_overwrite_new_owner_outcome(pg_clean):
         expected_claim_expires_at=claim_b.claim_expires_at,
     )
 
-    stored = outbox_a.mark_outcome(
-        row.correlation_id,
-        OutboxOutcome.RETRYABLE_FAILURE,
-        error="stale owner",
-        next_retry_at=now["t"],
-        expected_claim_expires_at=claim_a.claim_expires_at,
-    )
+    with pytest.raises(OutboxClaimLostError):
+        outbox_a.mark_outcome(
+            row.correlation_id,
+            OutboxOutcome.RETRYABLE_FAILURE,
+            error="stale owner",
+            next_retry_at=now["t"],
+            expected_claim_expires_at=claim_a.claim_expires_at,
+        )
 
+    stored = outbox_a.get(row.correlation_id)
+    assert stored is not None
     assert stored.status is WorkItemStatus.COMPLETED
     assert stored.last_outcome is OutboxOutcome.COMPLETED
 
 
+@pytest.mark.parametrize("instance_key", ["", "   "])
+def test_postgresql_outbox_rejects_empty_instance_before_connect(pg_clean, instance_key):
+    with pytest.raises(ValueError, match="non-empty"):
+        PostgreSQLLedgerOutbox(pg_clean, instance_key=instance_key)
+
+
 def test_original_owner_can_finalize_expired_unreclaimed_claim(pg_clean):
     now = {"t": datetime(2026, 6, 4, tzinfo=UTC)}
     outbox = PostgreSQLLedgerOutbox(
@@ -255,6 +268,22 @@ def test_writer_records_retryable_403_without_cross_instance_delivery(pg_clean):
     assert outbox_a.get(row_a.correlation_id).status is WorkItemStatus.PENDING
 
 
+def test_concurrent_workers_claim_each_row_at_most_once(pg_clean):
+    outbox_a = PostgreSQLLedgerOutbox(
+        pg_clean, instance_key="inst-a", worker_id="worker-a"
+    )
+    outbox_b = PostgreSQLLedgerOutbox(
+        pg_clean, instance_key="inst-a", worker_id="worker-b"
+    )
+    row = outbox_a.enqueue(_trade(execution_id="pg-concurrent"))
+
+    with ThreadPoolExecutor(max_workers=2) as pool:
+        claims = list(pool.map(lambda box: box.claim_pending(), (outbox_a, outbox_b)))
+
+    claimed_ids = [item.correlation_id for batch in claims for item in batch]
+    assert claimed_ids == [row.correlation_id]
+
+
 def test_writer_marks_duplicate_completed_from_adapter_response_on_pg(pg_clean):
     class _DuplicateLedger:
         def record_trade(self, _trade):

tests/application/test_execution_cancel.py

@@ -70,7 +70,7 @@ def _wire(broker, *, stale_sec=None):
     svc = SlotService(repo)
     intents = IntentService(svc, repo, InMemoryIntentStore(),
                             resolve_symbol=lambda pid: "005930")
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger()
     execution = ExecutionService(
         svc, repo, broker, ledger_outbox=outbox,

tests/application/test_execution_fill_price_backfill.py

@@ -120,7 +120,7 @@ def _wire(broker):
         resolve_symbol=lambda pid: "005930",
         instance_key=INST,
     )
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     execution = ExecutionService(svc, repo, broker, ledger_outbox=outbox, instance_key=INST)
     repo.reconcile_universe(INST, {"PF-1": "005930"})
     return repo, svc, intents, execution, outbox
@@ -966,7 +966,7 @@ class TestPaperBrokerRegression:
             resolve_symbol=lambda pid: "005930",
             instance_key=INST,
         )
-        outbox2 = InMemoryLedgerOutbox()
+        outbox2 = InMemoryLedgerOutbox(instance_key="test-instance")
         execution2 = ExecutionService(svc2, repo2, paper, ledger_outbox=outbox2, instance_key=INST)
         repo2.reconcile_universe(INST, {"PF-1": "005930"})
 

tests/application/test_execution_ledger.py

@@ -119,7 +119,11 @@ def _wire(broker, *, ledger=None, clock=None, alarm_emitter=None, ledger_skip_st
     svc = SlotService(repo)
     intents = IntentService(svc, repo, InMemoryIntentStore(),
                             resolve_symbol=lambda pid: "005930")
-    outbox = InMemoryLedgerOutbox(clock=clock) if clock else InMemoryLedgerOutbox()
+    outbox = (
+        InMemoryLedgerOutbox(clock=clock, instance_key="default")
+        if clock
+        else InMemoryLedgerOutbox(instance_key="default")
+    )
     execution = ExecutionService(
         svc,
         repo,
@@ -430,7 +434,7 @@ def test_duplicate_execution_id_results_in_zero_duplicate_in_ledger():
     # And even if the writer somehow re-processes (simulated by a fresh outbox
     # row for the same execution_id, e.g. after a manual replay), the ledger
     # adapter dedup'd it server-side and the outcome is duplicate_completed.
-    replay_outbox = InMemoryLedgerOutbox()
+    replay_outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     replay_writer = LedgerWriter(replay_outbox, ledger)
     replay_outbox.enqueue(first.trade)
     stats = replay_writer.run_once()

tests/application/test_ledger_option_d.py

@@ -39,7 +39,7 @@ def _store_with(pid="PF-1", cash=1_000_000.0, qty=0.0):
 # --- write-back (REQ-019/030) ----------------------------------------------
 def test_writeback_updates_snapshot_on_new():
     store = _store_with()
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger(cash_after=300_000.0, quantity_after=10.0)
     writer = LedgerWriter(outbox, ledger, snapshot_store=store)
     outbox.enqueue(_trade())
@@ -50,7 +50,7 @@ def test_writeback_updates_snapshot_on_new():
 
 def test_writeback_applies_on_duplicate_too():
     store = _store_with(cash=1_000_000.0)
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     # execution already recorded once → second write is duplicate, *_after still authoritative
     ledger = InMemoryLedger(cash_after=300_000.0, quantity_after=10.0)
     ledger.record_trade(_trade(execution_id="dup"))   # pre-record → next is duplicate
@@ -62,7 +62,7 @@ def test_writeback_applies_on_duplicate_too():
 
 def test_writeback_skipped_when_after_missing():
     store = _store_with(cash=1_000_000.0)
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger()  # no cash_after/quantity_after → None
     writer = LedgerWriter(outbox, ledger, snapshot_store=store)
     outbox.enqueue(_trade())
@@ -74,7 +74,7 @@ def test_writeback_best_effort_does_not_block_completion():
     class _BoomStore:
         def update_cash_position(self, *a, **k):
             raise RuntimeError("boom")
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger(cash_after=1.0, quantity_after=1.0)
     writer = LedgerWriter(outbox, ledger, snapshot_store=_BoomStore())
     outbox.enqueue(_trade())
@@ -83,7 +83,7 @@ def test_writeback_best_effort_does_not_block_completion():
 
 
 def test_no_writeback_without_store():
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger(cash_after=1.0, quantity_after=1.0)
     writer = LedgerWriter(outbox, ledger)  # no snapshot_store
     outbox.enqueue(_trade())
@@ -92,7 +92,7 @@ def test_no_writeback_without_store():
 
 # --- estimate overlay (REQ-029 / OQ-4) — via make_slots_rpc_handler ----------
 def test_pending_cash_delta_sums_by_asset():
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     outbox.enqueue(_trade(execution_id="b", side="buy", qty=10, price=1000.0, fee=1.0))
     deltas = outbox.pending_cash_delta()
     assert deltas["PF-1"] == trade_cash_delta(_trade(side="buy", qty=10, price=1000.0, fee=1.0))
@@ -101,7 +101,7 @@ def test_pending_cash_delta_sums_by_asset():
 
 def test_apply_cash_estimate_overlays_pending_delta():
     """Cash-estimated overlay applied through slots handler (REQ-029 OQ-4)."""
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     outbox.enqueue(_trade(execution_id="b", side="buy", qty=10, price=1000.0, fee=1.0))
     store = _store_with(pid="PF-1", cash=1_000_000.0, qty=0.0)
     repo = InMemorySlotRepository()
@@ -117,7 +117,7 @@ def test_apply_cash_estimate_overlays_pending_delta():
 def test_estimate_clears_after_writeback():
     """Once the ledger write completes (no longer pending), the estimate is gone."""
     store = _store_with(cash=1_000_000.0)
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger(cash_after=989_999.0, quantity_after=10.0)
     writer = LedgerWriter(outbox, ledger, snapshot_store=store)
     outbox.enqueue(_trade(side="buy", qty=10, price=1000.0, fee=1.0))

tests/application/test_ledger_outbox.py

@@ -1,16 +1,20 @@
 """LedgerOutbox + LedgerWriter direct unit tests (REQ-019)."""
 from __future__ import annotations
 
+from dataclasses import replace
 from datetime import UTC, datetime, timedelta
 
+import pytest
+
 from trex_core.adapters.ledger.in_memory import InMemoryLedger
 from trex_core.application.ledger_outbox import (
     InMemoryLedgerOutbox,
     LedgerWriter,
+    OutboxClaimLostError,
     OutboxOutcome,
     WorkItemStatus,
 )
-from trex_core.application.ledger_port import LedgerTrade
+from trex_core.application.ledger_port import LedgerPermanentError, LedgerTrade
 
 
 def _trade(execution_id="paper:PAPER-1:10") -> LedgerTrade:
@@ -22,7 +26,7 @@ def _trade(execution_id="paper:PAPER-1:10") -> LedgerTrade:
 
 
 def test_enqueue_is_idempotent_on_correlation_id():
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     a = outbox.enqueue(_trade())
     b = outbox.enqueue(_trade())
     assert a.correlation_id == b.correlation_id
@@ -30,7 +34,7 @@ def test_enqueue_is_idempotent_on_correlation_id():
 
 
 def test_writer_run_once_drains_pending_and_marks_completed():
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger()
     writer = LedgerWriter(outbox, ledger)
     outbox.enqueue(_trade(execution_id="e-1"))
@@ -42,7 +46,7 @@ def test_writer_run_once_drains_pending_and_marks_completed():
 
 
 def test_writer_marks_conflict_blocked_and_does_not_retry():
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger(conflict_executions={"e-x"})
     writer = LedgerWriter(outbox, ledger)
     outbox.enqueue(_trade(execution_id="e-x"))
@@ -58,7 +62,9 @@ def test_writer_retries_until_success():
     # dcy-015: retryable rows now back off; share an injected clock between the
     # outbox (list_pending "now") and the writer (next_retry_at) to advance past it.
     now = {"t": datetime(2026, 6, 4, tzinfo=UTC)}
-    outbox = InMemoryLedgerOutbox(clock=lambda: now["t"])
+    outbox = InMemoryLedgerOutbox(
+        clock=lambda: now["t"], instance_key="test-instance"
+    )
     ledger = InMemoryLedger(fail_n_then_succeed=1)
     writer = LedgerWriter(outbox, ledger, clock=lambda: now["t"])
     outbox.enqueue(_trade(execution_id="e-r"))
@@ -90,17 +96,29 @@ def test_process_correlation_id_only_drains_requested_row():
 def test_process_correlation_id_skips_foreign_instance_row():
     outbox = InMemoryLedgerOutbox(instance_key="inst-a")
     ledger = InMemoryLedger()
-    writer = LedgerWriter(outbox, ledger, instance_key="inst-b")
     item = outbox.enqueue(_trade(execution_id="e-foreign"))
 
-    outcome = writer.process_correlation_id(item.correlation_id)
-
-    assert outcome is None
+    with pytest.raises(ValueError, match="must match"):
+        LedgerWriter(outbox, ledger, instance_key="inst-b")
     assert ledger.count() == 0
-    assert writer.counters["claim:foreign_instance"] == 1
     assert outbox.get(item.correlation_id).status is WorkItemStatus.PENDING
 
 
+def test_process_correlation_id_rejects_legacy_row_without_instance_owner():
+    class _LegacyOutbox(InMemoryLedgerOutbox):
+        def get(self, correlation_id):
+            item = super().get(correlation_id)
+            return replace(item, instance_key=None) if item is not None else None
+
+    outbox = _LegacyOutbox(instance_key="inst-a")
+    ledger = InMemoryLedger()
+    writer = LedgerWriter(outbox, ledger, instance_key="inst-a")
+    item = outbox.enqueue(_trade())
+
+    assert writer.process_correlation_id(item.correlation_id) is None
+    assert ledger.count() == 0
+
+
 def test_writer_exposes_claim_and_outcome_counters():
     outbox = InMemoryLedgerOutbox(instance_key="inst-a")
     ledger = InMemoryLedger()
@@ -112,3 +130,81 @@ def test_writer_exposes_claim_and_outcome_counters():
     assert stats.completed == 1
     assert writer.counters["claim:claimed"] == 1
     assert writer.counters["outcome:completed"] == 1
+
+
+@pytest.mark.parametrize("instance_key", ["", "   "])
+def test_writer_rejects_explicit_empty_instance_key(instance_key):
+    outbox = InMemoryLedgerOutbox(instance_key="inst-a")
+    with pytest.raises(ValueError, match="non-empty"):
+        LedgerWriter(outbox, InMemoryLedger(), instance_key=instance_key)
+
+
+@pytest.mark.parametrize("instance_key", ["", "   "])
+def test_outbox_rejects_empty_instance_key(instance_key):
+    with pytest.raises(ValueError, match="non-empty"):
+        InMemoryLedgerOutbox(instance_key=instance_key)
+
+
+def test_in_memory_claim_is_exclusive_and_stale_lease_is_fenced():
+    now = {"t": datetime(2026, 6, 4, tzinfo=UTC)}
+    outbox = InMemoryLedgerOutbox(
+        instance_key="inst-a",
+        claim_ttl_sec=5.0,
+        clock=lambda: now["t"],
+    )
+    item = outbox.enqueue(_trade(execution_id="e-lease"))
+
+    [claim_a] = outbox.claim_pending()
+    assert outbox.claim_pending() == []
+    assert outbox.list_pending() == []
+
+    now["t"] += timedelta(seconds=6)
+    [claim_b] = outbox.claim_pending()
+    assert claim_b.claim_expires_at != claim_a.claim_expires_at
+    with pytest.raises(OutboxClaimLostError):
+        outbox.mark_outcome(
+            item.correlation_id,
+            OutboxOutcome.COMPLETED,
+            expected_claim_expires_at=claim_a.claim_expires_at,
+        )
+
+    stored = outbox.mark_outcome(
+        item.correlation_id,
+        OutboxOutcome.COMPLETED,
+        expected_claim_expires_at=claim_b.claim_expires_at,
+    )
+    assert stored.status is WorkItemStatus.COMPLETED
+    assert stored.claim_expires_at is None
+
+
+def test_outcome_logs_do_not_render_exception_text(caplog):
+    sentinel = "TOKEN_SENTINEL_ACCOUNT_DSN_PAYLOAD"
+
+    class _FailingLedger:
+        def record_trade(self, _trade):
+            raise LedgerPermanentError(sentinel)
+
+    outbox = InMemoryLedgerOutbox(instance_key="inst-a")
+    writer = LedgerWriter(outbox, _FailingLedger(), instance_key="inst-a")
+    outbox.enqueue(_trade(execution_id="e-sensitive"))
+
+    with caplog.at_level("INFO"):
+        stats = writer.run_once()
+
+    assert stats.delivery_failed == 1
+    assert sentinel not in caplog.text
+
+
+def test_lost_claim_is_counted_without_false_completion():
+    class _ClaimLosingOutbox(InMemoryLedgerOutbox):
+        def mark_outcome(self, correlation_id, outcome, **kwargs):
+            raise OutboxClaimLostError(correlation_id)
+
+    outbox = _ClaimLosingOutbox(instance_key="inst-a")
+    outbox.enqueue(_trade())
+    writer = LedgerWriter(outbox, InMemoryLedger(), instance_key="inst-a")
+
+    stats = writer.run_once()
+
+    assert stats.claim_lost == 1
+    assert stats.completed == 0

tests/application/test_ledger_outbox_backoff.py

@@ -47,7 +47,7 @@ class _ProgrammableLedger:
 
 
 def _wire(behavior, *, clock, **writer_kw):
-    outbox = InMemoryLedgerOutbox(clock=clock)
+    outbox = InMemoryLedgerOutbox(clock=clock, instance_key="test-instance")
     ledger = _ProgrammableLedger(behavior)
     writer = LedgerWriter(outbox, ledger, clock=clock, rng=lambda: 0.0, **writer_kw)
     return outbox, ledger, writer

tests/application/test_messaging_event_publisher.py

@@ -22,7 +22,7 @@ def _wire(transport):
     svc = SlotService(repo)
     intents = IntentService(svc, repo, InMemoryIntentStore(),
                             resolve_symbol=lambda pid: "005930")
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     ledger = InMemoryLedger()
     writer = LedgerWriter(outbox, ledger)
     repo.reconcile_universe("default", {"PF-1": "SYM"})
@@ -203,7 +203,7 @@ def test_sell_fill_dlq_sentinel_absent_when_strategy_id_set():
     intents = IntentService(svc, repo, InMemoryIntentStore(),
                             resolve_symbol=lambda pid: "005930")
     from trex_core.application.ledger_outbox import InMemoryLedgerOutbox
-    outbox = InMemoryLedgerOutbox()
+    outbox = InMemoryLedgerOutbox(instance_key="test-instance")
     execution = ExecutionService(svc, repo, PaperBroker(fill_after_polls=0),
                                  ledger_outbox=outbox, event_emitter=emitter)
     repo.reconcile_universe("default", {"PF-1": "SYM"})

trex_core/adapters/persistence/postgresql.py

@@ -21,6 +21,7 @@ from uuid import UUID, uuid4
 
 from trex_core.application.ledger_outbox import (
     LedgerWorkItem,
+    OutboxClaimLostError,
     OutboxOutcome,
     WorkItemStatus,
 )
@@ -569,6 +570,9 @@ _OUTBOX_COLUMNS = (
     "created_at, updated_at, next_retry_at, venue, order_venue, instance_key, "
     "claim_expires_at"
 )
+_OUTBOX_RETURN_COLUMNS = ", ".join(
+    f"o.{column.strip()}" for column in _OUTBOX_COLUMNS.split(",")
+)
 
 
 class PostgreSQLLedgerOutbox:
@@ -578,12 +582,16 @@ class PostgreSQLLedgerOutbox:
         self,
         dsn: str,
         *,
-        instance_key: str = "default",
+        instance_key: str,
         claim_ttl_sec: float = 60.0,
         worker_id: str | None = None,
         clock=None,
         psycopg_module: Any = None,
     ) -> None:
+        if not isinstance(instance_key, str) or not instance_key.strip():
+            raise ValueError("instance_key must be a non-empty string")
+        if claim_ttl_sec <= 0:
+            raise ValueError("claim_ttl_sec must be positive")
         self._psycopg = _require_psycopg(psycopg_module)
         self._dsn = dsn
         self._instance_key = instance_key
@@ -594,6 +602,10 @@ class PostgreSQLLedgerOutbox:
     def _now(self) -> datetime:
         return self._clock()
 
+    @property
+    def instance_key(self) -> str:
+        return self._instance_key
+
     def enqueue(self, trade: LedgerTrade) -> LedgerWorkItem:
         with self._psycopg.connect(self._dsn, autocommit=False) as conn:
             with conn.cursor() as cur:
@@ -625,16 +637,17 @@ class PostgreSQLLedgerOutbox:
         return existing
 
     def list_pending(self, limit: int = 32) -> list[LedgerWorkItem]:
+        now = self._now()
         with self._psycopg.connect(self._dsn) as conn:
             with conn.cursor() as cur:
                 cur.execute(
                     f"SELECT {_OUTBOX_COLUMNS} FROM ledger_outbox "
                     "WHERE status = 'pending' "
                     "AND instance_key = %s "
-                    "AND (next_retry_at IS NULL OR next_retry_at <= now()) "
-                    "AND (claim_expires_at IS NULL OR claim_expires_at <= now()) "
+                    "AND (next_retry_at IS NULL OR next_retry_at <= %s) "
+                    "AND (claim_expires_at IS NULL OR claim_expires_at <= %s) "
                     "ORDER BY created_at, correlation_id LIMIT %s",
-                    (self._instance_key, limit),
+                    (self._instance_key, now, now, limit),
                 )
                 return [self._row_to_item(r) for r in cur.fetchall()]
 
@@ -647,7 +660,7 @@ class PostgreSQLLedgerOutbox:
         now = self._now()
         lease_until = now + timedelta(seconds=self._claim_ttl_sec)
         correlation_sql = ""
-        params: list[Any] = [self._instance_key]
+        params: list[Any] = [self._instance_key, now, now]
         if correlation_id is not None:
             correlation_sql = "AND correlation_id = %s "
             params.append(correlation_id)
@@ -659,8 +672,8 @@ class PostgreSQLLedgerOutbox:
                     "SELECT correlation_id FROM ledger_outbox "
                     "WHERE status = 'pending' "
                     "AND instance_key = %s "
-                    "AND (next_retry_at IS NULL OR next_retry_at <= now()) "
-                    "AND (claim_expires_at IS NULL OR claim_expires_at <= now()) "
+                    "AND (next_retry_at IS NULL OR next_retry_at <= %s) "
+                    "AND (claim_expires_at IS NULL OR claim_expires_at <= %s) "
                     f"{correlation_sql}"
                     "ORDER BY created_at, correlation_id "
                     "LIMIT %s "
@@ -670,7 +683,7 @@ class PostgreSQLLedgerOutbox:
                     "SET claim_owner = %s, claim_expires_at = %s, updated_at = %s "
                     "FROM candidates "
                     "WHERE o.correlation_id = candidates.correlation_id "
-                    f"RETURNING {_OUTBOX_COLUMNS}",
+                    f"RETURNING {_OUTBOX_RETURN_COLUMNS}",
                     params,
                 )
                 rows = [self._row_to_item(r) for r in cur.fetchall()]
@@ -759,6 +772,8 @@ class PostgreSQLLedgerOutbox:
                     f"{where}",
                     params,
                 )
+                if cur.rowcount != 1:
+                    raise OutboxClaimLostError(correlation_id)
                 conn.commit()
         row = self.get(correlation_id)
         assert row is not None

trex_core/application/ledger_outbox.py

@@ -66,6 +66,10 @@ class WorkItemStatus(str, Enum):
     DELIVERY_FAILED = "delivery_failed"    # dcy-015 — permanent or dead-lettered (terminal)
 
 
+class OutboxClaimLostError(RuntimeError):
+    """The caller no longer owns the lease required for an outcome update."""
+
+
 # Outcomes that move a row to the terminal delivery_failed state.
 _DELIVERY_FAILED_OUTCOMES = frozenset(
     {OutboxOutcome.PERMANENT_FAILURE, OutboxOutcome.DEAD_LETTERED}
@@ -111,6 +115,11 @@ class LedgerWorkItem:
 class LedgerOutbox(Protocol):
     """Durable work-item store backing the ledger writer."""
 
+    @property
+    def instance_key(self) -> str:
+        """Immutable owner partition for rows handled by this outbox."""
+        ...
+
     def enqueue(self, trade: LedgerTrade) -> LedgerWorkItem:
         """Enqueue a trade. Idempotent on correlation_id (returns the
         existing row unchanged if it is already present)."""
@@ -163,6 +172,7 @@ class WriterStats:
     retryable: int = 0
     conflicts: int = 0
     delivery_failed: int = 0   # dcy-015: permanent + dead-lettered terminals
+    claim_lost: int = 0
 
 
 class LedgerWriter:
@@ -202,7 +212,12 @@ class LedgerWriter:
         # dcy-033 / REQ-100: best-effort alarm emitter.
         self._alarm_emitter = alarm_emitter
         self._ledger_skip_store = ledger_skip_store
-        self._instance_key = instance_key
+        resolved_instance_key = outbox.instance_key if instance_key is None else instance_key
+        if not isinstance(resolved_instance_key, str) or not resolved_instance_key.strip():
+            raise ValueError("instance_key must be a non-empty string")
+        if resolved_instance_key != outbox.instance_key:
+            raise ValueError("writer instance_key must match outbox instance_key")
+        self._instance_key = resolved_instance_key
         self.counters: dict[str, int] = {}
 
     def _count(self, key: str) -> None:
@@ -224,7 +239,19 @@ class LedgerWriter:
             limit,
         )
         for item in items:
-            outcome = self._process(item)
+            if not self._validate_instance(item):
+                continue
+            try:
+                outcome = self._process(item)
+            except OutboxClaimLostError:
+                self._count("claim:lost")
+                stats.claim_lost += 1
+                _log.warning(
+                    "ledger outbox claim lost: instance_key=%s correlation_id=%s",
+                    self._instance_key,
+                    item.correlation_id,
+                )
+                continue
             if outcome is OutboxOutcome.COMPLETED:
                 stats.completed += 1
             elif outcome is OutboxOutcome.DUPLICATE_COMPLETED:
@@ -242,19 +269,7 @@ class LedgerWriter:
         item = self._outbox.get(correlation_id)
         if item is None:
             return None
-        if (
-            self._instance_key is not None
-            and item.instance_key is not None
-            and item.instance_key != self._instance_key
-        ):
-            self._count("claim:foreign_instance")
-            _log.info(
-                "ledger outbox claim: instance_key=%s claim_result=foreign_instance "
-                "correlation_id=%s owner_instance_key=%s",
-                self._instance_key,
-                correlation_id,
-                item.instance_key,
-            )
+        if not self._validate_instance(item):
             return None
         if item.status is WorkItemStatus.PENDING:
             claimed = self._outbox.claim_pending(1, correlation_id=correlation_id)
@@ -276,7 +291,28 @@ class LedgerWriter:
                 len(claimed),
                 correlation_id,
             )
-        return self._process(item)
+        try:
+            return self._process(item)
+        except OutboxClaimLostError:
+            self._count("claim:lost")
+            _log.warning(
+                "ledger outbox claim lost: instance_key=%s correlation_id=%s",
+                self._instance_key,
+                correlation_id,
+            )
+            return None
+
+    def _validate_instance(self, item: LedgerWorkItem) -> bool:
+        if item.instance_key == self._instance_key:
+            return True
+        self._count("claim:foreign_instance")
+        _log.warning(
+            "ledger outbox claim rejected: instance_key=%s "
+            "claim_result=foreign_or_unowned correlation_id=%s",
+            self._instance_key,
+            item.correlation_id,
+        )
+        return False
 
     def _backoff_delay(self, attempts: int) -> float:
         """Exponential backoff with jitter: min(base*2^(n-1), cap) + up to 25%."""
@@ -288,7 +324,6 @@ class LedgerWriter:
         # outbox row. Skip the adapter call.
         existing = self._outbox.get(item.correlation_id)
         if existing is not None and existing.status is WorkItemStatus.COMPLETED:
-            self._outbox.mark_outcome(item.correlation_id, OutboxOutcome.DUPLICATE_COMPLETED)
             self._log_outcome(item, OutboxOutcome.DUPLICATE_COMPLETED)
             return OutboxOutcome.DUPLICATE_COMPLETED
         try:
@@ -300,7 +335,7 @@ class LedgerWriter:
                 error=str(exc),
                 expected_claim_expires_at=item.claim_expires_at,
             )
-            self._log_outcome(item, OutboxOutcome.CONFLICT_BLOCKED, error=str(exc))
+            self._log_outcome(item, OutboxOutcome.CONFLICT_BLOCKED)
             # dcy-033 / REQ-100: CRITICAL alarm — 409 conflict, operator resolution needed.
             if self._alarm_emitter is not None:
                 self._alarm_emitter.publish_alarm(
@@ -320,11 +355,11 @@ class LedgerWriter:
                 error=str(exc),
                 expected_claim_expires_at=item.claim_expires_at,
             )
-            self._log_outcome(item, OutboxOutcome.PERMANENT_FAILURE, error=str(exc))
+            self._log_outcome(item, OutboxOutcome.PERMANENT_FAILURE)
             _log.warning(
-                "ledger delivery_failed (permanent): instance_key=%s correlation_id=%s reason=%s",
+                "ledger delivery_failed (permanent): instance_key=%s correlation_id=%s",
                 item.instance_key or self._instance_key,
-                item.correlation_id, exc,
+                item.correlation_id,
             )
             # dcy-033 / REQ-100: CRITICAL alarm — permanent failure, no retry.
             if self._alarm_emitter is not None:
@@ -346,14 +381,13 @@ class LedgerWriter:
                     error=str(exc),
                     expected_claim_expires_at=item.claim_expires_at,
                 )
-                self._log_outcome(item, OutboxOutcome.DEAD_LETTERED, error=str(exc))
+                self._log_outcome(item, OutboxOutcome.DEAD_LETTERED)
                 _log.warning(
                     "ledger delivery_failed (dead-letter, attempts=%d): "
-                    "instance_key=%s correlation_id=%s reason=%s",
+                    "instance_key=%s correlation_id=%s",
                     new_attempts,
                     item.instance_key or self._instance_key,
                     item.correlation_id,
-                    exc,
                 )
                 # dcy-033 / REQ-100: CRITICAL alarm — dead-lettered after max retries.
                 if self._alarm_emitter is not None:
@@ -375,7 +409,7 @@ class LedgerWriter:
                 next_retry_at=next_retry_at,
                 expected_claim_expires_at=item.claim_expires_at,
             )
-            self._log_outcome(item, OutboxOutcome.RETRYABLE_FAILURE, error=str(exc))
+            self._log_outcome(item, OutboxOutcome.RETRYABLE_FAILURE)
             return OutboxOutcome.RETRYABLE_FAILURE
         # dcy-018 / REQ-019/030 (option D): write back portfolio-hub authoritative
         # cash/quantity to the snapshot store on new+duplicate (OQ-2 b). *_after
@@ -386,7 +420,11 @@ class LedgerWriter:
         # duplicate_completed; the ledger has exactly one entry for this
         # execution_id either way (REQ-019 AC-1, AC-4).
         if outcome.result is LedgerWriteResult.DUPLICATE_OK:
-            self._outbox.mark_outcome(item.correlation_id, OutboxOutcome.DUPLICATE_COMPLETED)
+            self._outbox.mark_outcome(
+                item.correlation_id,
+                OutboxOutcome.DUPLICATE_COMPLETED,
+                expected_claim_expires_at=item.claim_expires_at,
+            )
             self._resolve_ledger_skip(item.trade)
             self._log_outcome(item, OutboxOutcome.DUPLICATE_COMPLETED)
             return OutboxOutcome.DUPLICATE_COMPLETED
@@ -403,8 +441,6 @@ class LedgerWriter:
         self,
         item: LedgerWorkItem,
         outcome: OutboxOutcome,
-        *,
-        error: str | None = None,
     ) -> None:
         self._count(f"outcome:{outcome.value}")
         msg = "ledger outbox outcome: instance_key=%s correlation_id=%s outcome=%s"
@@ -413,9 +449,6 @@ class LedgerWriter:
             item.correlation_id,
             outcome.value,
         )
-        if error is not None:
-            msg += " error=%s"
-            args = (*args, error)
         _log.info(msg, *args)
 
     def _resolve_ledger_skip(self, trade: LedgerTrade) -> None:
@@ -454,7 +487,7 @@ class LedgerWriter:
         except Exception:  # noqa: BLE001 - write-back must not block the ledger path
             _log.warning(
                 "snapshot cash/position write-back failed: asset_id=%s",
-                trade.asset_id, exc_info=True,
+                trade.asset_id,
             )
 
 
@@ -467,7 +500,8 @@ class InMemoryLedgerOutbox:
         self,
         *,
         clock: Optional[Callable[[], datetime]] = None,
-        instance_key: str = "default",
+        instance_key: str,
+        claim_ttl_sec: float = 60.0,
     ) -> None:
         import threading
 
@@ -477,7 +511,16 @@ class InMemoryLedgerOutbox:
         # dcy-015: shared clock lets tests align backoff (next_retry_at) with
         # list_pending's "now" deterministically.
         self._clock = clock or (lambda: datetime.now(tz=timezone.utc))
+        if not isinstance(instance_key, str) or not instance_key.strip():
+            raise ValueError("instance_key must be a non-empty string")
+        if claim_ttl_sec <= 0:
+            raise ValueError("claim_ttl_sec must be positive")
         self._instance_key = instance_key
+        self._claim_ttl_sec = claim_ttl_sec
+
+    @property
+    def instance_key(self) -> str:
+        return self._instance_key
 
     def _now(self) -> datetime:
         return self._clock()
@@ -508,8 +551,11 @@ class InMemoryLedgerOutbox:
                 row = self._rows[cid]
                 # dcy-015 / REQ-019: skip rows still in backoff so they don't
                 # occupy the FIFO batch (head-of-line protection, NFR-3).
-                if row.status is WorkItemStatus.PENDING and (
-                    row.next_retry_at is None or row.next_retry_at <= now
+                if (
+                    row.status is WorkItemStatus.PENDING
+                    and row.instance_key == self._instance_key
+                    and (row.next_retry_at is None or row.next_retry_at <= now)
+                    and (row.claim_expires_at is None or row.claim_expires_at <= now)
                 ):
                     out.append(row)
                     if len(out) >= limit:
@@ -522,16 +568,27 @@ class InMemoryLedgerOutbox:
         *,
         correlation_id: str | None = None,
     ) -> list[LedgerWorkItem]:
-        if correlation_id is None:
-            return self.list_pending(limit)
         now = self._now()
+        lease_until = now + timedelta(seconds=self._claim_ttl_sec)
         with self._lock:
-            row = self._rows.get(correlation_id)
-            if row is None or row.status is not WorkItemStatus.PENDING:
-                return []
-            if row.next_retry_at is not None and row.next_retry_at > now:
-                return []
-            return [row]
+            claimed: list[LedgerWorkItem] = []
+            candidates = [correlation_id] if correlation_id is not None else self._order
+            for cid in candidates:
+                row = self._rows.get(cid)
+                if row is None or row.status is not WorkItemStatus.PENDING:
+                    continue
+                if row.instance_key != self._instance_key:
+                    continue
+                if row.next_retry_at is not None and row.next_retry_at > now:
+                    continue
+                if row.claim_expires_at is not None and row.claim_expires_at > now:
+                    continue
+                updated = replace(row, claim_expires_at=lease_until, updated_at=now)
+                self._rows[cid] = updated
+                claimed.append(updated)
+                if len(claimed) >= limit:
+                    break
+            return claimed
 
     def pending_cash_delta(self) -> dict[str, float]:
         with self._lock:
@@ -557,6 +614,11 @@ class InMemoryLedgerOutbox:
     ) -> LedgerWorkItem:
         with self._lock:
             row = self._rows[correlation_id]
+            if expected_claim_expires_at is None:
+                if row.claim_expires_at is not None:
+                    raise OutboxClaimLostError(correlation_id)
+            elif row.claim_expires_at != expected_claim_expires_at:
+                raise OutboxClaimLostError(correlation_id)
             if outcome is OutboxOutcome.COMPLETED:
                 new_status = WorkItemStatus.COMPLETED
             elif outcome is OutboxOutcome.CONFLICT_BLOCKED:
@@ -578,6 +640,7 @@ class InMemoryLedgerOutbox:
                     next_retry_at if outcome is OutboxOutcome.RETRYABLE_FAILURE else None
                 ),
                 updated_at=self._now(),
+                claim_expires_at=None,
             )
             self._rows[correlation_id] = updated
             return updated