fix(km): recover primary knowledge readback via direct fallback

apps/api/tests/test_knowledge_service_readback_degraded.py

@@ -3,9 +3,11 @@ import asyncio
 import pytest
 
 from src.models.knowledge import (
+    CategoryCount,
     EntrySource,
     EntryStatus,
     EntryType,
+    KnowledgeAssetTaxonomyCount,
     KnowledgeEntry,
     KnowledgeListResponse,
 )
@@ -148,6 +150,71 @@ async def test_knowledge_list_entries_bounds_primary_timeout_to_source_backed(mo
     assert response.total == 13
 
 
+@pytest.mark.asyncio
+async def test_knowledge_list_entries_uses_direct_readback_after_session_pool_timeout(monkeypatch) -> None:
+    service = KnowledgeService.__new__(KnowledgeService)
+    primary_calls = 0
+    direct_calls = 0
+    primary_entry = KnowledgeEntry(
+        id="km-primary-live-1",
+        title="Primary KM live row",
+        content="Persistent KM row recovered through direct readback",
+        entry_type=EntryType.RUNBOOK,
+        category="alert_handling",
+        tags=["telegram", "ai_agent", "log"],
+        source=EntrySource.AI_EXTRACTED,
+        status=EntryStatus.APPROVED,
+    )
+
+    async def pool_exhausted_primary(**_kwargs):
+        nonlocal primary_calls
+        primary_calls += 1
+        raise RuntimeError("db pool exhausted")
+
+    async def direct_readback(**kwargs):
+        nonlocal direct_calls
+        direct_calls += 1
+        assert kwargs["limit"] == 50
+        return KnowledgeListResponse(
+            items=[primary_entry],
+            total=727,
+            categories=[
+                CategoryCount(category="alert_handling", count=113),
+                CategoryCount(category="AI自動化/Ansible受控修復", count=87),
+            ],
+            asset_taxonomy=[
+                KnowledgeAssetTaxonomyCount(key="log", count=727),
+                KnowledgeAssetTaxonomyCount(key="alert", count=443),
+                KnowledgeAssetTaxonomyCount(key="mcp", count=36),
+            ],
+            readback_status="ready_direct_connection_after_session_timeout",
+            primary_readback_ready=True,
+            operator_stage="knowledge_readback_direct_connection_recovered",
+            next_step="repair_session_pool_readback_without_hiding_primary_km",
+            writes_on_read=False,
+            manual_review_required=False,
+        )
+
+    monkeypatch.setattr(service, "_list_entries_from_primary", pool_exhausted_primary)
+    monkeypatch.setattr(service, "_list_entries_from_direct", direct_readback)
+    monkeypatch.setattr(knowledge_service_module, "_PRIMARY_KM_RETRY_DELAY_SECONDS", 0)
+
+    response = await service.list_entries(limit=50)
+
+    assert primary_calls == 2
+    assert direct_calls == 1
+    assert response.total == 727
+    assert response.items == [primary_entry]
+    assert response.readback_status == "ready_direct_connection_after_session_timeout"
+    assert response.primary_readback_ready is True
+    assert response.operator_stage == "knowledge_readback_direct_connection_recovered"
+    assert response.next_step == "repair_session_pool_readback_without_hiding_primary_km"
+    assert response.categories[0].category == "alert_handling"
+    assert response.asset_taxonomy[0].key == "log"
+    assert response.writes_on_read is False
+    assert response.manual_review_required is False
+
+
 @pytest.mark.asyncio
 async def test_knowledge_list_entries_keeps_primary_items_when_taxonomy_degrades(monkeypatch) -> None:
     primary_entry = KnowledgeEntry(
@@ -247,6 +314,30 @@ async def test_knowledge_categories_fails_soft_when_readback_breaks(monkeypatch)
     assert all(row.count == 1 for row in categories)
 
 
+@pytest.mark.asyncio
+async def test_knowledge_categories_use_direct_readback_when_pool_is_busy(monkeypatch) -> None:
+    service = KnowledgeService.__new__(KnowledgeService)
+
+    async def pool_exhausted_categories():
+        raise RuntimeError("db pool exhausted")
+
+    async def direct_categories():
+        return [
+            CategoryCount(category="alert_handling", count=113),
+            CategoryCount(category="AI自動化/Ansible受控修復", count=87),
+        ]
+
+    monkeypatch.setattr(service, "_read_primary_categories", pool_exhausted_categories)
+    monkeypatch.setattr(service, "_read_categories_direct", direct_categories)
+
+    categories = await service.get_categories()
+
+    assert [(row.category, row.count) for row in categories] == [
+        ("alert_handling", 113),
+        ("AI自動化/Ansible受控修復", 87),
+    ]
+
+
 @pytest.mark.asyncio
 async def test_knowledge_asset_taxonomy_fails_soft_when_readback_breaks(monkeypatch) -> None:
     monkeypatch.setattr(
@@ -275,6 +366,32 @@ async def test_knowledge_asset_taxonomy_fails_soft_when_readback_breaks(monkeypa
     assert all(row.count >= 1 for row in taxonomy)
 
 
+@pytest.mark.asyncio
+async def test_knowledge_asset_taxonomy_uses_direct_readback_when_pool_is_busy(monkeypatch) -> None:
+    service = KnowledgeService.__new__(KnowledgeService)
+
+    async def pool_exhausted_taxonomy():
+        raise RuntimeError("db pool exhausted")
+
+    async def direct_taxonomy():
+        return [
+            KnowledgeAssetTaxonomyCount(key="log", count=727),
+            KnowledgeAssetTaxonomyCount(key="alert", count=443),
+            KnowledgeAssetTaxonomyCount(key="mcp", count=36),
+        ]
+
+    monkeypatch.setattr(service, "_read_primary_asset_taxonomy", pool_exhausted_taxonomy)
+    monkeypatch.setattr(service, "_read_asset_taxonomy_direct", direct_taxonomy)
+
+    taxonomy = await service.get_asset_taxonomy()
+
+    assert [(row.key, row.count) for row in taxonomy] == [
+        ("log", 727),
+        ("alert", 443),
+        ("mcp", 36),
+    ]
+
+
 @pytest.mark.asyncio
 async def test_knowledge_side_readbacks_bound_timeouts_to_source_backed(monkeypatch) -> None:
     service = KnowledgeService.__new__(KnowledgeService)