chat-agent-persistence

Chat Agent Persistence Guide

Chat 서비스의 데이터 영속화 레이어 (Checkpointer + Message Storage)

Architecture Overview

┌───────────────────────────────────────────────────────────────────────────┐
│                     Chat Persistence Architecture                          │
├───────────────────────────────────────────────────────────────────────────┤
│                                                                            │
│  ┌─────────────┐                                                          │
│  │ chat-worker │                                                          │
│  │ (LangGraph) │                                                          │
│  └──────┬──────┘                                                          │
│         │                                                                  │
│    ┌────┴────┐                                                            │
│    │         │                                                            │
│    ▼         ▼                                                            │
│  Checkpointer    Progress Events                                          │
│  (Multi-turn)    (Stage updates)                                          │
│    │                  │                                                   │
│    │                  │ XADD                                              │
│    ▼                  ▼                                                   │
│  ┌────────────┐  ┌──────────────────┐                                    │
│  │ PostgreSQL │  │  Redis Streams   │                                    │
│  │checkpoints │  │ chat:events:{0-3}│                                    │
│  └────────────┘  └────────┬─────────┘                                    │
│                           │                                               │
│              ┌────────────┼────────────┐                                 │
│              │            │            │                                 │
│              ▼            ▼            ▼                                 │
│      ┌─────────────┐ ┌─────────────┐ ┌──────────────────┐              │
│      │ eventrouter │ │chat-persist │ │  (other groups)  │              │
│      │   (group)   │ │  (group)    │ │                  │              │
│      └──────┬──────┘ └──────┬──────┘ └──────────────────┘              │
│             │               │                                           │
│             ▼               ▼                                           │
│      ┌─────────────┐ ┌──────────────┐                                  │
│      │event-router │ │chat-consumer │                                  │
│      │  (SSE fan)  │ │  (DB write)  │                                  │
│      └─────────────┘ └──────┬───────┘                                  │
│                             │                                           │
│                             ▼                                           │
│                      ┌──────────────┐                                  │
│                      │  PostgreSQL  │                                  │
│                      │chat.messages │                                  │
│                      └──────────────┘                                  │
│                                                                         │
└───────────────────────────────────────────────────────────────────────────┘

Two Persistence Paths

Quick Diagnosis

# 1. chat.messages 테이블 확인
kubectl exec -n postgres deploy/postgresql -- psql -U sesacthon -d ecoeco -c \
  "SELECT COUNT(*) FROM chat.messages;"

# 2. Checkpointer 상태 확인 (worker 로그)
kubectl logs -n chat deploy/chat-worker --tail=100 | grep -E "(CachedPostgresSaver|checkpointer|InMemory)"

# 3. Redis Streams consumer group 확인
kubectl exec -n redis rfr-streams-redis-0 -c redis -- redis-cli XINFO GROUPS chat:events:0

Common Issues

Issue 1: CachedPostgresSaver 초기화 실패

증상: InMemory checkpointer created (Redis fallback disabled)

로그:

[WARNING] CachedPostgresSaver failed, falling back to Redis only:
object _AsyncGeneratorContextManager can't be used in 'await' expression

원인: checkpointer.py:228에서 async context manager를 await로 호출

상세: See checkpointer.md

Issue 2: chat.messages 테이블 비어있음

증상: chat.conversations에 데이터 있지만 chat.messages는 0건

진단:

# Consumer group 확인 - "chat-persistence" 없으면 미배포
kubectl exec -n redis rfr-streams-redis-0 -c redis -- redis-cli XINFO GROUPS chat:events:0

원인: ChatPersistenceConsumer 미배포

상세: See message-consumer.md

Code Locations

Reference Files

• Checkpointer 상세: See checkpointer.md - Cache-Aside 패턴, LangGraph API
• Message Consumer 상세: See message-consumer.md - 배포 방법, Consumer Group
• 트러블슈팅 문서: docs/troubleshooting/chat-messages-not-persisted.md
• PostgreSQL 분석: docs/reports/postgres-chat-data-analysis.md