Architecture: LangGraph Persistence Backends: Redis, Postgres, MongoDB

Decision

Deploy a dedicated Redis Stack instance as the primary option, and provide PostgreSQL and MongoDB as additional backend choices for organizations with existing database infrastructure.

Alternative	Pros	Cons	Decision
Redis Stack + factory (primary)	Correct modules, clean separation, fast	Additional pod	Selected
PostgreSQL backend	ACID, relational, existing infra	Heavier, more setup	Selected
MongoDB backend	Document-native, flexible, Atlas/DocumentDB	Custom store wrapper	Selected
Upgrade rag-redis to Redis Stack	Reuse existing infra	Risk to RAG indexing	Rejected
Keep in-memory	Zero infra	State lost on restart	Rejected

Solution Architecture

Checkpointer Factory (checkpointer.py)

# Environment variables:
LANGGRAPH_CHECKPOINT_TYPE=memory    # memory (default) | redis | postgres | mongodb
LANGGRAPH_CHECKPOINT_REDIS_URL=     # Redis Stack connection string
LANGGRAPH_CHECKPOINT_POSTGRES_DSN=  # PostgreSQL DSN
LANGGRAPH_CHECKPOINT_MONGODB_URI=   # MongoDB connection URI
LANGGRAPH_CHECKPOINT_TTL_MINUTES=0  # 0 = no expiry

• create_checkpointer() returns the appropriate saver based on LANGGRAPH_CHECKPOINT_TYPE
• get_checkpointer() provides a global singleton
• Graceful fallback: if backend is unreachable or package missing, falls back to InMemorySaver

Store Factory (store.py)

# Environment variables:
LANGGRAPH_STORE_TYPE=memory         # memory (default) | redis | postgres | mongodb
LANGGRAPH_STORE_REDIS_URL=          # Redis Stack connection string
LANGGRAPH_STORE_POSTGRES_DSN=       # PostgreSQL DSN
LANGGRAPH_STORE_MONGODB_URI=        # MongoDB connection URI

Backend	Implementation	Package
Redis	_LazyAsyncRedisStore → AsyncRedisStore	langgraph-checkpoint-redis
Postgres	_LazyAsyncPostgresStore → AsyncPostgresStore	langgraph-checkpoint-postgres
MongoDB	_LazyAsyncMongoDBStore (custom, motor-based)	motor

Helm Chart (langgraph-redis)

• New subchart: charts/ai-platform-engineering/charts/langgraph-redis/
• Image: redis:8.0-alpine (Redis 8.0+ includes RedisJSON + RediSearch natively)
• PVC-backed persistence (2Gi default)
• Controlled via global.langgraphRedis.enabled condition
• Non-root container with security context hardening

Agent Wiring

• deep_agent.py: InMemorySaver() replaced with create_checkpointer()
• base_langgraph_agent.py: Module-level MemorySaver() replaced with get_checkpointer() singleton

Components Changed

Python

• ai_platform_engineering/utils/checkpointer.py (new) — Checkpointer factory (memory/redis/postgres/mongodb)
• ai_platform_engineering/utils/store.py — Store factory with _LazyAsyncRedisStore, _LazyAsyncPostgresStore, _LazyAsyncMongoDBStore
• ai_platform_engineering/multi_agents/platform_engineer/deep_agent.py — Uses create_checkpointer()
• ai_platform_engineering/utils/a2a_common/base_langgraph_agent.py — Uses get_checkpointer()
• pyproject.toml — Added langgraph-checkpoint-redis>=0.3.5, langgraph-checkpoint-postgres>=2.0.0, langgraph-checkpoint-mongodb>=0.3.0, motor>=3.4.0

Helm

• charts/ai-platform-engineering/charts/langgraph-redis/ (new) — Redis Stack subchart
• charts/ai-platform-engineering/Chart.yaml — Added langgraph-redis dependency
• charts/ai-platform-engineering/values.yaml — Added global.langgraphRedis.enabled, postgres/mongodb to checkpointPersistence and memoryPersistence
• charts/ai-platform-engineering/charts/supervisor-agent/values.yaml — Added postgres/mongodb to checkpointPersistence, mongodb to memoryPersistence
• charts/ai-platform-engineering/charts/supervisor-agent/templates/deployment.yaml — Postgres/MongoDB checkpoint and store env vars
• charts/ai-platform-engineering/charts/agent/values.yaml — Added postgres/mongodb to checkpointPersistence
• charts/ai-platform-engineering/charts/agent/templates/deployment.yaml — Postgres/MongoDB checkpoint env vars

Tests

• tests/test_checkpointer.py — 29 unit tests (config, factory, singleton for all backends)
• tests/test_redis_store.py — 36 unit tests (Redis, Postgres, MongoDB store wrappers + factory)

Related

• Spec: spec.md