How can Langflow be configured for high availability (HA)?

[lianlian1212]

How can Langflow be configured for high availability (HA)?

Version 1.4.2 still throws an error when starting 2 pods

Error Building Component
Network error. Please check the connection to the server.

[dosubot[bot]]

To configure Langflow version 1.4.2 for high availability without encountering a network error when starting two pods, ensure both instances are set up to share the same PostgreSQL database. Here's a step-by-step guide:

1. Configure the .env file with the necessary PostgreSQL database credentials and Langflow settings:

   POSTGRES_USER=langflow
   POSTGRES_PASSWORD=your_secure_password
   POSTGRES_DB=langflow
   POSTGRES_HOST=postgres
   POSTGRES_PORT=5432
   LANGFLOW_CONFIG_DIR=app/langflow
   LANGFLOW_PORT_1=7860
   LANGFLOW_PORT_2=7861
   LANGFLOW_HOST=0.0.0.0
   

2. Use the following docker-compose.yml configuration to deploy two Langflow instances (langflow-1 and langflow-2) that share the same PostgreSQL database:

   services:
     postgres:
       image: postgres:16
       environment:
         - POSTGRES_USER=${POSTGRES_USER}
         - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
         - POSTGRES_DB=${POSTGRES_DB}
       ports:
         - "${POSTGRES_PORT}:5432"
       volumes:
         - langflow-postgres:/var/lib/postgresql/data
   
     langflow-1:
       image: langflowai/langflow:latest
       pull_policy: always
       ports:
         - "${LANGFLOW_PORT_1}:7860"
       depends_on:
         - postgres
       environment:
         - LANGFLOW_DATABASE_URL=postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT}/${POSTGRES_DB}
         - LANGFLOW_CONFIG_DIR=${LANGFLOW_CONFIG_DIR}
         - LANGFLOW_HOST=${LANGFLOW_HOST}
         - PORT=7860
       volumes:
         - langflow-data-1:/app/langflow
   
     langflow-2:
       image: langflowai/langflow:latest
       pull_policy: always
       ports:
         - "${LANGFLOW_PORT_2}:7860"
       depends_on:
         - postgres
       environment:
         - LANGFLOW_DATABASE_URL=postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT}/${POSTGRES_DB}
         - LANGFLOW_CONFIG_DIR=${LANGFLOW_CONFIG_DIR}
         - LANGFLOW_HOST=${LANGFLOW_HOST}
         - PORT=7860
       volumes:
         - langflow-data-2:/app/langflow
   
   volumes:
     langflow-postgres:
     langflow-data-1:
     langflow-data-2:

This setup ensures both Langflow instances are connected to the same database, allowing for high availability. Verify that the network settings and ports are correctly configured to avoid network errors ^[1].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[lianlian1212]

When deploying Langflow in a Kubernetes cluster using langflow-helm-charts and configuring multiple Pods to share the same PostgreSQL database via environment variables, a network connection error occurs when starting two Pods.

#7914

[saber-wang]

Set environment variables 'LANGFLOW_EVENT_DELIVERY=direct'

[lianlian1212]

event_delivery EventDeliveryType
Possible values: [streaming, direct, polling]

Are there any differences between the three types? I don't see explain on the langflow website

[saber-wang]

@lianlian1212
The system internally uses an in-memory queue. Streaming and polling first create a task and then process new requests or task results, while direct mode creates a task and waits for the result directly.

[lianlian1212]

okay,thanks
If I use directly mode, will I get stuck doing multiple tasks at once?

[severfire]

as per 1.3.3 i experienced memory leaks, like every 2-4k api calls langflow shold be restarted - and i have ~24GB RAM.

[severfire]

as per latest nightly version, and as per latest release, HA right now is not possible with langflow due to memory leaks.

[lianlian1212]

Set environment variables 'LANGFLOW_EVENT_DELIVERY=direct'
https://github.com/langflow-ai/langflow/issues/7914
When pod = 2, network issues "Error Building Component Network error. Please check the connection to the server. " still exist.

langflow version:v1.4.3

2025/08/26 08:58:24 [error] 11#11: *704970 upstream prematurely closed connection while reading upstream, client: 10.42.0.160, server: , request: "POST /api/v1/build/74838bbc-bb5c-433e-9d89-2bc4670104e9/flow?start_component_id=ChatInput-BbGox&log_builds=true&event_delivery=direct HTTP/1.1", upstream: "http://10.99.148.138:7860/api/v1/build/74838bbc-bb5c-433e-9d89-2bc4670104e9/flow?start_component_id=ChatInput-BbGox&log_builds=true&event_delivery=direct", host: "xxx.com", referrer: "https://xxx.com/flow/74838bbc-bb5c-433e-9d89-2bc4670104e9/folder/639aa740-7d4f-4d07-9a36-84e430891950"