Troubleshooting Langflow in Hybrid Manager v1.4.0 (LTS)

Component configuration errors

Components show empty or invalid fields after importing a flow

Cause: Exported flow .json files don't include HM Machine User Keys, passwords, or Global Variable values. These are scoped to the Langflow instance they were exported from.

Fix:

1. In the new Langflow instance, go to Profile → Settings → Global Variables and create variables for the machine user key, database user, and database password using production values.

2. Open each EDB component and set HM Machine User Key to the correct Global Variable.

3. Reconfigure project, cluster, and database selections to point to the correct resources.

"Can't select a project" — portal domain defaults to the API endpoint

Cause: The EDB component is resolving the Hybrid Manager URL to the internal API endpoint instead of the frontend.

Fix: Select Controls at the top of the affected EDB Database or Knowledge Base component and set Hybrid Manager URL to:

http://istio-internalgateway.istio-system.svc.cluster.local:7080

EDB Knowledge Base component

Empty or irrelevant retrieval results

Symptoms: The component runs without error but returns no results, or returns results that don't match the query.

Diagnostic steps:

1. Verify the knowledge base exists and is populated:

   → WrapCopy

   SELECT * FROM aidb.kbstat WHERE name = 'your_knowledge_base_name';

2. Check that embeddings is greater than zero. If not, run:

   → WrapCopy

   SELECT aidb.run_pipeline('your_knowledge_base_name');

3. Confirm the Database Name and Knowledge Base fields in the component match exactly what was created in the cluster.

Common causes:

• Knowledge base was created but aidb.run_pipeline was never run.

• Source data was updated after the last embedding run — re-run aidb.run_pipeline to sync.

• The embedding model used to create the knowledge base doesn't match the one currently deployed. The model passed to aidb.knowledge_base_config(...) in aidb.create_pipeline(...) must correspond to a model cluster that's still active.

Connection failure to the Postgres cluster

Symptoms: Component shows a connection error when run.

Diagnostic steps:

1. Verify the machine user has Project Viewer role on the project:

   • Project → Users → Assign Project Roles → Project Viewer

2. Check that the Postgres cluster is active in the HM console.

3. Confirm Database Connection Type, User for the database, Password for the database, and Database Name all match what's configured in the cluster.

EDB Model Server component

Model cluster not available in the dropdown

Cause: No model cluster is deployed in the current HM project, or the machine user doesn't have the AI Model Manager role.

Fix:

1. Deploy a model cluster from Agent Factory → Model Library in the HM console. See Deploying a model cluster.

2. Confirm the machine user has AI Model Manager at organization level:

   • Profile → User management → Assign Organization Roles → AI Model Manager

Component runs but returns no response or an error

Symptoms: The EDB Model Server component finishes with an error, or returns an empty response.

Diagnostic steps:

1. Check the model cluster status in Agent Factory → Model Server Clusters. The cluster must show Active / Healthy before it can serve requests.

2. If the cluster is starting up (status: Pending), wait for it to become ready and retry.

3. Check the cluster's pod logs for initialization errors using the Hybrid Manager Grafana instance. See Observability for dashboard access.

4. Verify the Input field in the component is connected to an upstream component and isn't empty when the flow runs.

Context window exceeded

Symptoms: The model returns an error mentioning token limits or truncated input.

Fix: Reduce the amount of context being passed to the model. If the flow pipes EDB Knowledge Base results directly into the model:

• Lower the Limit results value on the EDB Knowledge Base component to retrieve fewer chunks.

• Use a chunking step (AIDB aidb.chunk_text_config) on source documents to reduce individual chunk size before indexing.

EDB Embeddings component

No model clusters available in the dropdown

Same cause and fix as EDB Model Server — see Model cluster not available in the dropdown above.

Embedding generation fails

Diagnostic steps:

1. Confirm the model cluster uses an embedding model (not an LLM) — the two are not interchangeable.

2. Check that the cluster is Active in Agent Factory → Model Server Clusters.

3. Review pod logs using the Hybrid Manager Grafana instance.

Flow-level issues

Flow works in Playground but fails after moving to another Hybrid Manager instance

Likely causes:

• Global Variables in the production Langflow instance aren't set up, or reference incorrect values.

• The production machine user has different (or missing) role assignments.

• The production knowledge base or model cluster has a different name than the development one.

Confirm every EDB component points to the correct project, cluster, and database, and that its Global Variables hold valid values. To run the flow as a managed service with parameters and credentials set at deploy time, see Flow deployment.

Flow runs but gives inconsistent or low-quality answers

Start with retrieval quality — most answer quality issues in RAG flows trace back to what the EDB Knowledge Base component is returning, not the model. See Empty or irrelevant retrieval results and the retrieval quality guidance in Langflow FAQ.

Further help

• Langflow FAQ — common questions about components, Global Variables, and flow deployment.
• Observability — Grafana dashboards and Prometheus monitoring for Langflow pod health and model cluster performance.
• Troubleshooting Agent Factory on Hybrid Manager — infrastructure-level issues (GPU, InferenceService, storage).