Why does my n8n workflow mix up messages between concurrent users?

Below is one possible approach to fixing your issue:

Ensure Unique Context per Execution

If multiple users are triggering the same workflow via a webhook, it’s important that each run carries a unique identifier. This way the workflow doesn’t “mix up” data between different orders or questions. For example, you can have your webhook URL or incoming request include a query parameter or a header (like a session or order ID) and then pass that value through your workflow. This makes sure every execution only works on its own context.

Check Your Workflow Design

Even though queue mode scales execution, the logic inside your workflow must be stateless or explicitly correlate one execution to one session. Consider:

• Storing: If you maintain any temporary state in a common resource—be it a temporary file, database, or even global variables in the workflow—you need to guarantee that they’re uniquely keyed.
• Splitting workflow paths: If one part of your workflow handles the incoming request and another part sends out responses, make sure they use the unique session data so that one user’s response is never delivered to another user.

Verify Queue Mode Configuration

Since you are running in queue mode, ensure that your configuration is complete:

• Shared Encryption Key: Verify that the same encryption key is set for all worker nodes (set via the environment variable N8N_ENCRYPTION_KEY), because sharing this key guarantees that sensitive execution data isn’t misinterpreted between workers.
• Webhook Processors: If your setup uses webhook processors (or if you use multiple main instances with webhooks), be sure the load balancer routes the requests correctly so that each request gets processed with its own context. (For example, see the documentation section on disabling webhook execution in the main process if you rely on dedicated webhook processors.)

Debug with Logs

Review your logs for messages indicating executions being added to or released from the queue. Look for any signs that workflow inputs might be getting combined or re-used unintentionally. This can help you pinpoint at which stage values are being mixed.

Summary

Your problem doesn’t seem to be an issue with n8n’s queue mode per se. Instead, it suggests that the current workflow design is allowing data from separate executions to be conflated. By ensuring that every incoming order or question is tagged with unique identifiers and by reviewing your state management and routing, you can prevent mix-ups across concurrent executions.

If needed, consider dividing the workflow into smaller parts with clear boundaries between processing incoming requests and sending responses so that each work item remains isolated.