Hi all,

I’ve been deploying a Flask + Gunicorn API on Railway, and I’m running into a frustrating problem where large JSON responses get truncated in production, but work perfectly fine in local dev.

Setup

• Backend: Flask app, running under Gunicorn

• Frontend: Gradio app making POST requests to /inference

• Deployment: Railway free tier (no custom proxy config)

• Procfile:

  web: gunicorn wsgi:app -b 0.0.0.0:$PORT --workers 2 --threads 4 --timeout 120
  

The Problem

When my model generates a long text response (multi-KB, Spanish UTF-8 output), the client receives only part of the JSON. This leads to errors like:

JSONDecodeError: Unterminated string starting at: line 1 column 16 (char 15)

If I log response.content, I see it ends mid-string — half a Base64 blob or mid-paragraph of text.

Locally, I can return responses >20k chars with no issue. In production on Railway, responses above some threshold (seems around a few KB) get truncated.

I also tested with a dummy /echo endpoint that returns "A" * 20000. Locally it works, in prod the body gets chopped.

What I’ve Tried

• Switched from Flask’s dev server to Gunicorn (Procfile).

• Increased Gunicorn limits: --timeout 120, --limit-request-line 0, --limit-request-field_size 0.

• Forced UTF-8 JSON (app.config["JSON_AS_ASCII"] = False).

• Even encoded messages in Base64 to avoid escaping issues.

Still, long responses are truncated.

Question

• Is this truncation a Railway proxy/body size limit (especially on the free tier)?

• If I upgrade my Railway plan, will that increase the allowed response size?

• Is there a config/setting I’m missing to let Railway pass through larger responses?

I want to know whether this is expected platform behavior, or if I’m doing something wrong with Flask/Gunicorn.

Any insights or advice would be greatly appreciated

Thanks!