502 Bad Gateway on REST API queries (Suspected Retool Response Header Size Limit)

💬 Queries and Resources
1

Hi Retool Community and Support Team,

We are experiencing sudden 502 Bad Gateway errors on our REST API queries. Since there have been absolutely no changes to our backend infrastructure, we strongly suspect this is an issue with Retool's internal proxy limits regarding incoming Response Headers.

Here is the exact situation:

When we send a request, our backend successfully processes it and returns a 200 OK along with a very large Response Header (which includes heavy Set-Cookie and APM/Trace headers).

Crucial Evidence: Our infrastructure logs confirm that ONLY 200 OK responses are being sent back to Retool. There are zero 502 logs on our server side.

Despite our server successfully dispatching the response, the Retool UI throws a 502 Bad Gateway.

This strongly indicates that Retool's internal proxy is choking on the large Response Headers returning from our server and dropping the connection, generating the 502 internally.

[Question]
Has there been any recent update to Retool's proxy or Node.js environment that tightened the max_http_header_size or proxy buffer limits for incoming responses?

Could you please investigate this issue and let us know if there is any workaround or fix?

Thank you!

3

Hi there,

Our team is also facing the same issue since this morning, not “all” API queries are affected, but some such as POST to Apollo are returning this:

Query run failed.

    • message:"502 - "

Same as described above, everything was working fine until this morning.

Also tested with POSTMAN directly and our POST are working fine / returning a success when tested outside of Retool.

If you could please look into this urgently, this is affecting all our teams.

Thank you

Jerome

4

It has been confirmed that it is currently functioning normally.

1 Like
5

I'll continue to dig into this, @finn. Thanks for the update. :+1:

6

Thanks @Darren working on our side this morning too

1 Like
7

It looks like we identified and rolled back the breaking change - thanks again for reporting!