Error while proxying, error: EOF

[frsantos]

Hi all,

We are migrating an application from Capistrano to Kamal, and everything seems fine. However, a few, very long requests always fail with a 502 error.

The application runs on Unicorn, and looking at the Rails logs it finishes properly (Completed 200 OK in 35123ms...), but as soon as the request finish, kamal-proxy shows the following log:

{
"time":"2025-09-23T14:52:35.341439256Z",
"level":"ERROR",
"msg":"Error while proxying",
"target":"8dd0dd95b40d:80",
"path":"/graphql/longRequest",
"error":"EOF"
}

{
"time":"2025-09-23T14:52:35.341652241Z",
"level":"INFO",
"msg":"Request",
"host":"staging.myserver.com",
"port":80,
"path":"/longRequest",
"request_id":"299df8b8-12c5-41c2-ac99-ae7afd8f37fa",
"status":502,
"service":"staging-web",
"target":"8dd0dd95b40d:80",
"duration":37545983590,
"method":"POST",
"req_content_length":24802,
"req_content_type":"application/json",
"resp_content_length":6214,
"resp_content_type":"text/html; charset=utf-8",
"client_addr":"10.189.52.35",
"client_port":"19022",
"remote_addr":"x.x.x.x",
"user_agent":"Mozilla/5.0 ...",
"proto":"HTTP/1.1",
"scheme":"http",
"query":"",
"req_cache_control":"",
"req_last_modified":"",
"req_user_agent":"Mozilla/5.0 ..."
}

We have noticed that it fails if the request takes more than 30 seconds, but we can't seem to change this timeout anywhere. I know 30 secs is too high, but until we optimise those pages we need to deploy as is.

We started setting the proxy -> response_timeout to a very high number (600), but to no avail. In the end we set all timeouts we could find to 600, but it still fails at 30 seconds.

• Unicorn: timeout 600 in config/unicorn/staging.rb
• Kamal: in deploy.staging.yml we have set response_timeout: 600 both in main proxy and web role proxy. Even heathcheck timeout is set to 600.

service: alkimii-staging

[...]

proxy:
  response_timeout: 600
  buffering: true

servers:
  web:
    hosts: ...
    env:
      clear:
        ROLE: web
    proxy:
      healthcheck:
        interval: 1
        path: /up
        timeout: 600
      response_timeout: 600
[...]

We have checked inside the kamal-proxy container that the file ~/.config/kamal-proxy/kamal-proxy.state has all changes applied.

What else have we missed?

Thanks in advance.

[kevinmcconnell]

@frsantos is there anything else running between Kamal Proxy and the Rails app that could be timing out the request?

For example, if you are using Thruster, it also defaults to a 30 second timeout.

[frsantos]

@kevinmcconnell Ah, very good point. As I only saw 2 containers (kamal-proxy and unicorn) and Thruster is so zero-config, I completely forgot about that one.

I'll try setting THRUSTER_HTTP_WRITE_TIMEOUT=600 too, and see.

Thanks!

[frsantos]

That was it, thanks!

[kevinmcconnell]

Excellent, glad to hear you got it sorted, and thanks for letting me know!