Unable to get worker running remotely

[AshifulBhuiyan]

I am not able to get worker running (waiting for worker permanently) on my personal website. But it works when I run locally (0.0.0.0). This is the log I get repeating when I run mercury run --verbose

Just spams
log not found : /ws/client/id

Could it be my nginx config? I didn't do anything special to it except for a proxypass to the port

---

EDIT

another note, when I use pip install mljar-mercury instead of just mercury. It just all works fine.

[Piotr]

Hi @AshifulBhuiyan,

Hard to say ... could you please send your nginx config?

[AshifulBhuiyan]

worker_processes 1;

events {
   worker_connections 1024; 
}

http {
include mime.types;
default_type application/octet-stream;
map $http_upgrade $connection_upgrade { default upgrade; '' close; }
sendfile on;
keepalive_timeout 65;



server {
   server name website.org;
   location / {
   proxy_set_headers Host $host;
   proxy_set_header X-Real-IP $remote_addr;
   proxy_set_header X-Forwarded-For $proxy_add_x_forawareded_for;
   aio threads;
   proxy_set_header X-Forwarded-Proto $scheme;

   proxy_pass http://0.0.0.0:8000;
}
}

[AshifulBhuiyan]

I managed to replicate the error on https://huggingface.co/spaces/ConquestAce/mathcodes

so most likely not the case of nginx

[Piotr]

Is you website public? What is the address? Can I look into web console logs? Hard to say where is the issue ...

The mercury package is the new version (V2) of Mercury, the mljar-mercury is old version (V1). Please use V2.

[AshifulBhuiyan]

I am using v2, also pages.mathcodes.org

[Piotr]

In HF Spaces Mercury cant connect to websocket (two red icons) - I was trying to debug the issue, but without any success ... It must be something with HF cloud settings.

I understand that in your case you have only one red icon?

[Piotr]

OK, I see that you have two red icons as well ...

[AshifulBhuiyan]

WebSocket:Disconnected and
Worker Unknown
Session ID: id

[Piotr]

@AshifulBhuiyan yes, similar problem as in HF ...

To summarize:

• when you run docker-compose locally it is all working
• when you deploy to cloud then it can't connect ...

I understand that you followed the docs to setup docker-compose https://runmercury.com/docs/docker-compose/

Any ideas what might be the issue?

[AshifulBhuiyan]

I am not using docker right now. I haven't been able to get it to run locally whatsoever.
Only thing I did was:
pip install mercury
mercury run 0.0.0.0:8000
and have nginx reverse proxy that port,

It works perfectly fine on my machine locally, but for remote connections, the worker does not run.

I am going to take a guess and say it has to do with my nginx config not forwarding /ws/client/id

(but mljar-mercury works as intended with all the same settings, in huggingface and in my remote site)

[AshifulBhuiyan]

Fixed.

Issue was with nginx and the websocket not being tunneled properly before.

Needed to config nginx as such:

location /wsapp/ {
    proxy_pass http://wsbackend;
    proxy_http_version 1.1;
    proxy_set_header Upgrade $http_upgrade;
    proxy_set_header Connection "Upgrade";
    proxy_set_header Host $host;
}