Again: «Looks like your connection to NodeBB was lost, please wait while we try to reconnect.» - in Docker Swarm

mwaeckerlin

Ok, I fixed the problem and created a pull request. Please accept it as soon as possible.

This is then a working Docker Swarm compose file:

version: '3.3'
services:

  mongodb:
    image: mongo
    volumes:
      - type: bind
        source: /srv/volumes/forum-mrw-sh/mongodb
        target: /data/db

  nodebb:
    image: mwaeckerlin/nodebb
    ports:
      - 8036:4567
    volumes:
      - type: bind
        source: /srv/volumes/forum-mrw-sh/nodebb/config
        target: /usr/src/app/config
      - type: bind
        source: /srv/volumes/forum-mrw-sh/nodebb/uploads
        target: /usr/src/app/public/uploads

The image here is mwaeckerlin/nodebb, as soon as my fix has been pulled into your repository, that can be changed to nodebb/docker.

In the online-setup, chose MongoDB and thet the db-url to mongodb.

Tom Rade

@mwaeckerlin The way I solved it was to mount /usr/src/app/temp, then start the server, move all files to the temporary "temp" directory, then stop the server and change the mount point to /usr/src/app. That worked quite nicely.

I think all your plugins and some settings will get deleted once you restart the server if you only mount config and uploads.

mwaeckerlin

@Tom_Rade, mounting, stopping, moving, moving mountpoint is not the way how a proper docker installation should work.

Could you give me a list of all files that should remain persistent?

Yes, in my nextcloud docker image I had to persist the apps (=plugins) path too

PitaJ

@mwaeckerlin socket origin is restricted to prevent what is called "cross site websocket hijacking". This is when another site connects to the socket server and acts as if it's the NodeBB client.

The reason your detected origin is wrong is almost certainly a configuration issue, most likely in your reverse proxy setup. This is why I suggest trying a cut down nginx config, as nginx is responsible for passing down the origin header. proxy_redirect off is one of the order-dependent options.

But I didn't find any documentation on how to use nconf

nconf options are set within config.json, documented here:

config.json - NodeBB Documentation

Documentation portal for NodeBB Forum Software

(docs.nodebb.org)

Dev mode sets this to "*" which enables connections from anywhere.

Tom Rade

@mwaeckerlin I just did the entire /usr/src/app directory, not sure if that's correct - let me know if you think otherwise.

mwaeckerlin

@Tom_Rade, no, that's not correct. Docker (and clean programming) require a strict separation between code and data.

mwaeckerlin

@PitaJ, the problem is, that not the url setting is *, but the origin that arrives in socket.io.

With a reverse proxy, I can handle same origin in nginx, centralized before it passes through to NodeBB, even more: NodeBB always gets all requests from only one source, from the reverse proxy server. So I suggest that the same origin feature should be optional: On by default, but it should be possible to disable it in reverse proxy environments.

I see that the origin reported from socket.io is *, but I don't understand yet, why and how.

mwaeckerlin

@PitaJ, exactly which option in the suggested nginx configuration passes the origin to the backend NodeBB?

I want to understand. Up to now, everything sounds like voodoo, such as «certain options which are dependent on the order» or «almost certainly a configuration issue». So what exactly is the problem? Which options are responsible for setting the origin? Which options depend in the order? What is why the correct order? AFAIK, order is not relevant for all options. I would like to have an educated understanding of what happens and why, not something that works by coincidence.

PitaJ

@mwaeckerlin said in Again: «Looks like your connection to NodeBB was lost, please wait while we try to reconnect.» - in Docker Swarm:

@PitaJ, the problem is, that not the url setting is *, but the origin that arrives in socket.io.

Yes I understood your problem.

With a reverse proxy, I can handle same origin in nginx, centralized before it passes through to NodeBB, even more: NodeBB always gets all requests from only one source, from the reverse proxy server. So I suggest that the same origin feature should be optional: On by default, but it should be possible to disable it in reverse proxy environments.

Did you go to the link? You can configure the socket.io origins by adding the following to config.json:

"socket.io": {
  "origins": "*"
}

Them you can add the same origin check to nginx.

I see that the origin reported from socket.io is *, but I don't understand yet, why and how.

I'm no nginx wizard either. It's pretty much voodoo to me as well. That's why I suggested trying a cut down config. If a config based closely on the one in our docs worked, then it's likely possible to figure out what the issue is based on differences between them.

I don't have enough experience to diagnose exactly what is wrong.

exactly which option in the suggested nginx configuration passes the origin to the backend NodeBB?

I want to understand. Up to now, everything sounds like voodoo, such as «certain options which are dependent on the order» or «almost certainly a configuration issue». So what exactly is the problem? Which options are responsible for setting the origin? Which options depend in the order? What is why the correct order? AFAIK, order is not relevant for all options. I would like to have an educated understanding of what happens and why, not something that works by coincidence.

I don't know enough about nginx to answer these questions. I just know that our config does work, and yours doesn't.

One more thing: none of us core devs know much about docker, which is why the documentation and defaults aren't very good for it.

mwaeckerlin

Ok, I suggest two thing:

I'll help you to elaborate a Dockerfile that works fine and stable, also in swarm environments (next step will be OpenShift and upgrades).
It would be helpful to add logging information when socket.io rejects a connection, i.e. log the detected vs the configured origin.

Agree, @PitaJ, @Tom_Rade?