How does the WEB-Console work?

[Pseudo]

I am having problems with the integrated console of the WEB Gui resetting roughly every minute or so, which is giving me a bit of trouble.

When opening the GUI via calling up a nodes IP directly and opening the console, it works without issue.
The trouble only arises when we are calling the GUI over our loadbalancer.

To configure the loadbalancer accordingly, I want to understand how the process of opening the console in the GUI works.
I would be happy to read an explanation from someone who knows this in-depth.

Thanks to anyone who can provide me with the requested information.

 

[dcsapak]

basically the browser opens a seperate webpage that does some basic api calls (check if the guest is on the correct node, if it's turned on etc.) and then opens a websocket to the node where you did open the web ui, which is then possibly proxied (via ssh) to the correct node
the browser application talks a very simple protocol with the backend to send the keystrokes/receive the output of the terminal

for novnc it's similar but the vnc protocol is tunneled over the websocket

for the xtermjs code, it's here: https://git.proxmox.com/?p=pve-xtermjs.git;a=summary

 

[Robert.H]

What endpoint should be used as health-check for the proxmox servers? I'm looking for something like /live and /ready in the API but I got nothing...

 

[dcsapak]

What endpoint should be used as health-check for the proxmox servers? I'm looking for something like /live and /ready in the API but I got nothing...

what do you mean with health check, what would you like to check?

 

[Robert.H]

Having multiple proxmox instances (in various state) behind a load balancer and wanting to pass API request to live and ready instances only, I need a simple endpoint (return 200 OK) to show that the required services on each instance is in a good state.

A proxmox instance that has no (or lost) cluster membership or is in maintenance for example should not return 200 OK on the /ready endpoint.

All load balancers I've tried only have very simple http health-check (return code or string match) so I don't see parsing multiple endpoints for cluster membership and maintenance status.

Maybe there's a way to extend the pve-proxy with a script at location /health/ready that will call the various endpoints or shell cmds needed, parse the outputs and take care of returning a success or failure.

 

[dcsapak]

ok for cluster health there is e.g. /cluster/status but it doesn't return an error if it's not quorate but rather a status info (which you could probably string match?)
see https://pve.proxmox.com/pve-docs/api-viewer/#/cluster/status
there in the api-viewer are possibly other api calls that could be interesting for you

 

[Robert.H]

Unfortunetly the /cluster/status page is not string match friendly, not for my use case anyways.
Wondering if everyone is simply using / and don't care if they reach an isolated node.
Could you point me in the general direction to extend pveproxy to add a /health/ready endpoint?
I'll try to do it and submit a pr when done. Started looking into /usr/share/perl5/PVE/ and my head hurts... it's not for me.