[BUG] No routing to Upstream Fallback Server

[mmeissner]

Describe the bug
I tried to add an Upstream Fallback Server for my Proxmox Managment UI.
I run a cluster of 2 Proxmox instances, the Managment UI can be reached from both servers.
I want Zoraxy to use the other server if the primary is down, therefor I configured a fallback server as upstream.
However Zoraxy does ignore the fallback upstream server and does show a 521 error on access.

Expected behavior
Zoraxy routes to fallback upstream when the primary upstream is down

Settings

Zoraxy knows that the primary is down

** Version **
Zoraxy v.3.1.6 on Debian amd64 Linux

[tobychui]

Huh interesting, as I recall I have tested this working a few releases back and never touched it. Will check it out next week to see what is the issue here.

[michael-r-elp]

Having the same issue here also an upstream server that requires TLS (HTTPS) and it's not falling back even though it correctly detects the main upstream as down.
Though in my case Zoraxy then shows a 404 Not found as message when trying to access the server.

[tobychui]

Ok I found the issue. The issue is that there was a PR that breaks the passive fallback load balancing logic (i.e. only switch to another upstream if the connecting one failed). I guess you guys are expecting it to be active fallback lb (i.e. automatically stop using an upstream when Zoraxy detects it is offline) as both of you screenshot me the uptime monitor.

Maybe I can take some time this weekend to implement both and see how it goes.

[michael-r-elp]

Well at least I showed the uptime monitor assuming that whatever mechanism it used there to determine if a upstream server is up would also be used when trying to connect to one, so it would for example first try connecting to the main upstream and if all the active upstreams fail finally use the fall back one if specified.
Though Zoraxy doesn't really make it clear whether the uptime monitor data is taken into account when determining which upstream to pick when someone attempts a connection. Though that does sound a bit like something that might make sense, but doesn't have to be the mechanism that's used in the end as long as the goal of someone connects and gets a response from one of the configured upstreams is achieved.

Whether that's through an active or passive fallback wouldn't matter to me.