proxy_protocol option breaks health check about docker-bungeecord

[itzg]

I didn't realize bungeecord had an option for PROXY protocol. A change will be needed for both repos, so I also created itzg/mc-monitor#6

[itzg]

I have added a variable called HEALTH_HOST that defaults to localhost, but in your case could be changed to point at the HAproxy's host/IP address. Would that work?

[nidotls]

I don't think this is a solution, because the HAProxy distributes the requests to all proxy servers at the same time, meaning if one proxy is offline all are classified as unhealthy

[itzg]

Oh, so you have several equivalent replicas of bungeecord instances behind HAproxy?

[nidotls]

Exactly. Distributed across multiple hosts

[itzg]

This is getting harder than I thought. It'll take an enhancement in https://github.com/Raqbit/mc-pinger also -- so three repos will need to be touched.

[itzg]

FYI, have a PR submitted ^

[itzg]

The latest image now includes the variable HEALTH_USE_PROXY, which you'll want to set to "true".

[nidotls]

Got /bin/sh: 1: /health.sh: Permission denied

[itzg]

Oops, try the latest image now.

[nidotls]

I don't know if this is related but on our production system (without HAProxy) the newest image doesn't start anymore sudo: /usr/lib/jvm/java-1.8-openjdk/bin/java: command not found.

Please give me feedback if I should create an issue for this.

On the HAProxy system the healthcheck runs now without problems, many thanks!

[itzg]

Are you using Portainer or something similar? If so, that product has an issue with incorrectly retaining image declared variables like PATH and JAVA_HOME. As a result, image upgrades that involve a change to those variables (like switching to Java 11) are not handled properly.

[nidotls]

Has clarified, thank you very much!