Load balancer counters are reset to 0 if there are load balancer changes

[weizhouapache]

CloudStack uses haproxy in VR to implement the load balancer feature. If there are any load balancer changes (add/remove VMs, add/remove rules, etc), the haproxy configuration will be regenerated and haproxy service in VR will be reloaded.

We've found an issue that counters are reset to 0 after haproxy reload, which is incorrect because the existing connections are remain alive.

Before haproxy reload:

After haproxy reload:

An issue has been created in haproxy community: https://discourse.haproxy.org/t/stats-are-reset-to-0-after-haproxy-reload/8051

Users who monitor the load balancer connections, please pay attention to it.

ISSUE TYPE

• Bug Report

COMPONENT NAME

VR

CLOUDSTACK VERSION

4.17
4.18

CONFIGURATION

Advanced network load balancer

OS / ENVIRONMENT

SUMMARY

STEPS TO REPRODUCE

EXPECTED RESULTS

ACTUAL RESULTS

[weizhouapache]

A possible solution for this https://discourse.haproxy.org/t/stats-are-reset-to-0-after-haproxy-reload/8051/9

However, cloudstack VR 4.17.1.0/4.18.0 uses haproxy 2.2.9, which is too old.

(It is verified when haproxy is upgraded to 2.6.6-1, following https://haproxy.debian.net/#distribution=Debian&release=bullseye&version=2.6)

[rohityadavcloud]

Does that impact usage records @weizhouapache ? That may be a blocker/critical issue if our usage records are historically incorrect.

[weizhouapache]

@rohityadavcloud no. Cloudstack does not generate usage records for load balancer connections. It only impacts autoscaling feature.