High memory limit needed for starting HAProxy Ingress Controller

[Fagioletti]

Description

Working on HAProxy Ingress Controller for Kubernetes (3.2.6) i see reloads every few seconds if the memory allocation (resources.limits) is not above 500Mib. When running the footprint is very low, but if allocated less (<500Mib) HAProxy keeps on reloading every few seconds. It does not enter ‘normal’ operations. This can be seen without any load on the controller.

Environment

HAProxy Ingress Controller Version: v3.2.12
HAProxy Ingress Controller chart Version: 1.49.0
Kubernetes Version: 1.32.6
HAProxy Ingress Deployment Type: DaemonSet

Expected Behavior

HAProxy Ingress Controller starting with lower requests/limits

Actual Behavior

Anything below 500Mib in requests.limits prevents haproxy to properly start up.

Steps to Reproduce

Start a fresh install of HAProxy with memory limits below 500Mib.
Set resources like:

  resources:
    limits:
      memory: 256Mi
    requests:
      cpu: 200m
      memory: 256Mi

Additional context

To reduce the base consumption I set maxconns to 2020, as well as tune.ssl.cachesizeglobally.
HTTP/HTTPS frontends then have a limit of 1000 connections each.

# snippet from configmap
apiVersion: v1
kind: ConfigMap
data:
  global-config-snippet: |
    tune.ssl.cachesize 2020
  maxconn: "2020"

This seems not to reduce to memory needed for starting up.

Logs

# container logs (continuous reloads)

(...)
[NOTICE]   (70) : Reloading HAProxy
[NOTICE]   (70) : Initializing new worker (516)
[NOTICE]   (516) : haproxy version is 3.2.12-6011f448e
(...)
[NOTICE]   (70) : Loading success.
[WARNING]  (473) : Proxy healthz stopped (cumulated conns: FE: 0, BE: 0).
Proxy http stopped (cumulated conns: FE: 0, BE: 0).
[WARNING]  (473) : Proxy http stopped (cumulated conns: FE: 0, BE: 0).
[WARNING]  (473) : Proxy https stopped (cumulated conns: FE: 0, BE: 0).
[WARNING]  (473) : Proxy stats stopped (cumulated conns: FE: 0, BE: 0).
Proxy https stopped (cumulated conns: FE: 0, BE: 0).
[WARNING]  (473) : Proxy schux_svc_dudut_http stopped (cumulated conns: FE: 0, BE: 0).
Proxy schux_svc_dudut_http stopped (cumulated conns: FE: 0, BE: 0).
(...)
2026/03/30 07:51:42 INFO    controller/controller.go:227 [transactionID=455c455e-6cc4-4ad8-bd71-5698d7eddabd] HAProxy reloaded
[NOTICE]   (70) : haproxy version is 3.2.12-6011f448e
[WARNING]  (70) : Former worker (473) exited with code 0 (Exit)
(...)

Workaround

None

[Fagioletti]

Workaround/Fix

managed to reduce the memory needs by modifying nbthread and tune.memory.hot-size (since the workload is running on big nodes it used nbthread 48, compare: metric_name=haproxy_process_nbthread)

    global-config-snippet: |
      nbthread 2
      tune.memory.hot-size