Haproxy Helm Chart configmap strips because the use of |-

[bmm-alc]

Hello,

I observe on all cluster where I deployed haproxy the configMap which contains haproxy configuration make the pods fail to start.
I constantly get this error

 haproxy version is 3.3.5-f0a2d1bf5                                                                                                
 config : parsing [/usr/local/etc/haproxy/haproxy.cfg:193]: Missing LF on last line, file might have been truncated at position 31.
 config : Error(s) found in configuration file : /usr/local/etc/haproxy/haproxy.cfg                                                
 config : Fatal errors found in configuration. 

I found the configMap is defined this way

data:
  haproxy.cfg: |2-

    global
      log stdout format raw local0
      stats timeout 1s
      user haproxy
      group haproxy

      ...
(omited)

so the last line is stripped

the characters 2- is added but I don't know which component do this (the haproxy chart, helm, etcd, kubernetes).
using haproxy.cfg: |+ in the values.yaml files does not fix the issue, the configMap still has haproxy.cfg: |2-.

but I found if I modify the file templates/configmap.yaml this way it works

diff --git a/haproxy/templates/configmap.yaml b/haproxy/templates/configmap.yaml
index e1847b2..35c8b83 100644
--- a/haproxy/templates/configmap.yaml
+++ b/haproxy/templates/configmap.yaml
@@ -23,7 +23,7 @@ metadata:
   labels:
   {{- include "haproxy.labels" . | nindent 4 }}
 data:
-  {{ .Values.configMount.subPath | default "haproxy.cfg" }}: |
+  {{ .Values.configMount.subPath | default "haproxy.cfg" }}: |+
   {{ tpl .Values.config . | nindent 4 }}
 {{- end }}

Do you have an idea on such bug.

[bmm-alc]

I forgot to say we use fluxcd to deploy manifests. We use it with layers where base haproxy configuration hold the main configuration

---
apiVersion: helm.toolkit.fluxcd.io/v2
kind: HelmRelease

metadata:
  name: haproxy-front

spec:
  interval: 10m
  chart:
    spec:
      chart: haproxy
      sourceRef:
        kind: HelmRepository
        name: haproxy
        namespace: flux-system
      version: "1.28.*"
  values:
    # HAProxy configuration
    kind: Deployment
    replicaCount: 3

and ./prd/ with a patch that declare only the config key, this way

---
apiVersion: helm.toolkit.fluxcd.io/v2
kind: HelmRelease
metadata:
  name: haproxy
spec:
  values:
    config: |+
      global
        log stdout format raw local0
        stats timeout 1s
        user haproxy
        group haproxy

      defaults
        log global
        mode http
        timeout connect 2s
(omitted)

I checked it does not affect the configuation with the flux build kustomization I get

(ommited)
        - labelSelector:
            matchLabels:
              app.kubernetes.io/instance: haproxy-front
              app.kubernetes.io/name: haproxy
          topologyKey: kubernetes.io/hostname
    config: |
      global
        log stdout format raw local0
        stats timeout 1s
        user haproxy
        group haproxy

(omitted)

[dkorunic]

Hi @bmm-alc, block-chomping characters |- present in your configuration for config are definitely not coming from the chart. I suspect this might be FluxCD issue, as this issue is not present in the normal Helm Chart when rendered through helm directly:

$ helm template ./haproxy | grep -A3 '^data:'
data:
  haproxy.cfg: |

    global
...

[bmm-alc]

hi @dkorunic

I suspect this might be FluxCD issue, as this issue is not present in the normal Helm Chart when rendered through helm directly:

$ helm template ./haproxy | grep -A3 '^data:'
data:
  haproxy.cfg: |

    global
...

as I said in my previous comment, building manifests locally does not show the chomping characters either.

[dkorunic]

@bmm-alc I am not sure either if it is FluxCD or some underlying set of libraries that does YAML normalisation, but your suggested fix is harmless and a perfectly fine workaround -- so I'm releasing fixed Helm Chart v1.28.1 with that fix as we speak.

[bmm-alc]

@bmm-alc I am not sure either if it is FluxCD or some underlying set of libraries that does YAML normalisation, but your suggested fix is harmless and a perfectly fine workaround -- so I'm releasing fixed Helm Chart v1.28.1 with that fix as we speak.

Thanks !