External haproxy

This example demonstrates how to configure HAProxy Ingress to manage an external haproxy instance deployed as a sidecar container. This approach decouple the controller and the running haproxy version, allowing the sysadmin to update any of them independently of the other.

Prerequisites

This document requires only a Kubernetes cluster. HAProxy Ingress doesn’t need to be installed, and if so, the installation process should use the Helm chart.

Configure the controller

The easiest and recommended way to configure an external haproxy is using the Helm chart with a customized values file. Create the haproxy-ingress-values.yaml file with the following content:

controller:
  config:
    syslog-endpoint: stdout
    syslog-format: raw
  haproxy:
    enabled: true
    securityContext:
      runAsUser: 0

These parameters are configuring an external haproxy and configuring haproxy to log to stdout. Also, haproxy is configured as root so it has permission to bind ports :80 and :443. By default haproxy container is started as UID 99.

A word about security

haproxy historically started as root so it has the permissions needed to bind to privileged ports, configure chroot, configure file descriptor limits, and other administrative tasks. haproxy then drops its own privileges just before starting its event loop. See Security Considerations from the documentation.

Since 2.4, haproxy container has been started as UID 99. There are a few ways to give it permissions to bind privileged port, none of them is provided by default by HAProxy Ingress Helm chart because all of them has some sort of limitation. Choose one of the options below that best suits the needs of your environment:

1. Configure haproxy to start as root, This is the configuration provided above, but it will not work if cluster policies deny containers running as root.

2. Some container runtime engines, like Docker 20.10 or newer, or Containerd embedded in k3s, reconfigure the starting of unprivileged ports so haproxy should work out of the box listening to :80 and :443 without the need to run as root. Give it a try by removing the securityContext configuration altogether:

   controller:
     config:
       syslog-endpoint: stdout
       syslog-format: raw
     haproxy:
       enabled: true

3. Change haproxy listening port to unprivileged ports, like 8080 and 8443:

   Note that, if exposing haproxy via hostNetwork, end users would need to connect to :8443 instead of the well known :443, so this is only an option if the cluster provides LoadBalancer services

   controller:
     config:
       syslog-endpoint: stdout
       syslog-format: raw
       http-port: "8080"
       https-port: "8443"
     service:
       httpPorts:
       - port: 80
         targetPort: 8080
       httpsPorts:
       - port: 443
         targetPort: 8443
       type: LoadBalancer
     haproxy:
       enabled: true

4. Change the haproxy image by adding the NET_BIND_SERVICE capability to the haproxy binary:

   FROM haproxy:X.X-alpine
   USER root
   RUN apk add -U libcap-utils
   RUN setcap 'cap_net_bind_service=+ep' /usr/local/sbin/haproxy
   USER haproxy

5. Reconfigure the start of unprivileged port to 80 or below using the following configuration:

   This configuration does not work if hostNetwork is configured as true, and does not work on Kernel versions older than 4.11.

   controller:
     config:
       syslog-endpoint: stdout
       syslog-format: raw
     haproxy:
       enabled: true
       securityContext:
         sysctls:
           name: net.ipv4.ip_unprivileged_port_start
           value: "1"

Install the controller

Add the HAProxy Ingress Helm repository if using HAProxy Ingress’ chart for the first time:

$ helm repo add haproxy-ingress https://haproxy-ingress.github.io/charts

Install or upgrade HAProxy Ingress using the haproxy-ingress-values.yaml parameters:

$ helm upgrade haproxy-ingress haproxy-ingress/haproxy-ingress\
  --install --create-namespace --namespace=ingress-controller\
  -f haproxy-ingress-values.yaml

Check if the controller successfully starts or restarts:

$ kubectl --namespace ingress-controller get pod -w

Test

Open two distinct terminals to follow haproxy-ingress and haproxy logs:

$ kubectl --namespace ingress-controller get pod
NAME                               READY   STATUS    RESTARTS   AGE
haproxy-ingress-6f8848d6fb-gxmrk   2/2     Running   0          13s

$ kubectl --namespace ingress-controller logs -f haproxy-ingress-6f8848d6fb-gxmrk -c haproxy-ingress

and

$ kubectl --namespace ingress-controller logs -f haproxy-ingress-6f8848d6fb-gxmrk -c haproxy

Do some curl to any exposed application, or just use the controller or service loadbalancer IP like the example below:

$ curl 192.168.1.11

HAProxy Ingress and the external haproxy should be logging their own events:

haproxy-ingress container:

...
I0117 17:30:27.282701       6 controller.go:87] HAProxy Ingress successfully initialized
I0117 17:30:27.282743       6 leaderelection.go:243] attempting to acquire leader lease  ingress-controller/ingress-controller-leader-haproxy...
I0117 17:30:27.335674       6 status.go:177] new leader elected: haproxy-ingress-6f8848d6fb-cxb6w
I0117 17:30:27.392372       6 controller.go:321] starting haproxy update id=1
I0117 17:30:27.392463       6 ingress.go:153] using auto generated fake certificate
I0117 17:30:27.437047       6 instance.go:309] haproxy successfully reloaded (external)
I0117 17:30:27.437217       6 controller.go:353] finish haproxy update id=1: parse_ingress=0.143483ms write_maps=0.149637ms write_config=0.971026ms reload_haproxy=43.498718ms total=44.762864ms
I0117 17:30:58.066768       6 leaderelection.go:253] successfully acquired lease ingress-controller/ingress-controller-leader-haproxy
I0117 17:30:58.066867       6 status.go:177] new leader elected: haproxy-ingress-6f8848d6fb-gxmrk

haproxy container:

...
192.168.1.10:61116 [17/Jan/2021:17:32:36.050] _front_http _error404/<lua.send-404> 0/0/0/0/0 404 190 - - LR-- 1/1/0/0/0 0/0 "GET / HTTP/1.1"

What was changed?

The sections below have details of what changed in the deployment compared with a default installation.

Sidecar

This example configures 2 (two) new containers in the controllers’ pod:

• haproxy is the external haproxy deployment with two mandatory arguments: -S with the master CLI unix socket, and -f with the configuration files path
• init, a Kubernetes’ initContainer used to create an initial and valid haproxy configuration.

The haproxy container references the official Alpine based image haproxy:2.3.12-alpine, but can be any other, provided that it isn’t older than the minimal required version. The minimal version for the external HAProxy can be found in the README file.

The init container just copy a minimum and valid haproxy.cfg. This file is used to properly starts haproxy and configures its master CLI that HAProxy Ingress uses to manage the instance.

A new command-line --master-socket was also added to the HAProxy Ingress container. This option enables an external haproxy instance, pointing to the unix socket path of its master CLI.

Shared filesystem

HAProxy Ingress sends configuration files to the haproxy instance using a shared filesystem. A Kubernetes’ emptyDir works well.

The following directories must be shared:

• /etc/haproxy: configuration and map files - init and haproxy-ingress need write access, haproxy need read access.
• /var/lib/haproxy: mostly ssl related files - haproxy-ingress need write access, haproxy need read access.
• /var/run/haproxy: unix sockets - haproxy-ingress and haproxy need write access.

Liveness probe

Default HAProxy Ingress deployment has a liveness probe to an haproxy’s health check URI. This example changes the liveness probe from the HAProxy Ingress container to the haproxy one.