Shutting down 1 master node keeps pods up but control plane not reachable

[devedse]

Expected Behavior

I've just started playing with this setup and got everything working through ansible. I hosted a simple nginx container that stays online when I shutdown master node 0.

Current Behavior

The control plane (normally reachable through kubectl / openlens) is not reachable anymore though:

C:\XGitPrivate\k3s-ansible\example>kubectl get pods
Unable to connect to the server: dial tcp 10.88.21.10:6443: connectex: A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

Steps to Reproduce

1. Ran the ansible script
2. Shutdown master node 0

Context (variables)

I changed as little as possible, I'm running VM's.

Operating system: Proxmox

Hardware: Proxmox hosts

Variables Used

all.yml

---
k3s_version: v1.25.12+k3s1
# this is the user that has ssh access to these machines
ansible_user: root
#ansible_ssh_common_args: '-o StrictHostKeyChecking=no'
ansible_ssh_common_args: '-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null'
ansible_python_interpreter: auto_silent
systemd_dir: /etc/systemd/system

# Set your timezone
system_timezone: "Europe/Amsterdam"

# interface which will be used for flannel
flannel_iface: "eth0"

# apiserver_endpoint is virtual ip-address which will be configured on each master
apiserver_endpoint: "10.88.21.10"

# k3s_token is required  masters can talk together securely
# this token should be alpha numeric only
k3s_token: "abcawefawefawefawefawefawefawefawef"

# The IP on which the node is reachable in the cluster.
# Here, a sensible default is provided, you can still override
# it for each of your hosts, though.
k3s_node_ip: '{{ ansible_facts[flannel_iface]["ipv4"]["address"] }}'

# Disable the taint manually by setting: k3s_master_taint = false
k3s_master_taint: "{{ true if groups['node'] | default([]) | length >= 1 else false }}"

# these arguments are recommended for servers as well as agents:
extra_args: >-
  --flannel-iface={{ flannel_iface }}
  --node-ip={{ k3s_node_ip }}

# change these to your liking, the only required are: --disable servicelb, --tls-san {{ apiserver_endpoint }}
extra_server_args: >-
  {{ extra_args }}
  {{ '--node-taint node-role.kubernetes.io/master=true:NoSchedule' if k3s_master_taint else '' }}
  --tls-san {{ apiserver_endpoint }}
  --disable servicelb
  --disable traefik
extra_agent_args: >-
  {{ extra_args }}

# image tag for kube-vip
kube_vip_tag_version: "v0.5.12"

# metallb type frr or native
metal_lb_type: "native"

# metallb mode layer2 or bgp
metal_lb_mode: "layer2"

# bgp options
# metal_lb_bgp_my_asn: "64513"
# metal_lb_bgp_peer_asn: "64512"
# metal_lb_bgp_peer_address: "192.168.30.1"

# image tag for metal lb
metal_lb_speaker_tag_version: "v0.13.9"
metal_lb_controller_tag_version: "v0.13.9"

# metallb ip range for load balancer
metal_lb_ip_range: "10.88.21.40-10.88.21.60"

# Only enable if your nodes are proxmox LXC nodes, make sure to configure your proxmox nodes
# in your hosts.ini file.
# Please read https://gist.github.com/triangletodd/02f595cd4c0dc9aac5f7763ca2264185 before using this.
# Most notably, your containers must be privileged, and must not have nesting set to true.
# Please note this script disables most of the security of lxc containers, with the trade off being that lxc
# containers are significantly more resource efficent compared to full VMs.
# Mixing and matching VMs and lxc containers is not supported, ymmv if you want to do this.
# I would only really recommend using this if you have partiularly low powered proxmox nodes where the overhead of
# VMs would use a significant portion of your available resources.
proxmox_lxc_configure: false
# the user that you would use to ssh into the host, for example if you run ssh some-user@my-proxmox-host,
# set this value to some-user
proxmox_lxc_ssh_user: root
# the unique proxmox ids for all of the containers in the cluster, both worker and master nodes
proxmox_lxc_ct_ids:
  - 1020
  - 1021
  - 1030
  - 1031
  - 1032

Hosts

host.ini

[master]
10.88.21.20
10.88.21.21

[node]
10.88.21.30
10.88.21.31
10.88.21.32

# only required if proxmox_lxc_configure: true
# must contain all proxmox instances that have a master or worker node
[proxmox]
10.88.20.10

[k3s_cluster:children]
master
node

Possible Solution

• [ ] I've checked the General Troubleshooting Guide

[devedse]

When I restart master 0 again things start working again. When I then shutdown the 2nd node (master 1) things break down in the exact same way.

The website I'm hosting still remains up though.

[devedse]

Also, after shutting down node-0 shouldn't it reschedule the pods to a different node?:

[devedse]

Ah it seems a few minutes later it actually evicted the pods. Any way to shorten this duration to like 30 seconds?:

[devedse]

I think I fixed the main issue by creating 3 master nodes rather then 2. It seems 3 is the minimum for HA (maybe someone can confirm?)