Feature/control plane vip via haproxy #34

Merged

kasun merged 2 commits from feature/control-plane-vip-via-haproxy into main

2026-10-03 00:02:12 +02:00

Author	SHA1	Message	Date
kasun	19bf92d6bc	docs(k8s-bootstrap): document control-plane VIP and pfSense HAProxy setup Deploy k8s Bootstrap / Bootstrap k3s Cluster (pull_request) Skipped Details Deploy k8s Bootstrap / Pulumi Preview (pull_request) Successful in 40s Details The VIP and HAProxy frontend/backend aren't exposed by the pfsense provider, so they must be created by hand before this stack is deployed.	2026-10-02 23:52:32 +02:00
kasun	afac1f338f	feature(k8s-bootstrap): join nodes and export kubeconfig via control-plane VIP Every join and the exported kubeconfig previously pointed at master1Ip, so losing master-1 cut off API access even with etcd quorum intact. Use controlPlaneVipAddress (a pfSense IP Alias fronted by HAProxy across all three masters) instead, add it as a TLS SAN on master-1, and wait for the VIP's /healthz before any other node joins. The VIP is set up manually on pfSense rather than via a second kube-vip/ARP VIP: API access is too critical to depend on the same failover mechanism that serves LoadBalancer Services. Also strips trailing whitespace from the Longhorn disk mount script.	2026-10-02 23:52:32 +02:00

Author

SHA1

Message

Date

kasun

19bf92d6bc

docs(k8s-bootstrap): document control-plane VIP and pfSense HAProxy setup

Deploy k8s Bootstrap / Bootstrap k3s Cluster (pull_request) Skipped

Details

Deploy k8s Bootstrap / Pulumi Preview (pull_request) Successful in 40s

Details

The VIP and HAProxy frontend/backend aren't exposed by the pfsense
provider, so they must be created by hand before this stack is deployed.

2026-10-02 23:52:32 +02:00

kasun

afac1f338f

feature(k8s-bootstrap): join nodes and export kubeconfig via control-plane VIP

Every join and the exported kubeconfig previously pointed at master1Ip,
so losing master-1 cut off API access even with etcd quorum intact. Use
controlPlaneVipAddress (a pfSense IP Alias fronted by HAProxy across all
three masters) instead, add it as a TLS SAN on master-1, and wait for
the VIP's /healthz before any other node joins.

The VIP is set up manually on pfSense rather than via a second
kube-vip/ARP VIP: API access is too critical to depend on the same
failover mechanism that serves LoadBalancer Services.

Also strips trailing whitespace from the Longhorn disk mount script.

2026-10-02 23:52:32 +02:00