Valheim GKE Server

This repo contains all things necessary to automate setting up a Valheim server in GKE.

Architecture

Requirements

Helm > v3.0
Terraform > v14.0

Getting started

Cluster setup

You can skip this if you already have a GKE cluster. Autopilot is not supported but may work. You can't add capabilities and this container modifies CAP_SYS_NICE so it may impact performance.

cp terraform.tfvars.example terraform.tfvars
edit terraform.tfvars (see below for inputs)
make init, make apply

Starting server

make start or ./start_server.sh

This script will invoke the Cloud Function to scale up the node pool, print logs if there is an error, otherwise wait for the pod to be marked as ready

Terraform Commands

make init
make plan
make apply
make docs - Generate TF docs

Inputs

Name	Description	Type	Default	Required
apis_required	a list of APIs to enable	`list(string)`	n/a	yes
artifactrepo	Name of AR to create	`string`	`"blank"`	no
backup_cf_memory	MB to give the backup function. Must exceed the size of your backup file	`number`	n/a	yes
backup_function_schedule	cron schedule for backup in UTC	`string`	`"0 9 * * *"`	no
bucket_name	Bucket name to create. Used to store world backups	`string`	n/a	yes
cloudscheduler_location	Location for cloud scheduler	`string`	n/a	yes
cluster_name	Name of GKE cluster	`string`	n/a	yes
container_name	Name of the container in the pod	`string`	n/a	yes
create_artifact_registry	n/a	`bool`	n/a	yes
dr_region	The DR region to host the DR bucket in	`string`	n/a	yes
ip_range_pods	The NAME of the secondary range to use for pods. Note this MUST exist already.	`string`	`"pods"`	no
ip_range_services	The NAME of the secondary range to use for services. Note this MUST exist already.	`string`	`"services"`	no
machine_type	Type to use for node pool	`string`	`"e2-standard-2"`	no
master_authorized_networks	A list of cidrs to allow to talk to the control plane	`list(object({ cidr_block = string, display_name = string }))`	n/a	yes
metrics_baseline	Threshold CPU value to scaledown. Defines cluster idle state. Uses container/cpu/core_usage_time metric	`string`	n/a	yes
namespace	Kubernetes namespace to use	`string`	`"default"`	no
network	The VPC network to host the cluster in	`string`	`"default"`	no
project	GCP Project to create resources in	`string`	n/a	yes
region	The region to host the cluster in	`string`	n/a	yes
scaledown_function_schedule	cron schedule for backup in UTC	`string`	`"0 * * * *"`	no
scaleup_users	A list of principals which are allowed to invoke the scale up function	`list(string)`	n/a	yes
subnetwork	The subnetwork to host the cluster in	`string`	`"default"`	no
zones	Which zones to make the cluster in. Set to a single zone for free tier	`list(string)`	n/a	yes

Server deployment

Get kubectl config

gcloud container clusters get-credentials (cluster name) --region (region)

Deploy helm chart It will prompt for a password so it is not stored in your history.

scripts/create_helm_server.sh (install|upgrade)

Server config data (world, config) is stored on a PVC and server backups are copied to GCS. This means it doesn't really matter if you lose your cluster you just need your backup zip.

Components

GKE Cluster and node pool - these run the workloads Backup Cloud Function + Cloud Scheduler job - grabs a daily backup and optionally copies it to a secondary bucket Scaledown Cloud Function + Cloud Scheduler job - checks the pod CPU metrics and determines when to scale node pool to 0 Scaleup Cloud Function - scales node pool to 1 upon invocation

Credits

This uses a modified helm chart from https://addyvan.github.io/valheim-k8s/ This uses the container from https://github.com/lloesche/valheim-server-docker

ScaleSec / valheim-gke-server