Skip to content

Latest commit

 

History

History
322 lines (198 loc) · 20.4 KB

readme.md

File metadata and controls

322 lines (198 loc) · 20.4 KB
Raw

Complete Factorio Server Deployment (CloudFormation)

The template contained within this repository can be used to deploy a Factorio server to Amazon Web Services (AWS) in minutes. As the solution leverages "Spot Pricing", the server should cost less than a cent an hour to run, and you can even turn it off when you and your friends aren't playing - saving even more money.

If you wish to deploy multiple factorio servers with the one CloudFormation template (perhaps because of a team event or you just have different saves that you want to be able to pick and choose from), then a multi-server fork of this repository exists here: https://github.com/robertmassaioli/factorio-multi-server-spot-pricing

Prerequisites

  1. A basic understanding of Amazon Web Services.
  2. An AWS Account.
  3. Basic knowledge of Linux administration (no more than what would be required to just use the factoriotools/factorio Docker image).

Overview

The solution builds upon the factoriotools/factorio Docker image, so generously curated by the folks over at FactorioTools (thank you!).

In a nutshell, the CloudFormation template launches an ephemeral instance which joins itself to an Elastic Container Service (ECS) Cluster. Within this ECS Cluster, an ECS Service is configured to run a Factorio Docker image. The ephemeral instance does not store any saves, mods, Factorio config, data etc. - all of this state is stored on a network file system (Elastic File System - EFS).

The CloudFormation template is configured to launch this ephemeral instance using spot pricing. What is spot pricing you might ask? It's a way to save up to 90% on regular "on demand" pricing in AWS. There are drawbacks however. You're effectively participating in an auction to get a cheap instance. If demand increases and someone else puts in a higher bid than you, your instance will terminate in a matter of minutes.

A few notes on the services we're using...

  • EFS - Elastic File System is used to store Factorio config, save games, mods etc. None of this is stored on the server itself, as it may terminate at any time.
  • Auto Scaling - An Auto Scaling Group is used to maintain a single instance via spot pricing.
  • VPC - The template deploys a very basic VPC, purely for use by the Factorio server. This doesn't cost you a cent.

Getting Started

Launch Stack

  1. Click the above link, you'll need to log into your AWS account if you haven't already.
  2. Ensure you've selected a suitable AWS Region (closest to you) via the selector at the top right.
  3. Click Next to proceed through the CloudFormation deployment, provide parameters on the following page. You'll need a Key Pair and your Public IP address if you want to access the instance remotely via SSH (recommended). Refer to the Remote Access section below. There should be no need to touch any other parameters unless you have reason to do so. Continue through the rest of the deployment.

On Demand vs Spot

You may switch between On Demand / Spot via the InstancePurchaseMode CloudFormation parameter. When using Spot, it is not necessary to specify an InstanceType. Simply adjust the SpotMinMemoryMiB and SpotMinVCPUCount to specify how much Memory and CPU you would like on your instance. AWS will find you the cheapest spot instance available below the SpotPrice that you have specified. Should you wish to use a specific instance, you can specify it via the InstanceType parameter. If you are using "On Demand", you must specify the InstanceType.

Next Steps

All things going well, your Factorio server should be running in five minutes or so. Wait until CloudFormation reports the stack status as CREATE_COMPLETE. Go to the EC2 dashboard in the AWS console and you should see a Factorio server running. Take note of the public IP address. You should be able to fire up Factorio, and join via this IP address. No need to provide a port number, we're using Factorio's default. Bonus points - Public IP addresses are ugly. Refer to Custom Domain Name within Optional Features for a better solution.

At this point you should really configure remote access as per the below section, so that you can access the server and modify server-settings.json (e.g. add a password, add to the Factorio server browser, whitelist admins etc.).

Optional Features

Remote Access

You will likely want to SSH onto the Linux instance to make server changes / add a game password. You might also want to do this to upload your existing save. For security, SSH should be locked down to a known IP address (i.e. you), preventing malicious users from trying to break in. You'll need to create a Key Pair in AWS, find your public IP address, and then provide both of the parameters in the Remote Access (SSH) Configuration (Optional) section.

Note that this assumes some familiarity with SSH. The Linux instance will have a user ec2-user which you may connect to via SSH. If you want to upload saves, it's easiest to upload them to /home/ec2-user via SCP as the ec2-user user (this is ec2-user's home directory), and then sudo mv these files to the right location in the factorio installation via SSH.

For remote access, you'll need to:

  1. Create a Key Pair (Services > EC2 > Key Pairs). You'll need to use this to connect to the instance for additional setup.
  2. Find your public IP address. You'll need this to connect to the instance for additional setup.

If you're creating a new Factorio deployment, provide these parameters when creating the stack. Otherwise, update your existing stack and provide these parameters.

Uploading and Downloading an existing save.

Fast save upload (Recommended)

Warning: Makes sure that your server is live and all EC2 and ECS healthchecks are green before trying this.

Use the automation in util/upload-save.bash to upload your save file to your server, like so:

bash util/upload-save.bash ~/path/to/my/save.zip $your_ec2_ip_or_remote_name

Optionally, you can specify a path to a private key with a bash variable rather than relying on default ssh keys.

FACTORIO_PEM=~/path/to/my.pem bash util/upload-save.bash ~/path/to/my/save.zip $your_ec2_ip_or_remote_name

This is just an automated implementation of the slower version below.

Fast save download (Recommended)

Use the automation in util/download-latest-save.bash to download the latest (most recently played) save from a factorio server:

bash util/download-latest-save.bash $your_ec2_ip_or_remote_name

Optionally, you can specify a path to a private key with a bash variable rather than relying on default ssh keys.

FACTORIO_PEM=~/path/to/my.pem bash util/download-latest-save.bash $your_ec2_ip_or_remote_name

Your server needs to be running for this to work and it should download your latest save to your local directory.

Manual upload process (for understanding the system)

This procedure involves uploading your new save and then force killing the docker container. When the container is force killed it won't auto save, and the default logic is that on restart, the latest save will be loaded. To do this you must have SSH enabled via the CloudFormation deployment. The container must be running, otherwise you can't access EFS (where the save resides) from the EC2 instance.

  1. From your computer, upload your save to the EC2 instance. scp MySave.zip ec2-user@<my-domain-or-EC2-ip>:~/

  2. SSH into the EC2 instance. ssh ec2-user@<my-domain-or-EC2-ip>

  3. Identify the first 3 digits of the factorio docker container ID. We're doing this in advance, as we need to be quick in later steps. docker ps

Output will look something like this. The container ID is 19d3e1743e5c, so just note down 19d for future use.

CONTAINER ID   IMAGE                            COMMAND                  CREATED          STATUS                    PORTS                                                                                          NAMES
19d3e1743e5c   factoriotools/factorio:stable    "/docker-entrypoint.…"   2 minutes ago    Up 2 minutes              0.0.0.0:27015->27015/tcp, :::27015->27015/tcp, 0.0.0.0:34197->34197/udp, :::34197->34197/udp   ecs-factorio-EcsTask-i4J15601Hvkr-1-factorio-f6e3809ad5d6d8848501
01c68c702f42   amazon/amazon-ecs-agent:latest   "/agent"                 27 minutes ago   Up 26 minutes (healthy)                                                                                                  ecs-agent

  1. Find the save directory with the below command. savedir=$(mount | grep nfs4 | cut -f3 -d ' ' | xargs -I {} echo "{}/saves")

  2. Move your uploaded save to the right location. sudo mv ~/MySave.zip $savedir

  3. Touch your save to ensure it's timestamp is the latest out of all saves. If you take too long between this step and the next (and the server auto-saves), it will instead load the auto save. If this happens you got very unlucky... just try again. touch $savedir/MySave.zip

  4. Force kill the factorio docker container. It must be killed and not stopped, otherwise it will auto-save and load that save on restart. Use part of the container ID we noted before. docker kill 19d

  5. That should be it. Wait 30s for the container to restart, and when you connect it should load the save you just uploaded.

Custom Domain Name

Every time your Factorio server starts it'll have a new public IP address. This can be a pain to keep dishing out to your friends. If you're prepared to register a domain name (maybe you've already got one) and create a Route 53 hosted zone, this problem is easily fixed. You'll need to provide both of the parameters under the DNS Configuration (Optional) section. Whenever your instance is launched, a Lambda function fires off and creates / updates the record of your choosing. This way, you can have a custom domain name such as "factorio.mydomain.com". Note that it may take a few minutes for the new IP to propagate to your friends computers. Have patience. Failing that just go to the EC2 console, and give them the new public IP address of your instance.

FAQ

Do I need a VPC, or Subnets, or other networking config in AWS?

Nope. The stack creates everything you need.

What if my server is terminated due to my Spot Request being outbid?

Everything is stored on EFS, so don't worry you won't lose anything (well, that's partially true - you might lose up to 5 minutes of gameplay depending on when the server last saved). There is every chance your instance will come back in a few minutes. If not you can either select a different instance type, increase your spot price, or completely disable spot pricing and revert to on demand pricing. All of these options can be performed by updating your CloudFormation stack parameters.

My server keeps getting terminated. I don't like Spot Pricing. Take me back to the good old days.

That's fine; update your CloudFormation stack and set the SpotPrice parameter to an empty value. Voila, you'll now be using On Demand pricing (and paying significantly more).

How do I change my instance type?

Update your CloudFormation stack. Enter a different instance type.

What is the best instance type to run Factorio

For the source of truth, we can look at the minimum/recommended game settings for Factorio which says:

  • Processor Speed: 3Ghz+ minimum / 4Ghz+ recommended
  • Processor Cores: Quad Core
  • Memory: 8GB minimum / 16GB recommended

Given that, we can use a handy tool, like Cloud Price to find us the cheapest instances that also meet these requirements and sort them by Clock Speed.

At the time of writing this would indicate that the best instances are:

  • For Minimum Spec: m6a.large (2vCPUs, 8GB, 3.6Ghz) => USD$0.0317/hour spot
  • For Mid-Range Spec: m5zn.large (2vCPUs, 8GB, 4.5Ghz) => USD$0.0582/hour spot
  • For High-Range Spec: m5zn.xlarge (4vCPUs, 16GB, 4.5Ghz) => USD$0.1787/hour spot

It is recommended that you start off on minimum spec and then, when you notice that you need more power, stop your server, swap instance type, and start again. This is one of the best benefits of factorio spot pricing, being able to get a better computer to run your game at a moment's notice.

How do I change my spot price limit?

Update your CloudFormation stack. Enter a different limit.

I'm done for the night / week / month / year. How do I turn off my Factorio server?

Update your CloudFormation stack. Change the server state parameter from "Running" to "Stopped".

How do I turn my stack on and off from the terminal?

You can write a bash script, using the CLI like so:

#!/bin/bash

update_stack() {
    local state=$1
    aws cloudformation update-stack \
        --stack-name factorio-2024 \
        --use-previous-template \
        --parameters ParameterKey=ServerState,ParameterValue=$state \
        --capabilities CAPABILITY_IAM \
        --profile AdministratorAccess-111111111111
}

case "$1" in
    start)
        update_stack "Running"
        ;;
    stop)
        update_stack "Stopped"
        ;;
    *)
        echo "Usage: $0 {start|stop}"
        exit 1
        ;;
esac

If you put that in a file called update_factorio.bash then you could run:

$ bash update_factorio.bash <start|stop>

That does require that you run aws configure sso first and set up an IAM account with all the right perms.

I'm done with Factorio, how do I delete this server?

Delete the CloudFormation stack. Except for the EFS, Done. The EFS is retained when the CloudFormation stack is deleted to preserve your saves, but can then be manually deleted.

How can I upgrade the Factorio version?

Update your CloudFormation stack. Set the required tag value for FactorioImageTag. Use the tags specified here: https://hub.docker.com/r/factoriotools/factorio/. Your Factorio server will stop momentarily.

I'm running the "latest" version, and a new version has just been released. How do I update my server?

You can force a redeployment of the service via ECS. Update the service, and select Force new deployment.

How can I change map settings, server settings etc.

You'll need to have remote access to the server (refer to Optional Features). You can make whatever changes you want to the configuration in /opt/factorio/config. Once done, restart the container:

  1. Go to ECS (Elastic Container Service) in the AWS Console
  2. Click the factorio cluster
  3. Tick the factorio service, and select update
  4. Tick "Force new deployment"
  5. Click Next 3 times, and finally Update service

I can no longer connect to my instance via SSH?

Your public IP address has probably changed. Check your public IP address and update the stack, providing your new public IP address.

How do I load an existing save?

Be advised that whenever the Factorio container is terminated, it creates a new autosave just prior to terminating. For this reason, restarting the container directly on the host via SSH isn't advised.

In order to load an existing save, follow the below steps:

  1. Go to ECS (Elastic Container Service) in the AWS Console
  2. Click the factorio cluster
  3. Tick the factorio service, and select update
  4. Set Number of tasks to 0
  5. Click Next 3 times, and finally Update service
  6. Access the instance via SSH, placing your save in /opt/factorio/saves.
  7. Repeat steps above, setting the Number of tasks to 1.

What's Missing / Not Supported?

  • Scenarios - you can probably figure out a way to get this working... I've just never tried :-).

Expected Costs

The two key components that will attract charges are:

  • EC2 - If you're using spot pricing (and the m3.medium instance as per the default in the template), I doubt you would attract more than a cent an hour in fees for EC2. Even if you ran it 24 hours a day for a whole month, that's about 7 bucks.
  • EFS - Charged per Gigabyte stored per month (GB-Month). Varies based on region, but typically less than 50c per gigabyte. My EFS file system for Factorio is only about 100MB (incl. mods and 5 saves), so maybe 5 cents per month? To lower storage costs when not actively utilzied, items in EFS are automatically moved to Infrequent Access after 7 days and also moved back to Standard if subsequently accessed.

AWS do charge for data egress (i.e. data being sent from your Factorio server to clients), but again this should be barely noticeable.

Help / Support

This has been tested in both the Sydney and Oregon AWS regions (verify your AWS region of choice includes m3.medium @ https://aws.amazon.com/ec2/spot/pricing and/or change as needed). Your mileage may vary. If you get stuck, create an issue and myself or someone else may come along and assist.

Be sure to check out factoriotools's repositories on Docker Hub and GitHub. Unless your question is specifically related to the AWS deployment, you may find the information you're after there:

Stack gets stuck on CREATE_IN_PROGRESS

There might be multiple reasons.

Selected Instance Type not available in your region

NOTE: This should no longer be an issue when using the SpotMinMemoryMiB and SpotMinVCPUCount parameters instead of InstanceType.

It may be because there are no suitable Instance Types available in your region. As a result of this, an auto-scaling group gets successfully created - however it never launches an instance. This means the ECS service cannot ever start, as it has nowhere to place the container. I would suggest going to the AWS Console > EC2 > Spot Requests > Pricing History, and find a suitable instance type that's cost effective and has little to no fluctuation in price.

In the below example (Paris), m5.large looks like a good option. Try to create the CloudFormation stack again, changing the InstanceType CloudFormation parameter to m5.large. See: #10

Spot pricing history

Restarting the Container

Visit the ECS dashboard in the AWS Console.

  1. Clusters
  2. Click on the factorio Cluster
  3. Tick the factorio Service, click Update
  4. Tick the "Force new deployment" option
  5. Click Next step (three times)
  6. Click Update Service

Basic Docker Debugging

If you SSH onto the server, you can run the following commands for debugging purposes:

  • sudo docker logs $(docker ps -q --filter ancestor=factoriotools/factorio) - Check Factorio container logs.

DO NOT restart the Factorio docker container via SSH. This will cause ECS to lose track of the container, and effectively kill the restarted container and create a new one. Refer to Restarting the Container above for the right method.

Changelog

06-Oct-2024

  • Migrate from Launch Configuration to Launch Template, as Launch Configuration is unavailable in AWS accounts created after 01-Oct-2024.
  • Remove requirement for user to specify an instance type, but rather specify the Memory and vCPU that they require. AWS will figure out the best instance type.

Thanks

Thanks goes out to FactorioTools (and contributors) for maintaining the Factorio Docker images.

Thank you to all who have contributed to this repository.