Skip to content

Setup and run your own clusters on your own infrastructure in minutes (Eg. ECS or EKS). Roo is a zero config distributed ingress, edge-router & reverse-proxy (supporting multiple letsencrypt/https hosts) using Docker Swarm. No dependencies.

License

sfproductlabs/roo

master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
roo
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Roo

TL;DR

This basically lets you run your own encrypted and load balanced Amazon AWS clusters on your own hardware, and is a 5-minute replacement for Kubernetes, with no configuration (no additional setup for clustered kv stores, no janky config files, no defining providers, no dodgy second hand helm charts, no ssl setup, no manual ssl certification, no single point failures etc). You can setup a cluster, and publish a new domain in around 30 seconds (once you get the one time, 5 minute setup out of the way).

All you need to do is add a few lines to a docker-compose file and Roo does the rest see a full configuration example:

      OriginHosts: example.com,www.example.com
      OriginScheme: https
      OriginPort: 443
      DestinationHost: test_test
      DestinationScheme: http
      DestinationPort: 80

See a 5 minute line-by-line run-through example setup (includes setting up physical infrastructure, virtual docker swarm infrastructure, and roo cluster that runs on top of the swarm, that handles requests and passes it to your underlying clustered services)

Introduction

Roo functional diagram

This aims to be a free replacement of Amazon's ECS (Elastic Compute Service), EKS (Kubernetes), CertificateManager, Load-Balancer and CloudWatch using your own Docker Swarm. It IS a complete replacement for nginx, traefik, haproxy, and a lot of kubernetes. The idea is to give developers back the power and take it back from ridiculous self-complicating dev-ops tools that get more complicated and less useful (for example Traefik 2 just removed support for clustered Letsencrypt from their open source version to spruik their enterprise version. Nginx and HAProxy do the same). I wasted a lot of time on their software before writing this. I truly hope it benefits others too.

If you are unfamiliar with swarm/kubernetes and are a developer and want a quick intro into how powerful and easy swarm can be, see how you can setup a thousand-machine cluster in just 20 lines (just copy + paste from there) or check out my command notes. In a day I was scaling clusters up and down on my own infrastructure with single commands.

The power Roo gives you is to add HTTPS://example1.com and HTTPS://example2.com to your clustered services with zero configuration. Let's encrypt allocates your service's certificates. It works across every machine, docker node, service, in your cluster.

Roo itself is clustered. Every machine it runs on shares the load to your services. It's distributed store shares certificates from Letsencrypt used across all your nodes. Now apple is denying certificates older than a year, I feel as a dev, that lets encrypt is almost mandatory as it creates a lot of admin.

Getting Started (on docker)

  • Want to just run it?
docker run sfproductlabs/roo:latest

Getting Started (complete run-through example on Hetzner Cloud)

Setup the physical nodes

This will set you up with a cluster on Hetzner Cloud (change the first 20 lines to suit your own cloud provider). I use this on my own production servers. I don't love Hetzner - the service isn't as good as I'd like - but it is improving and is CHEAP (~$3 per server).

# mac
brew install hcloud 
brew install jq
#debian/ubuntu
#sudo apt install hcloud-cli jq
#create a project in hetzner called test (https://console.hetzner.cloud/projects)
#create a api key in the project you setup on hetzner
#hcloud context create test #connect the api key to the project
#test the connection
hcloud server-type list 
#start with an empty project (check this is empty)
hcloud server list 
hcloud ssh-key create --name andy --public-key-from-file ~/.ssh/id_rsa.pub  
hcloud network create --ip-range=10.1.0.0/16 --name=aftnet
hcloud network add-subnet --ip-range=10.1.0.0/16 --type=server --network-zone=eu-central aftnet
#If you want a lot more machines see the horizontal web scraper project commands (https://github.com/sfproductlabs/scrp)
#for n in {1..30}; do (hcloud server create --name docker$RANDOM$RANDOM$RANDOM$RANDOM --type cx11 --image debian-9 --datacenter nbg1-dc3 --network aftnet --ssh-key andy 2>&1 >/dev/null &) ; done
#watch -n 5 "echo "Press Ctrl-c to exit when your server count meets the desired amount. You will need to copy and paste just the following instructions to proceed." && hcloud server list | grep 'running' | awk 'END {print NR}'"
hcloud server create --name docker1 --type cx11 --image debian-9 --datacenter nbg1-dc3 --network aftnet --ssh-key andy 
hcloud server create --name docker2 --type cx11 --image debian-9 --datacenter nbg1-dc3 --network aftnet --ssh-key andy 
hcloud server create --name docker3 --type cx11 --image debian-9 --datacenter nbg1-dc3 --network aftnet --ssh-key andy 
rm *.txt
hcloud server list -o columns=name -o noheader > worker-names.txt
hcloud server list -o columns=ipv4 -o noheader > worker-ips.txt
cat worker-names.txt | xargs -I {} hcloud server describe -o json {} | jq -r '.private_net[0].ip' >> worker-vips.txt
hcloud server create --name manager1 --type cx11 --image debian-9 --datacenter nbg1-dc3 --network aftnet --ssh-key andy
hcloud server describe -o json manager1 | jq -r '.private_net[0].ip' > manager-vip.txt
scp -o StrictHostKeyChecking=no *.txt root@$(hcloud server list -o columns=ipv4,name -o noheader | grep manager1 | awk '{print $1}'):~/
scp -o StrictHostKeyChecking=no ansible/* root@$(hcloud server list -o columns=ipv4,name -o noheader | grep manager1 | awk '{print $1}'):~/
scp -o StrictHostKeyChecking=no *-docker-compose.yml root@$(hcloud server list -o columns=ipv4,name -o noheader | grep manager1 | awk '{print $1}'):~/

If it stuffs up run DANGEROUS it will delete all your servers for the project:

hcloud server list -o columns=name -o noheader | xargs -P 8 -I {} hcloud server delete {}

Setup the docker swarm

Get on the manager1 node:

#only required on a mac
eval `ssh-agent` && ssh-add ~/.ssh/id_rsa 
#now login to manager1
ssh -l root -A $(hcloud server list -o columns=ipv4,name -o noheader | grep manager1 | awk '{print $1}')

Then run:

apt-get update && \
apt-get upgrade -y && \
apt-get install apt-transport-https ca-certificates curl gnupg-agent software-properties-common -y && \
curl -fsSL https://download.docker.com/linux/debian/gpg | sudo apt-key add - && \
apt-key fingerprint 0EBFCD88 && \
add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/debian $(lsb_release -cs) stable" && \
apt-get update && \
apt-get install docker-ce docker-ce-cli containerd.io ansible -y && \
docker swarm init --advertise-addr=ens10 && \
docker swarm join-token worker | xargs | sed -r 's/^.*(docker.*).*$/\1/' > join.sh && \
chmod +x join.sh && \
printf "\n[defaults]\nhost_key_checking = False\n" >> /etc/ansible/ansible.cfg && \
printf "\n[managers]\n" >> /etc/ansible/hosts && \
cat manager-vip.txt >> /etc/ansible/hosts && \
printf "\n[dockers]\n" >> /etc/ansible/hosts && \
cat manager-vip.txt >> /etc/ansible/hosts && \
cat worker-vips.txt >> /etc/ansible/hosts && \
printf "\n[workers]\n" >> /etc/ansible/hosts && \
cat worker-vips.txt >> /etc/ansible/hosts && \
ansible dockers -a "uptime" && \
printf "\n            $(cat join.sh | awk '{print $0}')" >> swarm-init.yml && \
ansible-playbook swarm-init.yml && \
ansible dockers -a "docker stats --no-stream" && \
docker node ls  

Setup roo

#It's required to run roo on a manager node to get the automatic updates from the docker-compose files. You don't need to serve content from it though.
docker node update --label-add load_balancer=true manager1 && \
docker node update --label-add load_balancer=true docker1 && \
docker node update --label-add load_balancer=true docker2 && \
docker node update --label-add load_balancer=true docker3 && \
docker network create -d overlay --attachable forenet --subnet 192.168.9.0/24 && \
docker stack deploy -c roo-docker-compose.yml roo

You can watch roo boot & status using docker service logs roo_roo -f otherwise wait about a minute for the services & raft-log to auto-start.

Start a service (test)

  • Setup a test-domain and and IPs in your host-name record.
  • Edit test-docker-compose.yml and replace test.sfpl.io with the test-domain, (make sure to set it up in your host records and use a load-balancer!).

Then run:

docker stack deploy -c test-docker-compose.yml test

Go to your domain. All Done

Debug

Check roo's distributed keystore
docker run -it --net=forenet alpine ash
apk add curl
curl -X GET http://tasks.roo_roo:6299/roo/v1/kvs
Check the service log
docker service logs roo_roo -f
docker service ps roo_roo
Check using Ansible
ansible workers -a "ip addr"
More Help

Check the FAQ

Getting Started (on swarm)

Run this all in the docker swarm manager look here to get started.

  • Create the default network, for example:
docker network create -d overlay --attachable forenet --subnet 192.168.9.0/24 
  • Add label to the nodes you want to run it on (load_balancer describes the label for roo to attach to):
docker node update --label-add load_balancer=true docker1-prod
  • You'll need to run roo on at least one manager node if you want docker services to auto-sync with Let's Encrypt. Don't worry though as this is not a security issue like in other platforms (like Traefik) as the manager node need not be publicly accessible.
  • Run the docker-compose file on swarm (WARNING: LetsEncrypt hates load, so update the ROO_ACME_STAGING=true if you plan to muck about):
# docker stack deploy -c roo-docker-compose.yml roo
  • It takes a minute to bootstrap
  • Notice the roo images are available at dockerhub sfproductlabs/roo:latest. I don't use the Github repository service as it requires a key to just get the image.
  • Then in your own docker-compose file do something like (note the label used in the zeroconfig, and the destination, its usually swarmstack_swarmservicename):
version: "3.7"
services:
  test:
    image: test
    deploy:
      replicas: 1
    networks:
      - forenet
    labels:
      OriginHosts: test.sfpl.io
      DestinationHost: test_test
      DestinationScheme: http
      DestinationPort: 80
    
networks:
  forenet:
    external: true  
  • NOTE: Give it a minute before you query your service, as the service blocks unknown traffic (the docker watcher updates every 30 seconds or so, and it will need a bit to find your new service). JOB DONE!

Zeroconfig of docker swarm services

You need to tell roo what the incoming hostname etc is and where to route it to in the docker-compose file (if you want to go fully automoatic)

Letsencrypt caveats

Please be extremely careful, letsencrypt/acme have weekly limits (which are super low) that will ban you from getting certificates for a week. Please use the staging setting in the roo-docker-compose-file.

https://letsencrypt.org/docs/rate-limits/

Going Manual

Roo comes with a clustered Distributed Key-Value (KV) Store (See the API below for access). You can use this to manually configure roo. To add the route and do the zeroconfig example above manually, do this instead:

  • You'll need to first deploy the test stack
docker stack deploy -c test-docker-compose.yml test
  • Then let roo know the route (do this inside any machine in the forenet network):
curl -X PUT roo_roo:6299/roo/v1/kv/com.roo.host:test.sfpl.io:https -d 'http://test_test:80'

Building from source

  • You need to get the dependencies. Check the roo-docker-compose.yml file for a current list.
  • Run make in the root directory (not the roo directory).

Schema Definitions

Adding a route to the routing table

  • Use the Put API below
com.roo.host:<requesting_host<:port (optional)>:scheme>  <destination_url>

See an example

API

Post a single key to the distributed store

curl -X PUT -d'test data' http://localhost:6299/roo/v1/kv/test
  • Returns a json object { "ok" : true} if succeded

A Real Example Route

Put in a test route. For our example we are serving goole.com at our public endpoint, and routing to a swarm service with an endpoint inside our network on port 9001. The swarm stack is cool, the swarm service name is game. In docker swarm world this equates to a mesh network load balanced DNS record of tasks.cool_game. (you'll need to find your stack, service, and replace that and port 9001 with your details to get it working). The overall result to add this to our routing tables is:

curl -X PUT -d'http://tasks.cool_game:9001' http://localhost:6299/roo/v1/kv/com.roo.host:google.com:https

  • Note: as we don't specify a port in our requests we remove the optional port :443

To test you can run something like this (this just makes your localhost pretend like its responding to a request to google.com): curl -H "Host: google.com" https://localhost/

So to summarize, google.com:443 is the incoming route to roo from the internet, and tasks.cool_game:9001 is your service and port to your internal service (in this case its an internal intranet docker swarm service).

Get a SINGLE key from the store

curl -X GET http://localhost:6299/roo/v1/kv/test
  • Returns the raw bytes (you'll see this as a string if you stored it like that)

List multiple keys from the store (Using a SCAN/Prefix filter query)

curl -X GET http://localhost:6299/roo/v1/kvs/te #Searches the prefix _te_
curl -X GET http://localhost:6299/roo/v1/kvs/tes #Searches the prefix _tes_
curl -X GET http://localhost:6299/roo/v1/kvs/test #Searches the prefix _test_
curl -X GET http://localhost:6299/roo/v1/kvs #Gets everything in the _entire_ kv store (filter is on nothing)
  • Returns the rows searched using the SCAN query (in KV land its a prefix filter) in JSON
  • The resulting values are encoded in base64, so you may need to convert them (unlike the GET single query above which returns raw bytes)
  • Use window.atob("dGVz......dCBkYXRh") in javascript. Use json.Unmarshall or string([]byte) in Golang Go if you want a string... OR just use GET instead.

Help!

FAQ

It hangs

  • You've reached the maximum requests to LetsEncrypt probably. You can check this by running curl -X GET http://localhost:6299/roo/v1/kvs/ and finding com.roo.cache:your.domain.com. If it exists it's not this.
  • Or your stack_service name isn't correct (the DestinationHost label in your docker-compose file).

I don't see my site!

  • You may have tried to go to the site before the service updated the docker routing table (it updates every 60 seconds). If you went to your site before it updates, roo thinks that your domain is a dodgy request and puts similar requests in the bin for 4 minutes.

Something might be up!

  • Get a status of your containers (you first need to get ansible, and add your docker nodes to the docker-hosts group [you may need to look into ssh-agent too]):
ansible docker-hosts -a "docker stats --no-stream"
  • Or go deeper:
ansible docker -a "bash -c 'ps waux'" | grep -E "(rood)"

This will get a realtime snapshot on all your machines in your swarm.

Get some profiling info

NOTE: This might be removed in a future release, but it's super useful now

Install golang on a computer in your network and get profiler insights (see more https://golang.org/pkg/runtime/pprof/):

On swarm run go tool pprof http://roo_roo_host_from_internal_forenet_network:6299/debug/pprof/heap then type svg (or output to something you may like better).

On my local development machine I do:

go tool pprof http://localhost:6299/debug/pprof/heap?debug=1

then type:

(pprof) svg

Example of a roo heap profile

Or for intermittent results

Run from source directory (./roo):

mkdir pprof
cd pprof
curl http://localhost:6299/debug/pprof/heap > heap.0.pprof
sleep 30
curl http://localhost:6299/debug/pprof/heap > heap.1.pprof
sleep 30
curl http://localhost:6299/debug/pprof/heap > heap.2.pprof
sleep 30
curl http://localhost:6299/debug/pprof/heap > heap.3.pprof

Then:

go tool pprof pprof/heap.3.pprof
(pprof) top20
(pprof) list NewWriterSize

Or for manual results

Go to http://roo_roo_host_from_internal_forenet_network:6299/debug/pprof/

Get detailed memory info using Valgrind

The docker image contains apt install valgrind

Run:

#valgrind --tool=memcheck --vgdb=yes --leak-check=yes --track-origins=yes --progress-interval=600 /app/roo/rood /app/roo/roo/config.json

Then run gdb (you may need to apt install gdb on swarm node first):

#gdb /app/roo/rood
(gdb) target remote | vgdb
(gdb) set logging on
(gdb) set logging redirect on
(gdb) set logging overwrite on
(gdb) set logging debugredirect on
(gdb) show logging
(gdb) monitor leak_check full reachable any

Or alternatively pipe it to a gdb.txt logfile:

gdb -c /app/roo/rood -ex "target remote | vgdb" -ex "set logging on" -ex "set logging redirect on" -ex "set logging debugredirect on" -ex "set logging overwrite on" -ex "monitor leak_check full reachable any" -ex detach -ex quit 0 2>&1 >/dev/null

Inspecting the roo containers

  • Inspect the logs
docker service logs roo_roo -f
  • Inspect the containers
docker run -it --net=forenet alpine ash
nslookup tasks.roo_roo.
curl -X GET http://<result_of_nslookup>:6299/roo/v1/kvs

Andrew's DevOps Setup (Yuck! But we have to do it)

  • I use floating IPs that transition from swarm worker to swarm worker upon failure. (Highly Available)
  • with round robin DNS (setup a few workers and share the IPs) (Load Balanced, a bit dodgy but works)
  • You need to make the swarm redundant with more than one node, the mesh network load balances internally. Make it more than 2 nodes so that the raft cluster doesn't fight over leadership. Odd numbers are great. (HA & LB)
  • It may not cost as much as an AWS ELB, but it probably won't saturate either. Yes, I've run into issues where you need to "warm up" the Amazon elastic load balancer before. I wouldn't be surprised if this handled as much traffic without the cost.
  • No need to say this is $$$$ cheap. I'm saving 10x as much as I would be if I used Amazon AWS using this setup. It's nice to give my startups the same tech my enterprise clients get, but they can actually afford. I don't want to share which cloud provider I used as it took 30 days to request 1 extra machine. But it was cheap. Let us all know if you find something better! You can get these benefits too if you look!

TODO

  • Add jwt token inspection (optional) as replacement for client ssl auth
  • Add support for Elastic's APM Monitoring Service https://www.elastic.co/guide/en/apm/get-started/current/quick-start-overview.html
  • Add an option to whitelist hostnames only in the store (this will prevent dodgy requests)
  • Add a synchronized scheduler so that only one docker manager runs the auto-update script (it currently depends on 1 manager node notifying the slaves indirectly via the kv store)
  • Memory api checker needs to be cached in hourly, replace kvcache, docker update, add node to hosts during join so if it fails it can be deleted, cache host whitelist
  • Autoscaling raft
  • Could add rejoin once kicked out of raft https://github.com/lni/dragonboat/blob/master/config/config.go#L329
  • Autoscale Docker
  • Autoscale Physical Infratructure
  • Move flaoting IPs (Load balance, service down) https://github.com/sfproductlabs/floater thanks @costela
  • SSL in API
  • HTTP for Proxying Origin (Only SSL Supported atm)
  • Add end to end encryption of kv-store and distributed raft api and api:6299
  • Investigate the possibilty of a race condition between the in memory certificate/proxy cache right when letsencrypt should be renewing (might be a 10 minute window of inoperability)? Interesting thought...
  • Use Kubernetes container management (I'd like help with this one esp.!)

Support

Roo is in use commercially and in production. If you want support, email us at support@sfpl.io.

About

Setup and run your own clusters on your own infrastructure in minutes (Eg. ECS or EKS). Roo is a zero config distributed ingress, edge-router & reverse-proxy (supporting multiple letsencrypt/https hosts) using Docker Swarm. No dependencies.

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published