Nebula on Packet global deployment example

The docker-compose tutorial is great for getting a feeling and doing local development on the Nebula API but as a repeated request has been to have a simple way to confirm Nebula global scalability a terrform module has been created which will do the following:

  • Install a MongoDB & Nebula manager on a single packet devices
  • Create multiple devices (by default 5) over multiple packet facilities (a list of 5 different facilities by default)
  • Create and configure an example app & device_group on Nebula

Running instructions

  1. clone the repo
  2. cd into the repo folder
  3. Change the (in repo root folder) to your variables
  4. run terraform apply.

You now have a nebula cluster that includes a single manager and as many workers you gave it (default 9) spread out across packet regions (default 9 regions).


  • terraform 0.12.0 or higher
  • PACKET_AUTH_TOKEN envvar set to your packet token
  • a packet organization
  • curl installed on the machine running terraform

Configuration variables :

  • server_size - the size of the Nebula manager server - defaults to c1.small.x86
  • server_region_device - the region where the nebula manager server will reside - defaults to ams1
  • worker_count - the number of Nebula workers to create - defaults to 5
  • worker_region_device - the regions where the workers will be deployed on - defaults to ["ams1","dfw2","ewr1","nrt1","sjc1"]
  • worker_size - the size of the servers used in Nebula - note that at least 1GB is needed as it uses a RancherOS for simplicity sake - defaults to c1.small.x86
  • PACKET_AUTH_TOKEN - the token used to connect to your packet account - needs to be declared as an envvar - required

Should you want to ssh to the devices for any reason (not needed for the system to work) you can do so with the private key & the root user for the manager and rancher user for the workers.

this can take 10-15 minutes for the manager to start as it has to install docker & docker-compose then download and run all containers, the initial boot of the workers is considerably faster but it might still be a couple of minutes after the terraform run completes.