I've been spending a lot of time lately working on nodes for various blockchain projects (Polkadot, Cardano, Tron, Binance Chain, Ethereum, ...). The rosetta api spec is super interesting, but like most things in crypto the documentation is sometimes wrong or incomplete and there are bugs and undocumented features.
Each of the nodes runs on a separate server, and we typically have one node for mainnet and another for testnet. I'm working across mutiple servers, doing difficult stuff, and I want it to be as easy as possible.
I need to reduce friction and cognitive load.
Accessing the servers is easy - I use Tmux with the continuum and resurrect plugins and maintain different sessions for each type of server. This makes accessing multiple servers during the same work day really simple and effortless. But working on the servers is still awkward.
On my dev machine I have
zsh with syntax highlighting, command completion, and various tools, like z to make navigation supper easy. I also have a lot of aliases defined. E.g.
cd ... Working on a remote server should be as convenient and familiar as working on my local machine, so I want to find a way to configure a server the same way as my laptop, and I want to do it automatically so that it can be done many times, with no additional effort.
Ansible seems to be popular because:
- It's free
- It's got all the features and capabilities you're going to need
- It's agentless - you don't need to install anything on the machine you want to control - you can use Ansible with anything that you can ssh into.
I used the following resources to get started:
This useful video gave me some orientation and helped me figure out what I was aiming for and how to get started. Before watching it, I didn't know "which way was up".
This blog post showed me how to create an inventory using the
I spent a lot of time being unnecessarily confused about service accounts. I guess until you have 1 success at understanding something you don't know if you've misunderstood by a little or a lot.
Once you have an inventory of servers that you want to connect to, you still need to specify (and prepare for) how you will connect to them. I'd hoped that the
gcp_computeplugin would do some heavy lifting for me in this step, but it seems not. It can do lots of useful stuff like creating instances and specifying disk space and networks, but it won't really help you ssh into an instance. No matter though...
This blog post turned out to be just what I needed. I found it at the beginning of my search when I was trying to create an inventory, and discarded it as almost useful. Turns out that
OS Loginis the best way to ssh into a GCE instance and once you've got your inventory taken care of, this blog post really helps.
When I was installing python modules, I had some errors about pyenv shims being incorrect. The scripts were looking for versions that weren't present. Running
pyenv reshash fixed it. Kind of magically, but annoying.
Setting up a service account and giving it the correct permissions took more time and was more confusing than anything to do with Ansible.
I found this blog post about setting up vim for
The preferred way to install ansible on Mac is using
When you use
OS Login the username you have when you ssh into the compute instance will change. This SO question explains why.
gcloud auth list
ansible-inventory -i inventory.compute.gcp.yml --graph
ansible -i inventory.compute.gcp.yml all -m ping