innmind/tower

Application deployment tool

2.2.0 2021-02-20 11:07 UTC

This package is auto-updated.

Last update: 2024-11-20 19:46:57 UTC


README

Build Status codecov Type Coverage

This is a command line tool to deploy your code base with a new approach. Instead of building another tool sending a set of shell commands over ssh with a logic of point to point, Tower takes the approach of servers as nodes of a tree where from a node you trigger the tower of sub-nodes.

The setup:

  • one install of Tower per server
  • the config to deploy the server is localised on it
  • a node* only knows its neighbours' servers

Advantages:

  • One only knows what can be deployed on a server (not how)
  • Keep the know how to deploy on the server
  • If the actions are updated, no one is impacted
  • Easily cascade deployment (if a node retrieve sources from its parent)
  • Nodes at the same level can be deployed in parrallel (via background jobs, loose direct output)
  • Commands run locally, everything's logged so you can trace what/when an env is deployed

Drawbacks:

  • A node need to know how to connect to its neighbours (if node is hacked the subtree is compromised) (specific to the ssh transport, tcp not affected)
  • A node need to know how to connect to its parent to retrieve sources (lower the load on master repository) (only if you decide such a strategy)
  • Cross relation between parent and neighbours

Example:

              A (you)
             / \
            /   \
           /     \
   (prod) B       C (staging)
         / \       \
        /   \       \
       /     \       \
      D       E       F
      |       |
      |       |
      G       H

Say here the tree below B is also for production, with this tool you could easily place the same Tower setup on the five servers and to deploy all of them, you would run a command like tower ping B and done! Or if you want to only deploy H, just to be sure everythings deploying fine, connect to the machine and locally run tower trigger or add this as one of the neighbour on your local machine (A) and run tower ping H. Once again it's a tree, you can start from wherever you want.

But in a normal case you would just have B and C as neighbours of your machine.

To be really efficient (meaning not overloading your VCS server), setup neighbours to retrieve code from their parent. For instance, setup B and C with your VCS server as git remote; D and E have B as git remote, and so on...

Note: I talk about git here, but you're not forced to use it

Note: When cascading, if a neighbour fail to deploy, its subtree won't be deployed

Another use case would be you have a single server for your app, and other servers for related services required by your first server. You could imply, with this notion of tree, that every time you deploy your app it triggers the deployment of those related services (so everything is always up-to-date).

Installation

composer global require innmind/tower

Configuration

neighbours:
    _name_:
        url: ssh://example.com:80/path/to/config/on/neighbour/server.yml # scheme can be tcp or ssh, path only used via ssh
        tags: [foo, bar] # optional

exports: # optional
    - 'echo "ENV=value"' # contains list of env variables that will be available to each action

actions:
    - 'some bash command'

The actions set will be the commands run on you server when the server the configuration is on is pinged. Each actions will have as environment variables the ones built via the exports section, this array must be commands that will produce an output of the form ENV=value.

Finally the neighbours section is the list of the servers that will be pinged when the one the configuration is on is pinged. You can decide to ping a server either by ssh (leaving the server vulnerable to the outside world, but is the easiest) or by a tcp socket (allows to close all direct access to the machine but opens to a DOS attack).

Usage

Via ssh

On your machine configure a tower.yml file with the following:

neighbours:
    gateway:
        url: ssh://gateway.com:22/path/to/configuration/on/gateway

Once done you can run tower ping gateway in this folder. This will connect to gateway.com via the port 22, move to the folder /path/to/configuration/on/gateway and run the command tower trigger; you'll need to create a file tower.yml in the folder that will either describe the actions to do or the neighbours to ping.

Via tcp

On your machine configure a tower.yml file with the following:

neighbours:
    gateway:
        url: tcp://gateway.com:1337

On the server you need to create a similar file named tower.yml containing either the actions to run or the neighbours to ping; then in this folder run tower listen 1337 -d. This last command opens a tcp connection on port 1337 and waits for incoming pings.

Now you can run on your machine tower ping gateway.