This is a reverse proxy that which will shard pads based on padId(within query[currently in socket-namespace branch of Etherpad core]) to a pool of backends based on the availability of the backends which is based on the number of concurrent active Pads being edited.
It's likely that this project will only get to proof of concept stage(see V0) and then something that integrates with HAProxy/Varnish et al will replace it as NodeJS is probably not the right tool for the job but having the high level management system written in NodeJS makes sense to discover potential pitfalls and best practices.
Copy settings.json.template to settings.json and modify the values.
node app.js
Visit http://localhost:9000
Settings come from settings.json, see settings.json.template for an example to modify for your environment.
backends is your Backend Etherpad instances.
maxPadsPerInstance is how many active pads you want to allow per instance. This value should be between 1 and 20000 depending on the number of authors and words per minute that you limit or you wish to allow. Once this limit is met then random instances will be used.
checkInterval is how often to check every backend for availability. You should set this to a low number if you have lower number of very active instances with short pad life. You should set this to a high number if you have lost of instances with relatively long pad life expectancy.
port is the port this service listens on.
For database settings/options please see UeberDB https://github.com/ether/ueberdb
- Test in production.
- Figure out why changing
1000to200forcheckIntervalmakes tests fail. - Abstract http-proxy out / introduce support for other proxy software/services.
- API to create/destroy backends - REF: https://github.com/colyseus/proxy/blob/master/proxy.ts
- If no backends are available, send a message explaining "we're full up"
Apache 2